CN114359685A

CN114359685A - Training method, device and equipment for logistics piece recognition model

Info

Publication number: CN114359685A
Application number: CN202011032568.3A
Authority: CN
Inventors: 彭显明
Original assignee: SF Technology Co Ltd
Current assignee: SF Technology Co Ltd
Priority date: 2020-09-27
Filing date: 2020-09-27
Publication date: 2022-04-15

Abstract

The utility model provides a training method, a device and equipment of a logistics piece recognition model, the trained logistics piece recognition model can efficiently recognize the number of logistics pieces in an automatic sorting assembly line of the logistics pieces, so that the related equipment of the automatic sorting assembly line can analyze whether the abnormal sorting condition exists or not in the first time according to the recognized number of the logistics pieces. The method comprises the following steps: acquiring initial images, wherein the initial images are a plurality of images obtained by shooting logistics pieces for training, and the number of the logistics pieces in the images is marked on the initial images; identifying the logistics piece surface list in the initial image to obtain a logistics piece surface list image; filtering the single image of the logistics part surface in the initial image to obtain a target image; and training the initial neural network model through a target image, and taking the trained model as a physical distribution piece recognition model, wherein the physical distribution piece recognition model is used for recognizing the number of physical distribution pieces existing in the input image.

Description

Training method, device and equipment for logistics piece recognition model

Technical Field

The application relates to the field of logistics, in particular to a training method, a training device and training equipment for a logistics piece identification model.

Background

In the logistics transportation operation, under the background of optimizing the transportation efficiency, the transportation link and the sorting link of the logistics pieces are two operation links with larger optimization space.

Newer sorting equipment has a one-to-one scanning, weighing capability to check whether the physical distribution pieces themselves match the information of the physical distribution sheets attached to the physical distribution pieces.

In the existing research process of the related technology, the inventor finds that the existing sorting equipment often has abnormal sorting conditions, and the automatic sorting efficiency of the sorting operation is influenced.

Disclosure of Invention

The utility model provides a training method, a device and equipment of a logistics piece recognition model, the trained logistics piece recognition model can efficiently recognize the number of logistics pieces in an automatic sorting assembly line of the logistics pieces, so that the related equipment of the automatic sorting assembly line can analyze whether the abnormal sorting condition exists or not in the first time according to the recognized number of the logistics pieces.

In a first aspect, the present application provides a method for training a logistics piece recognition model, where the method includes:

acquiring initial images, wherein the initial images are a plurality of images obtained by shooting logistics pieces for training, and the number of the logistics pieces in the images is marked on the initial images;

identifying the logistics piece surface list in the initial image to obtain a logistics piece surface list image;

filtering the single image of the logistics part surface in the initial image to obtain a target image;

and training the initial neural network model through a target image, and taking the trained model as a physical distribution piece recognition model, wherein the physical distribution piece recognition model is used for recognizing the number of physical distribution pieces existing in the input image.

With reference to the first aspect of the present application, in a first possible implementation manner of the first aspect of the present application, the identifying a logistics item surface sheet in an initial image, and obtaining an image of the logistics item surface sheet includes:

calling a logistics item list identifier configured by a user, wherein the logistics item list identifier is used for identifying the existence area of the logistics item list in the initial image;

and determining a single face image of the logistics piece in the initial image according to the single face identifier of the logistics piece.

With reference to the first aspect of the present application, in a second possible implementation manner of the first aspect of the present application, before training the initial neural network model by using the target image, the method further includes:

and carrying out image enhancement processing on the initial image or the target image so as to improve the image quality.

With reference to the second possible implementation manner of the first aspect of the present application, in a third possible implementation manner of the first aspect of the present application, the performing image enhancement processing on the initial image or the target image includes:

separating a first brightness channel image corresponding to a brightness channel in a color space from an image to be enhanced, wherein the image to be enhanced is an initial image or a target image;

carrying out contrast enhancement processing on the first brightness channel image to obtain a second brightness channel image;

and fusing the second brightness channel image with the channel images except the first brightness channel image in the image to be enhanced.

With reference to the first aspect of the present application, in a fourth possible implementation manner of the first aspect of the present application, before filtering a single image of a logistics surface in an initial image to obtain a target image, the method includes:

and carrying out filtering processing on image information with high-frequency signal characteristics in the initial image.

With reference to the fourth possible implementation manner of the first aspect of the present application, in a fifth possible implementation manner of the first aspect of the present application, the filtering process is specifically a median filtering process, a smoothing ratio of the median filtering process is K/N, a maximum labeling error of the median filtering process is K/M, K is used to indicate a kernel size of the median filtering process, N is used to indicate a single minimum pixel size of a logistics piece surface preset in an initial image, and M is used to indicate a preset average pixel size of the logistics piece.

With reference to the first aspect of the present application or any one of the possible implementation manners of the first aspect, in a sixth possible implementation manner of the first aspect of the present application, the initial image is an image obtained by shooting a weighing device, the weighing device is a dynamic weighing device on a logistics sorting line, the weighing device is configured to weigh a logistics item to be transported while transporting the logistics item to be transported, train the initial neural network model through a target image, and after taking the trained model as a logistics item identification model, the method further includes:

acquiring an image to be identified, which is obtained by shooting a weighing device;

inputting the image to be identified into a logistics piece identification model so as to identify the number of logistics pieces existing in the image to be identified;

and when the number of the physical distribution pieces existing in the image to be identified is more than 1, determining that the weighing equipment has an abnormal weighing state.

In a second aspect, the present application provides a training device for a logistics piece identification model, the device including:

the acquisition unit is used for acquiring initial images, wherein the initial images are a plurality of images obtained by shooting logistics pieces for training, and the number of the logistics pieces existing in the images is marked on the initial images;

the identification unit is used for identifying the logistics piece surface list in the initial image to obtain a logistics piece surface list image;

the filtering unit is used for filtering the single image of the logistics part surface in the initial image to obtain a target image;

and the training unit is used for training the initial neural network model through the target image and taking the trained model as a logistics piece recognition model, wherein the logistics piece recognition model is used for recognizing the number of logistics pieces existing in the input image.

With reference to the second aspect of the present application, in a first possible implementation manner of the second aspect of the present application, the identifying unit is specifically configured to:

With reference to the second aspect of the present application, in a second possible implementation manner of the second aspect of the present application, the apparatus further includes an enhancing unit, configured to:

With reference to the second possible implementation manner of the second aspect of the present application, in a third possible implementation manner of the second aspect of the present application, the enhancing unit is specifically configured to:

With reference to the second aspect of the present application, in a fourth possible implementation manner of the second aspect of the present application, the apparatus further includes a filtering unit, configured to:

With reference to the fourth possible implementation manner of the second aspect of the present application, in a fifth possible implementation manner of the second aspect of the present application, the filtering process is specifically a median filtering process, a smoothing ratio of the median filtering process is K/N, a maximum labeling error of the median filtering process is K/M, K is used to indicate a kernel size of the median filtering process, N is used to indicate a single minimum pixel size of a logistics piece surface preset in an initial image, and M is used to indicate a preset average pixel size of the logistics piece.

With reference to the second aspect of the present application or any one of the possible implementation manners of the first aspect, in a sixth possible implementation manner of the second aspect of the present application, the initial image is an image obtained by shooting a weighing device, the weighing device is a dynamic weighing device on a logistics sorting line, the weighing device is configured to weigh the logistics items to be transported while transporting the logistics items to be transported, and the apparatus further includes an application unit configured to:

In a third aspect, the present application further provides a training device for a logistics item identification model, which includes a processor and a memory, where the memory stores a computer program, and the processor executes the steps in any one of the methods provided in the first aspect of the present application when calling the computer program in the memory.

In a fourth aspect, the present application further provides a computer-readable storage medium storing a plurality of instructions adapted to be loaded by a processor to perform the steps of any of the methods provided in the first aspect of the present application.

From the above, the present application has the following advantageous effects:

in the training process of the logistics piece identification model, after an initial image obtained by shooting a training logistics piece is obtained, the logistics piece surface list in the initial image is identified to obtain a corresponding logistics piece surface single image, the logistics piece surface single image is filtered in the initial image to obtain a target image, the initial neural network model is trained according to the target image to obtain the logistics piece identification model, the model can be used for identifying the number of the logistics pieces in an input image, in the training mode, because the identification precision requirement of the number of the logistics pieces is only required to be met, the local image characteristics are not required to be concerned, the image processing workload of image identification on the logistics pieces can be obviously reduced, the high-efficiency identification efficiency of the number of the logistics pieces can be achieved, and related equipment of an automatic sorting production line can identify the number of the logistics pieces according to the identified number of the logistics pieces, analyzing whether abnormal sorting conditions exist or not within the first time; secondly, the logistics piece surface list in the image is filtered, so that the image processing workload required for identifying the logistics piece surface list is directly avoided, the identification efficiency is further improved, sensitive information existing in the logistics piece surface list can be filtered, the effects of desensitization and user privacy protection are achieved, and the logistics piece identification model is beneficial to popularization and application.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

Fig. 1 is a schematic flow chart of a training method of a physical distribution piece recognition model in the present application;

FIG. 2 is a schematic flow chart illustrating the identification of a surface sheet of a physical distribution article according to the present application;

FIG. 3 is a schematic view of a single sign scenario of the present application;

FIG. 4 is a schematic flow chart of the contrast enhancement process of the present application;

FIG. 5 is a schematic structural diagram of a training device for a physical distribution piece recognition model according to the present application;

fig. 6 is a schematic structural diagram of a training device of the logistics part identification model of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

In the description that follows, specific embodiments of the present application will be described with reference to steps and symbols executed by one or more computers, unless otherwise indicated. Accordingly, these steps and operations will be referred to, several times, as being performed by a computer, the computer performing operations involving a processing unit of the computer in electronic signals representing data in a structured form. This operation transforms the data or maintains it at locations in the computer's memory system, which may be reconfigured or otherwise altered in a manner well known to those skilled in the art. The data maintains a data structure that is a physical location of the memory that has particular characteristics defined by the data format. However, while the principles of the application have been described in language specific to above, it is not intended to be limited to the specific form set forth herein, and it will be recognized by those of ordinary skill in the art that various of the steps and operations described below may be implemented in hardware.

The principles of the present application may be employed in numerous other general-purpose or special-purpose computing, communication environments or configurations. Examples of well known computing systems, environments, and configurations that may be suitable for use with the application include, but are not limited to, hand-held telephones, personal computers, servers, multiprocessor systems, microcomputer-based systems, mainframe-based computers, and distributed computing environments that include any of the above systems or devices.

The terms "first", "second", and "third", etc. in this application are used to distinguish between different objects and not to describe a particular order. Furthermore, the terms "include" and "have," as well as any variations thereof, are intended to cover non-exclusive inclusions.

First, before the present application is introduced, the relevant contents of the present application with respect to the application background will be described.

The training method and device for the logistics piece identification model and the computer readable storage medium can be applied to training equipment of the logistics piece identification model, the trained logistics piece identification model can efficiently identify the number of logistics pieces in an automatic sorting assembly line of the logistics pieces, and therefore whether the abnormal sorting condition exists or not can be analyzed within the first time according to the number of the identified logistics pieces by relevant equipment of the automatic sorting assembly line.

According to the training method of the logistics item identification model, an execution main body of the training method can be a training device of the logistics item identification model, or different types of training Equipment of the logistics item identification model such as server Equipment, a physical host or User Equipment (UE) integrated with the training device, wherein the training device of the logistics item identification model can be realized in a hardware or software mode, and the UE can be terminal Equipment such as a smart phone, a tablet computer, a notebook computer, a palm computer, a desktop computer or a Personal Digital Assistant (PDA). The training device of the logistics piece identification model can be further divided into a plurality of devices and jointly executes the method provided by the application.

In the prior art, a weighing link in an automatic sorting assembly line is taken as an example, in the weighing link, a logistics list on a logistics piece can be scanned in a scanning mode, the logistics list can identify the weight of the logistics piece, the weight of the scanned logistics piece identification is compared with the weight weighed by weighing equipment at the moment, if the weight of the logistics piece identification is consistent with the weight of the weighing equipment, the weight of the logistics piece identification can be determined to be normal and effective, and if the weight of the logistics piece identification is inconsistent with the weight of the weighing equipment, the weight of the logistics piece identification can be determined to be abnormal and invalid, manual inspection is needed, or the logistics piece is returned for sorting again.

In the process of actual automatic sorting operation, other working links such as an initial sorting link and a transportation link are also configured at the upstream of the weighing link, and if abnormal conditions such as logistics piece accumulation, sorting abnormity, transportation abnormity and the like occur in the upstream link, the condition that more than two logistics pieces occur in the weighing link within the same detection time is possible to occur, and under the condition, the weight of a scanned logistics list is inevitably not matched with the weight of more than two logistics pieces.

Taking A, B logistics pieces appearing on the weighing device in the same detection time as an example, the logistics list scanned on the logistics piece a gets the identified weight of 5kg, meanwhile, the actual weight of the logistics piece a is also 5kg, and the actual weight of the logistics piece B is 3kg, it is easy to see that although the 5kg weight of the logistics list identification on the logistics piece a matches the actual weight thereof (the weight of the logistics list identification on the logistics piece a is normal and effective), since the weighing device weighs A, B two logistics pieces (total 8kg), the two logistics pieces do not match, at this time, if the report is wrong, manual inspection is performed, or the reflow is sorted again, obviously, the sorting efficiency of the automated sorting operation is affected.

Based on the above defects of the prior related art, the present application provides a training method for a logistics piece identification model, which overcomes the defects of the prior related art at least to a certain extent.

Wherein, in this application, the commodity circulation piece specifically can be the express delivery piece, and automatic sorting operation is the automatic sorting of going on the express delivery piece naturally, and commodity circulation piece face list on the commodity circulation piece specifically can be the express delivery face list on the express delivery piece, and commodity circulation piece face list is used for the relevant information of this commodity circulation piece of sign, for example bar code, two-dimensional code, commodity circulation identifier, sender information, addressee information, weight information, volume information, commodity circulation mode of transportation, article type etc..

Next, a method for training a logistics piece recognition model provided by the present application is described.

First, referring to fig. 1, fig. 1 shows a schematic flow chart of a training method for a logistics item identification model in the present application, and as shown in fig. 1, the training method for a logistics item identification model in the present application may specifically include the following steps:

step S101, obtaining an initial image, wherein the initial image is a plurality of images obtained by shooting logistics pieces for training, and the number of the logistics pieces in the image is marked on the initial image;

step S102, identifying the logistics piece surface list in the initial image to obtain a logistics piece surface list image;

s103, filtering the single image of the logistics part surface in the initial image to obtain a target image;

and S104, training the initial neural network model through the target image, and taking the trained model as a logistics piece identification model, wherein the logistics piece identification model is used for identifying the number of logistics pieces existing in the input image.

According to the scheme, in the training process of the logistics piece identification model, after the initial images obtained by shooting the training logistics pieces are obtained, the logistics piece surface single images in the initial images are identified to obtain the corresponding logistics piece surface single images, the logistics piece surface single images are filtered in the initial images to obtain the target images, the initial neural network model is trained according to the target images to obtain the logistics piece identification model, the model can be used for identifying the number of the logistics pieces in the input images, in the training mode, the identification precision requirement of the number of the logistics pieces is met, local image characteristics are not required to be concerned, the image processing workload of image identification on the logistics pieces can be obviously reduced, the identification efficiency of the number of the logistics pieces is high, and related equipment of an automatic sorting production line can identify the number of the logistics pieces according to the identified number of the logistics pieces, analyzing whether abnormal sorting conditions exist or not within the first time; secondly, the logistics piece surface list in the image is filtered, so that the image processing workload required for identifying the logistics piece surface list is directly avoided, the identification efficiency is further improved, sensitive information existing in the logistics piece surface list can be filtered, the effects of desensitization and user privacy protection are achieved, and the logistics piece identification model is beneficial to popularization and application.

The following specifically describes the steps of the training method of the logistic part identification model shown in fig. 1.

In the application, the training logistics piece in the initial image can be not only a sample logistics piece configured for a training model by a worker, but also a real logistics piece involved in the actual logistics operation of a logistics company.

Similarly, the initial image obtained by shooting the logistics piece for training can be obtained by shooting a training model for a worker, and can also be an image of a related business of shooting the logistics piece in the actual logistics operation of a logistics company, so that a ready-made initial image can be obtained, and the intervention of manual operation of the worker is further reduced.

For example, the training device of the logistics piece identification model provided by the application can establish a data channel with a server or a database of a logistics company under the configuration of a worker, monitor logistics piece images updated in the server or the database within a period of time, wherein the logistics piece images are stored in related businesses with the requirement of shooting logistics pieces in actual logistics work of the logistics company, for example, sorting processes of shooting logistics pieces can be performed in an automatic sorting pipeline, whether sorting is normally completed or not is analyzed through image identification, for example, when the logistics pieces are received and dispatched, the logistics pieces can be shot and stored so as to obtain evidence and verify the good appearance of the logistics pieces, and when the logistics piece images updated in the server or the database are monitored, the images can be called to trigger the training method of the logistics piece identification model provided by the application, and training a logistics piece recognition model.

Of course, the training device of the physical distribution piece recognition model may be a server device of the physical distribution company itself, and the server device itself may also function as a server or a database of the physical distribution company itself, so that the physical distribution piece image in the server or the database may be directly read from the local.

After the initial image used for training the model is obtained, the filtering processing of the logistics piece surface list is started by the initial image.

Specifically, the material flow sheet in the initial image may be identified first, and the material flow sheet image included in the initial image is segmented from the initial image, where the material flow sheet image is the corresponding image of the material flow sheet identified in the initial image.

In the present application, the identification of the logistics item list in the initial image may be implemented on the one hand by a neural network, such as an object detection network.

For example, in a logistics company, there may be some business or application scenarios involving identification of a logistics item list, for example, in an automated sorting line, information in the logistics item list on the logistics item, such as an identification weight, may be read by means of code scanning, and then the actual weight weighed by the logistics item is combined to check whether there is a weight anomaly in the logistics item.

On the other hand, the identification of the logistics item list in the initial image can also be realized through manual marking.

In an exemplary implementation, referring to a schematic flow chart of the present application for identifying a logistics item list shown in fig. 2, in the present application, identifying a logistics item list in an initial image may specifically include:

step S201, a logistics item list identifier configured by a user is called, wherein the logistics item list identifier is used for identifying the existence area of the logistics item list in an initial image;

it can be understood that after the training device of the physical distribution piece recognition model calls the initial images, the user may identify the physical distribution piece lists for the images, or after the user identifies the physical distribution piece lists for the images, the training device of the physical distribution piece recognition model obtains the initial images.

In the process of manually identifying the logistics item list, a user or a staff of the logistics company can see the logistics item list in the image and identify the existence area of the logistics item list in the image, for example, as shown in fig. 3, a scene schematic diagram of the logistics single-side list identification of the present application is shown, in fig. 3, the logistics single-side list identification specifically exists in the form of an identification frame, and the surrounding area of the identification frame is the existence area of the logistics item list.

Certainly, in practical application, the identification manner of the identification frame is only a specific implementation manner, and other identification manners may also be adopted, for example, configuration information related to image configuration may be said to be, and in the configuration information, the existence area of the logistics piece surface in the image is described by taking a pixel as a unit; for another example, a polygon formed by multi-segment folding lines attached to the material flow element surface sheet may be used to mark the material flow element surface sheet in the image, and the polygon has no other objects except the material flow element surface sheet, and specifically, the marking manner may be adjusted according to actual needs, and is not limited herein.

Step S202, according to the logistics piece surface single identification, a logistics piece surface single image is determined in the initial image.

After the training equipment of the logistics piece identification model obtains the logistics piece surface list identification, the corresponding logistics piece surface list image can be determined according to the existence area identified by the logistics piece surface list.

The determining of the single image of the surface of the physical distribution piece in the initial image may be determining the position of the single image of the surface of the physical distribution piece in the initial image, or segmenting a single image of the surface of the physical distribution piece from the initial image through an image.

After the face sheet image of the logistics piece is determined, the filtering process of the face sheet of the logistics piece in the application can be continued.

Specifically, the logistics piece surface single image can be deleted in the initial image, the corresponding image information does not exist in the existing region of the original logistics piece surface single image in the target image obtained after deletion, and after the target image is subsequently input into the model, useful image information cannot be extracted from the region of the model in the initial image, so that the purpose of filtering the logistics piece surface single image is achieved;

or, a filtering identifier may be added to the initial image as the target image, where the filtering identifier is used to indicate that the existing area of the logistics single image in the initial image is an invalid image area and needs to be filtered. After the target image is input into the model, the model determines the image area of the single image of the logistics piece surface according to the identification and directly ignores the area, so that the aim of filtering the single image of the logistics piece surface is fulfilled.

Alternatively, the single image of the material flow surface may be subjected to image overlay processing in the initial image, for example, the single image of the material flow surface is "blackened", the single image of the material flow surface is overlaid with Mosaic (Mosaic), and the like, so as to overlay the single image of the material flow surface, thereby achieving the purpose of filtering the single image of the material flow surface.

Or, a filtering identifier may be added to the single logistics surface image divided from the initial image, where the filtering identifier is used to indicate that the single logistics surface image is an invalid image, and at this time, the target image includes the initial image and the single logistics surface image to which the filtering identifier is added.

After the single image of the logistics piece surface is filtered to obtain the target image, the target image can be used for training a logistics piece identification model, wherein before the model is input for training, a worker can mark the logistics pieces and the number of the logistics pieces in the target image so as to check the identification result of the model on the number of the logistics pieces in the image. Specifically, semantic information configuration can be performed on pixels of the logistics pieces in the image, for example, semantic information of "logistics piece 1, where the total number of logistics pieces is 1" is configured for each pixel point of the logistics piece 1 in a certain image.

Specifically, images in a target image can be sequentially input into an initialized initial neural network model to be propagated in a forward direction, a loss function is calculated by combining the number of logistics pieces recognized from the input image output by the model, and back propagation is performed according to the loss function to optimize parameters of the model.

Specifically, the initial neural network model may specifically adopt different types of models such as a YOLOv3 model, a ResNet model, an R-CNN model, a Fast R-CNN model, a Faster R-CNN model, a Mask R-CNN model, and an SSD model.

The aforementioned neural network capable of identifying the logistics item surface list from the initial image, such as the target detection network, may also be referred to as a neural network model, which may be obtained by training different logistics item surface list images, and the specific training mode is similar to the model training mode provided above, and is not described herein again in detail.

It is understood that, in order to improve the training effect of the model, before the images are input to the model for training, the images may be subjected to image quality enhancement processing.

The image quality enhancement processing, which may also be referred to as image preprocessing, is used to improve the image quality of the image, so that the image has richer and more prominent image features, and thus the model has finer-grained and more accurate image recognition processing on the image after the image quality enhancement.

In the present application, in addition to performing image quality enhancement processing on a target image, that is, an effective image obtained after an initial image is subjected to filtering processing on a logistics surface single image (for example, the above-mentioned remaining image after the logistics surface single image is deleted from the initial image, the initial image with a filtering identifier added thereto, an image obtained after image overlay processing is performed on the logistics surface single image in the initial image, and an image set including the initial image and the logistics surface single image with the filtering identifier added thereto), before performing filtering processing on the logistics surface single image on the initial image, the initial image may be subjected to image enhancement processing, and a specific processing timing of the enhancement processing on the image may be adjusted according to actual needs, and is not limited herein.

Correspondingly, before training the neural network model of the nurse through the target image, the training method of the logistic item identification model provided by the application can further include:

In practical applications, the image enhancement processing may specifically include a normalization processing of an image format (or a unification processing of an image format), an overexposure processing, a sharpening processing, a contrast enhancement processing, an orientation correction processing, and the like.

The standardized processing of the image formats can be understood that different image formats may exist, for example, images with different image formats such as image size and different color spaces are subjected to processing such as scaling and color space conversion, and the image formats are unified, so that the models can conveniently perform consistent processing on input images, and the influence on the recognition and training of the models due to the different image formats is avoided.

The overexposure processing is to be understood that, in a part of the image, overexposure may occur due to limitations of shooting environment conditions such as a shooting angle and ambient light, or due to the influence of shooting technology, and therefore, the overexposure processing may be performed to correct overexposure, for example, by calculating an inverse (255-image) of the current image and then taking the smaller of the current image and the inverse image as a value of the current pixel position.

The sharpening process is mainly used for compensating the outline of an image, enhancing the edge of the image and the part with gray level jump to make the image clear, and is divided into two types, namely spatial domain processing and frequency domain processing. The sharpening process is to highlight the edge, the contour or the feature of some linear target elements in the image, for example, 8 neighborhood laplacian with center 5 can be used to convolve with the image to achieve the purpose of sharpening the enhanced image.

Contrast enhancement processing can be understood, in a partial image, the brightness values of most of the pixels are concentrated in a narrow dynamic interval, so that the contrast of the image is small, the color tone is single, and more image information is difficult to distinguish from the image, therefore, the contrast enhancement processing can be performed, the gray value range of the image is stretched or compressed into the gray value range of a display system, so as to expand the brightness difference, distinguish as many brightness levels as possible, improve and enhance the contrast of the image, for example, improve the contrast degree between a conveying belt and a logistics part. Specifically, contrast enhancement processing of a type such as gamma conversion processing, histogram equalization processing, and exponential conversion processing may be employed.

And posture correction processing, namely, in the partial image, the posture of the logistics piece may be inclined greatly, or the shooting angle of the whole image is inclined, and the posture of the image can be corrected through posture correction processing, such as rotation, stretching or geometric deformation of the image, so that the posture of the logistics piece in the image is more consistent with the posture of the logistics piece in the actual logistics operation.

Further, as another exemplary implementation manner, in the present application, referring to a flowchart of a contrast enhancement process of the present application shown in fig. 4, an image enhancement process performed in the present application may include, for a contrast enhancement aspect:

step S401, separating a first brightness channel image corresponding to a brightness channel in a color space from an image to be enhanced, wherein the image to be enhanced is an initial image or a target image;

it is understood that the following description will be given by taking the contrast enhancement processing on the initial image as an example, and in the present application, the separation of the channel images may be performed on the initial image before the contrast enhancement processing is performed.

It will be appreciated that the image itself may be implemented in different color spaces, each color space having different image channels, which may represent features of the image in different aspects.

For example, a common Red-Green-Blue (RGB) color space, which has three image channels of R channel, G channel and B channel, indicates the color characteristics of an image from Red, Green and Blue.

For another example, in the YUV color space, a Y channel is used to indicate brightness of an image, a U channel is used to indicate hue of the image, and a V channel is used to indicate color saturation of the image.

In the application, the corresponding brightness image can be extracted from the brightness channel in the initial image, for example, the single-channel image corresponding to the Y channel is extracted from the YUV color space, so that the brightness of the pixel point of the image is directly reflected, the light intensity information of the image is reflected, the subsequent contrast enhancement processing is facilitated, and the enhancement effect is improved.

If the initial image does not have a luminance channel, the color space conversion may be performed, for example, the color space conversion between the RGB color space and the YUV color space may refer to the following conversion formula:

step S402, carrying out contrast enhancement processing on the first brightness channel image to obtain a second brightness channel image;

after a single channel image of the luminance channel is obtained, the image may then be subjected to contrast enhancement processing.

Taking histogram Equalization processing as an example, in the present application, in practical application, a Contrast Limited histogram Equalization (CLAHE) method or other Histogram Equalization (HE) methods may be specifically adopted.

And S403, fusing the second brightness channel image with the channel images except the first brightness channel image in the image to be enhanced.

After the contrast enhancement processing is carried out on the brightness channel image, the single-channel image obtained by the processing can be fused with the channel image which is not subjected to the contrast enhancement processing in the initial image to obtain the image in the form of the normal image.

For example, the Y-channel image in the YUV color space is subjected to contrast enhancement processing to obtain a Y 'channel image, and the remaining U-channel images and V-channel images are not subjected to contrast enhancement processing, so that the Y', U, and V-channel images can be fused to obtain a complete image as an output.

Further, as another exemplary implementation manner, in the present application, a filtering process may be further introduced, and in particular, the training method of the logistic item identification model provided in the present application may further include, before filtering the logistic item surface single image, the following steps:

image information having a high-frequency signal characteristic is subjected to a filtering process in the initial image.

In the application, the whole initial image can be subjected to filtering processing to smooth image noise (with high-frequency signal characteristics) existing in the image, so that the influence of the noise on the image quality is further reduced.

In practical applications, the barcode, the two-dimensional code, and the text on the material distribution list generally have characteristics of a high-frequency signal in an image (for example, edge contours of the barcode, the two-dimensional code, and the text tend to cause drastic changes in image characteristics such as image brightness and gray scale, and at this time, image information of the edge contours may be regarded as a high-frequency signal). For some images with obvious single high-frequency signal characteristics of the logistics piece surface, the single images of the logistics piece surface in the images can be filtered in advance, so that the initial images with the condition are filtered, subsequent identification and filtering processing of the single images of the logistics piece surface are not needed, and the effect of assisting in filtering the single images of the logistics piece surface is achieved.

Secondly, certain noise may be generated by the above image enhancement processing, especially the contrast enhancement processing, and the filtering processing performed at this time may also have a smoothing effect on the noise to a certain extent, so as to further reduce the influence of the noise on the image quality.

In practical application, the filtering process may specifically adopt a median filtering process, a gaussian filtering process, or other types of filtering processes.

Further, as another exemplary implementation manner, in the median filtering process, the smoothing ratio is K/N, the maximum labeling error of the median filtering process is K/M, K is used to indicate the kernel size of the median filtering process, N is used to indicate the preset minimum pixel size of the logistics piece surface list in the initial image, and M is used to indicate the preset average pixel size of the logistics piece.

Further, as another exemplary implementation manner, in the present application, the model trained by the training method for the logistics piece identification model provided by the present application may be specifically applied to a weighing link in an automated sorting production line, where the above-mentioned weighing link in the automated sorting production line may have a weighing abnormality, and the model trained by the training method for the logistics piece identification model of the present application efficiently determines the number of logistics pieces at the weighing device, and determines whether the weighing state of the weighing device is normal according to the number of the logistics pieces, for example: if the number of the logistics pieces is larger than 1, determining that the weighing state is an abnormal weighing state; and if the number of the physical distribution pieces is equal to 1, determining that the weighing state is a normal weighing state.

That is, after the physical distribution piece recognition model is obtained through training, the method may further include:

After the model inputs the image to be recognized, the image features can be extracted through a corresponding network structure in the model (for example, the network structure of the model can generally comprise a convolutional layer and a full-connection layer, and relevant parameters of the network structure of the model can be optimized and adjusted in the training process of the model), and finally, the recognition result of the model is summarized and output through the feature extraction of the multi-layer network structure.

So, combine image recognition, judge from the image aspect more directly perceivedly, accurately whether normal waiting to weigh the state of weighing of commodity circulation piece, especially when the commodity circulation piece appears in the upper reaches link of weighing equipment and sorts unusually, when the commodity circulation piece transmission more than two to weighing equipment, can detect this abnormal conditions the very first time, carry out accurate reply and handle, further guaranteed automatic sorting operation's stability and letter sorting efficiency.

Correspondingly, the initial image for training the model is specifically an image obtained by shooting a training logistics piece on the weighing equipment.

In the application, the weighing device is a device configured in an automatic sorting operation of a logistics company and used for weighing logistics pieces on an automatic sorting production line so as to check whether the identification weight and the actual weight of the logistics pieces are consistent.

The weighing device can be a dynamic weighing device, and the dynamic weighing device can be used for weighing the logistics piece while transporting the logistics piece, namely, the logistics piece in a transportation state can be dynamically weighed. For example, the Weighing apparatus may be embodied as a Dynamic Weighing System (DWS), which is also known as a DWS Dynamic scale.

Wherein, the weighing equipment can be configured in the automatic sorting assembly line and weigh the logistics piece passing through or placed. During weighing, the logistics piece is placed on a weighing platform, a weighing surface or a weighing area of the weighing device, and the weighing device senses the weight of the logistics piece through the sensor and converts the weight into weight data in a data form.

In order to better implement the training method of the logistics piece identification model provided by the application, the application also provides a training device of the logistics piece identification model.

Referring to fig. 5, fig. 5 is a schematic structural diagram of a training device of a logistic item identification model according to the present application, in which the training device 500 of the logistic item identification model specifically includes the following structure:

the acquisition unit 501 is configured to acquire an initial image, where the initial image is a plurality of images obtained by shooting physical distribution pieces for training, and the number of the physical distribution pieces existing in the image is marked on the initial image;

the identification unit 502 is configured to identify a logistics piece surface list in the initial image to obtain a logistics piece surface list image;

the filtering unit 503 is configured to filter the single image of the logistics part surface in the initial image to obtain a target image;

the training unit 504 is configured to train the initial neural network model through the target image, and use the trained model as a logistics piece recognition model, where the logistics piece recognition model is used to recognize the number of logistics pieces existing in the input image.

In an exemplary implementation manner, the identifying unit 502 is specifically configured to:

In yet another exemplary implementation, the apparatus further includes an enhancing unit 505 for:

In another exemplary implementation manner, the enhancing unit 505 is specifically configured to:

In yet another exemplary implementation, the apparatus further includes a filtering unit 506, configured to:

In another exemplary implementation, the filtering process is specifically a median filtering process, a smoothing ratio of the median filtering process is K/N, a maximum labeling error of the median filtering process is K/M, K is used to indicate a kernel size of the median filtering process, N is used to indicate a single minimum pixel size of a material flow in an initial image, and M is used to indicate a preset average pixel size of the material flow.

In another exemplary implementation manner, the initial image is an image obtained by shooting a weighing device, the weighing device is a dynamic weighing device on the logistics sorting line, the weighing device is used for weighing the logistics item to be transported while transporting the logistics item to be transported, and the apparatus further includes an application unit for:

Referring to fig. 6, fig. 6 shows a schematic structural diagram of a training device of a physical distribution piece recognition model according to the present application, specifically, the training device of a physical distribution piece recognition model according to the present application includes a processor 601, a memory 602, and an input/output device 603, where the processor 601 is configured to implement steps of a training method of a physical distribution piece recognition model according to any embodiment corresponding to fig. 1 to 4 when executing a computer program stored in the memory 602, and the memory 602 is configured to store a computer program required by the processor 501 to execute the training method of a physical distribution piece recognition model according to any embodiment corresponding to fig. 1 to 4.

Illustratively, a computer program may be partitioned into one or more modules/units, which are stored in the memory 602 and executed by the processor 601 to accomplish the present application. One or more modules/units may be a series of computer program instruction segments capable of performing certain functions, the instruction segments being used to describe the execution of a computer program in a computer device.

The training device of the logistics item identification model can include, but is not limited to, a processor 601, a memory 602, and an input-output device 603. Those skilled in the art will appreciate that the illustration is only an example of the training device of the physical distribution piece recognition model, and does not constitute a limitation of the training device of the physical distribution piece recognition model, and may include more or less components than those illustrated, or combine some components, or different components, for example, the device may further include a network access device, a bus, etc., and the processor 601, the memory 602, the input and output device 603, and the network access device, etc., are connected through the bus.

The Processor 601 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, discrete hardware component, etc. The general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, the processor is a control center of the training device of the physical distribution piece recognition model, and various interfaces and lines are used for connecting various parts of the whole device.

The memory 602 may be used for storing computer programs and/or modules, and the processor 601 may implement various functions of the computer apparatus by executing or executing the computer programs and/or modules stored in the memory 602 and calling data stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program (such as an image playing function) required by at least one function, and the like; the storage data area may store data (such as image data and the like) created from use of a training apparatus of the physical distribution piece recognition model, and the like. In addition, the memory may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), at least one magnetic disk storage device, a Flash memory device, or other volatile solid state storage device.

The processor 601, when executing the computer program stored in the memory 602, may specifically implement the following functions:

It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the training apparatus and the device for the logistics item identification model and the corresponding units thereof described above may refer to the description of the training method for the logistics item identification model in any embodiment corresponding to fig. 1 to fig. 4, and are not described herein again in detail.

It will be understood by those skilled in the art that all or part of the steps of the methods of the above embodiments may be performed by instructions or by associated hardware controlled by the instructions, which may be stored in a computer readable storage medium and loaded and executed by a processor.

For this reason, the present application provides a computer-readable storage medium, where a plurality of instructions are stored, where the instructions can be loaded by a processor to execute steps in the training method for the physical distribution piece identification model in any embodiment corresponding to fig. 1 to fig. 4 in the present application, and specific operations may refer to descriptions of the training method for the physical distribution piece identification model in any embodiment corresponding to fig. 1 to fig. 4, which are not described herein again.

Wherein the computer-readable storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.

Since the instructions stored in the computer-readable storage medium can execute the steps in the method for training the physical distribution piece identification model in any embodiment of the present application, such as those shown in fig. 1 to fig. 4, the beneficial effects that can be achieved by the method for training the physical distribution piece identification model in any embodiment of the present application, such as those shown in fig. 1 to fig. 4, can be achieved, and are described in detail in the foregoing description, and are not repeated herein.

The method, the device, the equipment and the computer-readable storage medium for training the logistics item identification model provided by the application are introduced in detail, and a specific example is applied in the description to explain the principle and the implementation of the application, and the description of the above embodiment is only used for helping to understand the method and the core idea of the application; meanwhile, for those skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.

Claims

1. A training method of a logistics piece recognition model is characterized by comprising the following steps:

acquiring an initial image, wherein the initial image is a plurality of images obtained by shooting physical distribution pieces for training, and the number of the physical distribution pieces existing in the image is marked on the initial image;

training an initial neural network model through the target image, and taking the trained model as a physical distribution piece recognition model, wherein the physical distribution piece recognition model is used for recognizing the number of physical distribution pieces existing in the input image.

2. The method of claim 1, wherein the identifying the logistics item list in the initial image and obtaining the logistics item list image comprises:

calling a logistics item bill identification configured by a user, wherein the logistics item bill identification is used for identifying an existing area of the logistics item bill in the initial image;

and determining the logistics piece surface single image in the initial image according to the logistics piece surface single identification.

3. The method of claim 1, wherein prior to training an initial neural network model with the target image, the method further comprises:

4. The method of claim 3, wherein the image enhancement processing of the initial image or the target image comprises:

separating a first brightness channel image corresponding to a brightness channel in a color space from an image to be enhanced, wherein the image to be enhanced is the initial image or the target image;

performing contrast enhancement processing on the first brightness channel image to obtain a second brightness channel image;

5. The method of claim 1, wherein the filtering the single image of the logistics surface in the initial image to obtain the target image comprises:

6. The method according to claim 5, wherein the filtering process is specifically a median filtering process, a smoothing ratio of the median filtering process is K/N, a maximum labeling error of the median filtering process is K/M, K is used for indicating a kernel size of the median filtering process, N is used for indicating a single minimum pixel size of a logistics surface preset in the initial image, and M is used for indicating a preset average pixel size of the logistics surface.

7. The method according to any one of claims 1 to 6, wherein the initial image is an image obtained by shooting a weighing device, the weighing device is a dynamic weighing device on a logistics sorting line, the weighing device is used for weighing a logistics piece to be transported while transporting the logistics piece, the initial neural network model is trained through the target image, and after the trained model is used as a logistics piece identification model, the method further comprises:

acquiring an image to be identified, which is obtained by shooting the weighing equipment;

inputting the image to be identified into the logistics piece identification model so as to identify the number of logistics pieces existing in the image to be identified;

and when the number of the physical distribution pieces in the image to be identified is more than 1, determining that the weighing equipment has an abnormal weighing state.

8. A training device for a logistics piece recognition model is characterized by comprising:

the device comprises an acquisition unit, a storage unit and a processing unit, wherein the acquisition unit is used for acquiring an initial image, the initial image is a plurality of images obtained by shooting physical distribution pieces for training, and the number of the physical distribution pieces existing in the image is marked on the initial image;

9. Training device of a logistic item identification model, characterized in that it comprises a processor and a memory, in which a computer program is stored, which when called by the processor executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a plurality of instructions adapted to be loaded by a processor to perform the method of any one of claims 1 to 7.