CN114943864A - Tobacco leaf grading method integrating attention mechanism and convolutional neural network model - Google Patents
Tobacco leaf grading method integrating attention mechanism and convolutional neural network model Download PDFInfo
- Publication number
- CN114943864A CN114943864A CN202210666171.2A CN202210666171A CN114943864A CN 114943864 A CN114943864 A CN 114943864A CN 202210666171 A CN202210666171 A CN 202210666171A CN 114943864 A CN114943864 A CN 114943864A
- Authority
- CN
- China
- Prior art keywords
- convolution
- tobacco leaf
- channel
- attention
- channels
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 235000002637 Nicotiana tabacum Nutrition 0.000 title claims abstract description 69
- 238000000034 method Methods 0.000 title claims abstract description 33
- 230000007246 mechanism Effects 0.000 title claims abstract description 32
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 17
- 244000061176 Nicotiana tabacum Species 0.000 title 1
- 241000208125 Nicotiana Species 0.000 claims abstract description 68
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims abstract description 4
- 238000011176 pooling Methods 0.000 claims description 59
- 238000004364 calculation method Methods 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 7
- 230000002146 bilateral effect Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 230000004931 aggregating effect Effects 0.000 claims description 2
- 230000006835 compression Effects 0.000 claims description 2
- 238000007906 compression Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 8
- 230000008569 process Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 239000000779 smoke Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a tobacco leaf grading method integrating an attention mechanism and a convolutional neural network model, which comprises the following steps of: s1, acquiring a tobacco leaf picture, constructing a tobacco leaf sample data set based on the national flue-cured tobacco leaf grading standard and an expert experience method, and preprocessing; step S2: based on a convolutional neural network, introducing a depth separable convolution method, decomposing standard convolution into depth convolution and point-by-point convolution, constructing a depth separable convolution model, and training based on a preprocessed tobacco leaf sample data set; and step S3, introducing a method of an attention mechanism module, and optimizing the depth separable convolution model to obtain a final tobacco leaf grading model. The invention can realize accurate and efficient tobacco leaf grading.
Description
Technical Field
The invention relates to the field of machine learning, in particular to a tobacco leaf grading method integrating an attention mechanism and a convolutional neural network model.
Background
In the tobacco industry chain, grading of flue-cured tobacco is an important link and is related to the practical benefit of tobacco growers. The transmission manual grading is recognized by naked eyes, the error is large, and the efficiency is low. And forty-two levels are adopted in China, and the division of the refinement level brings greater difficulty for detection and grading of tobacco leaves. With the continuous development of machine vision, the tobacco leaf grading technology is changing from inefficient manual grading to computer automatic grading. With the great increase of the detection amount, the traditional manual grading algorithm cannot meet the requirement of rapid production, and the application of the machine vision technology and the deep learning to the grading identification of the tobacco leaf images has great research space and application value. According to a large amount of research data and years of accumulated practice in the tobacco industry of China, the tobacco leaves at different parts of the tobacco leaf structure generally have different appearance characteristics, and the chemical components, the smoke quality and the physical characteristics of the tobacco leaves with different colors have larger differences. The quality of the tobacco leaves is closely related to the appearance color and the appearance shape of the tobacco leaves and has regularity.
Disclosure of Invention
In view of this, the invention aims to provide a tobacco leaf grading method combining an attention mechanism and a convolutional neural network model, which can realize accurate and efficient tobacco leaf grading.
In order to achieve the purpose, the invention adopts the following technical scheme:
a tobacco leaf grading method integrating an attention mechanism and a convolutional neural network model comprises the following steps:
s1, acquiring a tobacco leaf picture, constructing a tobacco leaf sample data set based on the national flue-cured tobacco leaf grading standard and an expert experience method, and preprocessing;
step S2: based on a convolutional neural network, introducing a depth separable convolution method, decomposing standard convolution into depth convolution and point-by-point convolution, constructing a depth separable convolution model, and training based on a preprocessed tobacco leaf sample data set;
and step S3, introducing a method of an attention mechanism module, and optimizing the depth separable convolution model to obtain a final tobacco leaf grading model.
Further, the preprocessing comprises cutting, denoising and standardization processing of the tobacco leaf image, and specifically comprises the following steps:
cutting the tobacco leaf image, and denoising the cut image by adopting a bilateral filtering denoising method; after denoising, standardizing the tobacco leaf image, wherein the formula of the standardization is as follows:
where u is the mean of the image, x represents the image matrix, σ represents the standard deviation, N represents the number of pixels of the image x, and adjusted _ stddev is the adjustment coefficient.
Further, the depth separable convolution model, the layer 1, is depth convolution, performs calculation of spatial correlation, and performs effective extraction on features; the layer 2 is a point-by-point convolution, and the number of output characteristic channels is adjusted through linear combination of input channels, specifically:
firstly, deep convolution operation is carried out, each input channel is filtered by a single filter, all multi-channel feature maps from the upper layer are split into feature maps of single channels, single-channel convolution is carried out on the feature maps respectively, and then the feature maps are stacked together again;
setting the input and output characteristic graphs to be the same, and adopting convolution kernel as D k ·D k ，D k Representing the length and width of the convolution kernel k, M being the number of input channels, N being the number of output channels, D f And (3) representing the width and height of the input and output characteristic graphs, performing convolution on the M convolution kernels and the M channels respectively, wherein the calculation amount of the deep convolution is as follows (2):
D k ·D k ·M·D f ·D f (2)
to is D f ·D f M is combined by convolution, the convolution of 1.1. N is carried out, the conventional convolution is combined, and the calculation amount of the point-by-point convolution is as shown in formula (3):
M·N·D f ·D f (3)
the total computation of the depth separable convolution is the sum of the depth convolution and the point-by-point convolution, as shown in equation (4):
D k ·D k ·M·D f ·D f +M·N·D f ·D f (4)
and the calculated amount of the standard convolution is as follows:
D k ·D k ·M·D f ·D f ·N (5)
the ratio of the depth separable convolution to the standard convolution is as follows (6):
further, the attention mechanism module adopts a convolution attention module, combines a space and a channel attention mechanism module, and is particularly suitable for the attention mechanism module
The channel attention module firstly inputs the characteristics, respectively performs maximum pooling and average pooling, then respectively performs element-by-element addition operation on the characteristics output by the shared full-connection layer through the multilayer perceptron, and generates a final channel attention weight through an activation function; finally, performing element-by-element multiplication operation on the channel attention weight and the input feature weight to generate input features required by the space attention module;
the spatial attention module takes the feature map output by the channel attention module as an input feature map of the module; firstly, performing maximum pooling and average pooling based on channels, then performing merging operation on the two results based on the channels, performing convolution operation to reduce the dimension into 1 channel, and generating space attention characteristics through an activation function; and finally, multiplying the space attention characteristic and the input characteristic of the module to obtain the finally generated characteristic.
Further, the channel attention mechanism module compresses the feature map in the spatial dimension to obtain a one-dimensional vector, then performs operation, considers not only the average pooling but also the maximum pooling when performing compression in the spatial dimension, the average pooling and the maximum pooling are used for aggregating spatial information of the feature map, sends the spatial information to a shared network, compresses the spatial dimension of the input feature map, sums and combines element by element to generate the channel attention map,
the channel attention calculation formula is shown in formula (7)
In formula 7:
and
respectively representing the average pooling characteristic and the maximum pooling characteristic of the channels; MLP represents a multi-layer perceptron; avgpoolRepresents average pooling; max Pool denotes maximum pooling; w 1 And W 0 Parameters in the represented multi-layer perceptron; sigma represents sigmoid activation function
Further, the space attention mechanism module compresses the channel, and performs average pooling and maximum pooling on the channel dimension respectively; the operation of maximum pooling MaxPool is to extract the maximum value on a channel, and the extraction times are the height multiplied by the width; the operation of average value pooling AvgPool is to extract an average value on a channel, and the extraction times are also height times width; and then combining the extracted feature maps to obtain a 2-channel feature map, which specifically comprises the following steps:
in the formula:
and
respectively representing the average pooling characteristic and the maximum pooling characteristic of the channels; avg Pool represents average pooling; max Pool represents maximum pooling; 7 × 7 denotes the size of the convolution kernel; σ denotes a sigmoid activation function.
Compared with the prior art, the invention has the following beneficial effects:
the invention establishes a model integrating an attention mechanism and a convolution neural network, introduces a depth separable convolution method, simultaneously adds the attention mechanism, continuously focuses on the most discriminative region to realize the classification of the image, has better prediction effect and effectively improves the identification accuracy.
Drawings
FIG. 1 is an original image of tobacco leaves in an embodiment of the present invention;
FIG. 2 is a block diagram of a filtered denoised image according to an embodiment of the present invention;
FIG. 3 is a standard convolution process in accordance with an embodiment of the present invention;
FIG. 4 is a process for depth separable convolution according to an embodiment of the present invention;
FIG. 5 is a diagram of an attention module in accordance with an embodiment of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
Referring to the drawings, the invention provides a tobacco leaf grading method integrating an attention mechanism and a convolutional neural network model, which comprises the following steps:
s1, acquiring a tobacco leaf picture, constructing a tobacco leaf sample data set based on the national flue-cured tobacco leaf grading standard and an expert experience method, and preprocessing;
in this example, the tobacco leaf samples studied were all from fujian gming with a collection time of 2021 years and 6 months. The tobacco leaves growing in the upper part, the middle part and the lower part are respectively selected for research, each tobacco leaf sample is relatively good, poor and general tobacco leaves manually selected by tobacco growers, and the tobacco leaf samples are sorted and graded by tobacco experts into three groups, namely B4F, C2F and X2F, and the grading mainly depends on the appearance characteristics of the tobacco leaves.
Shoot 600 many high definition tobacco leaf images through the high appearance of shooing, select 420 qualified tobacco leaf images and use as this experiment, the standard condition of tobacco leaf sample is as follows:
the pictures of the samples of the tobacco leaves with different grades are shown in figure 1;
in this embodiment, the preprocessing includes cutting, denoising, and standardizing the tobacco leaf image, and specifically includes: cutting the tobacco leaf image, and denoising the cut image by adopting a bilateral filtering denoising method; after denoising, standardizing the tobacco leaf image, wherein the formula of the standardization is as follows:
where u is the mean of the image, x represents the image matrix, σ represents the standard deviation, N represents the number of pixels of the image x, and adjusted _ stddev is the adjustment coefficient.
Step S2: based on a convolutional neural network, introducing a deep separable convolution method, decomposing the standard convolution into a deep convolution and a point-by-point convolution, constructing a deep separable convolution model, and training based on a preprocessed tobacco leaf sample data set;
in this embodiment, the calculation process of the convolution model is as shown in fig. 3, the scale and the calculation amount are huge, and it is necessary to perform convolution operation on all input features to obtain a series of outputs. Further, a depth separable convolution model is adopted, the calculation process is shown in fig. 4, the layer 1 is depth convolution, and the calculation of spatial correlation is mainly carried out to effectively extract features; the layer 2 is point-by-point convolution, and the number of output characteristic channels is adjusted mainly through linear combination of input channels.
Firstly, deep convolution operation is carried out, each input channel is filtered by a single filter, all multi-channel feature maps from the upper layer are split into feature maps of single channels, single-channel convolution is carried out on the feature maps respectively, and then the feature maps are stacked together again;
let the input and output characteristic graphs be the same, and use convolution kernel as D k ·D k ，D k Representing the length and width of the convolution kernel k, M being the number of input channels, N being the number of output channels, D f And (3) representing the width and the height of the input and output characteristic graphs, performing convolution on the M convolution kernels and the M channels respectively, wherein the deep convolution calculation quantity is as shown in a formula (2):
D k ·D k ·M·D f ·D f (2)
to is D f ·D f M is combined by convolution, the convolution of 1.1. N is carried out, the conventional convolution is combined, and the calculation amount of the point-by-point convolution is as shown in formula (3):
M·N·D f ·D f (3)
the total computation of the depth separable convolution is the sum of the depth convolution and the point-by-point convolution, as shown in equation (4):
D k ·D k ·M·D f ·D f +M·N·D f ·D f (4)
and the calculated amount of the standard convolution is as shown in formula (5):
D k ·D k ·M·D f ·D f ·N (5)
the ratio of the depth separable convolution to the standard convolution is as follows (6):
and step S3, introducing a method of an attention mechanism module, optimizing the depth separable convolution model, and obtaining a final tobacco leaf grading model.
In this embodiment, the attention mechanism module is a convolution attention module, and combines a space and a channel attention mechanism module, specifically
The channel attention module firstly inputs the characteristics, respectively performs maximum pooling and average pooling, then respectively performs element-by-element addition operation on the characteristics output by the shared full-connection layer through the multilayer perceptron, and generates a final channel attention weight through an activation function; finally, performing element-by-element multiplication operation on the channel attention weight and the input feature weight to generate input features required by the space attention module;
the spatial attention module takes the feature map output by the channel attention module as an input feature map of the module; firstly, performing maximum pooling and average pooling based on channels, then performing merging operation on the two results based on the channels, performing convolution operation to reduce the dimension into 1 channel, and generating space attention characteristics through an activation function; and finally, multiplying the space attention characteristic and the input characteristic of the module to obtain the finally generated characteristic.
Preferably, the channel attention mechanism module compresses the feature map in the spatial dimension to obtain a one-dimensional vector, and then operates, when compressing in the spatial dimension, not only considering average pooling but also considering maximum pooling, the average pooling and the maximum pooling are used to aggregate spatial information of the feature map and are sent to a sharing network, the spatial dimension of the input feature map is compressed, and element-by-element summation and combination are performed to generate the channel attention map,
the channel attention calculation formula is shown in formula (7)
In formula 7:
and
respectively representing the average pooling characteristic and the maximum pooling characteristic of the channels; MLP denotes a multilayer perceptron; AvgPool represents average pooling; max Pool denotes maximum pooling; w 1 And W 0 Parameters in the represented multi-layer perceptron; sigma represents sigmoid activation function
Preferably, the spatial attention mechanism module compresses the channels, and performs average pooling and maximum pooling on the channel dimensions respectively; the operation of maximum pooling MaxPool is to extract the maximum value on a channel, and the extraction times are height multiplied by width; the operation of average value pooling AvgPool is to extract an average value on a channel, and the extraction times are also height times width; then, combining the extracted feature maps to obtain a 2-channel feature map, which specifically comprises the following steps:
in the formula:
and
respectively representing the average pooling characteristic and the maximum pooling characteristic of the channels; avg Pool represents average pooling; max Pool denotes Max PoolMelting; 7 × 7 denotes the size of the convolution kernel; σ denotes the sigmoid activation function.
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.
Claims (6)
1. A tobacco leaf grading method integrating an attention mechanism and a convolutional neural network model is characterized by comprising the following steps of:
s1, acquiring a tobacco leaf picture, constructing a tobacco leaf sample data set based on the national flue-cured tobacco leaf grading standard and an expert experience method, and preprocessing;
step S2: based on a convolutional neural network, introducing a depth separable convolution method, decomposing standard convolution into depth convolution and point-by-point convolution, constructing a depth separable convolution model, and training based on a preprocessed tobacco leaf sample data set;
and step S3, introducing a method of an attention mechanism module, optimizing the depth separable convolution model, and obtaining a final tobacco leaf grading model.
2. The tobacco leaf grading method combining the attention mechanism and the convolutional neural network model according to claim 1, wherein the preprocessing comprises cropping, denoising and standardizing the tobacco leaf image, and specifically comprises:
clipping the tobacco leaf image, and denoising the clipped image by adopting a bilateral filtering denoising method; after denoising, standardizing the tobacco leaf image, wherein the formula of the standardization is as follows:
where u is the mean of the image, x represents the image matrix, σ represents the standard deviation, N represents the number of pixels of the image x, and adjusted _ stddev is the adjustment coefficient.
3. The tobacco leaf grading method integrating the attention mechanism and the convolutional neural network model according to claim 1, wherein the depth separable convolutional model is adopted, the layer 1 is depth convolution, spatial correlation calculation is performed, and features are effectively extracted; layer 2 is a point-by-point convolution, and the number of output characteristic channels is adjusted through linear combination of input channels, specifically:
firstly, deep convolution operation is carried out, each input channel is filtered by a single filter, all multi-channel feature maps from the upper layer are split into feature maps of single channels, single-channel convolution is carried out on the feature maps respectively, and then the feature maps are stacked together again;
let the input and output characteristic graphs be the same, and use convolution kernel as D k ·D k ，D k Representing the length and width of a convolution kernel k, M being the number of input channels, N being the number of output channels, D f And (3) representing the width and height of the input and output characteristic graphs, performing convolution on the M convolution kernels and the M channels respectively, wherein the calculation amount of the deep convolution is as follows (2):
D k ·D k ·M·D f ·D f (2)
to is D f ·D f M is combined by convolution, the convolution of 1.1. N is carried out, the conventional convolution is combined, and the calculation amount of the point-by-point convolution is as shown in formula (3):
M·N·D f ·D f (3)
the total computation of the depth separable convolution is the sum of the depth convolution and the point-by-point convolution, as shown in equation (4):
D k ·D k ·M·D f ·D f +M·N·D f ·D f (4)
and the calculated amount of the standard convolution is as follows:
D k ·D k ·M·D f ·D f ·N (5)
the ratio of the depth separable convolution to the standard convolution is as follows (6):
4. the tobacco leaf grading method combining an attention mechanism and a convolutional neural network model according to claim 1, wherein the attention mechanism module is a convolutional attention module, and is combined with a space and channel attention mechanism module, in particular
The channel attention module firstly inputs the characteristics, respectively performs maximum pooling and average pooling, then respectively performs element-by-element addition operation on the characteristics output by the shared full-connection layer through the multilayer perceptron, and generates a final channel attention weight through an activation function; finally, performing element-by-element multiplication operation on the channel attention weight and the input feature weight to generate input features required by the space attention module;
the spatial attention module takes the feature map output by the channel attention module as an input feature map of the module; firstly, performing maximum pooling and average pooling based on channels, then performing merging operation on the two results based on the channels, performing convolution operation to reduce the dimension into 1 channel, and then generating spatial attention characteristics by an activation function; and finally, multiplying the space attention characteristic and the input characteristic of the module to obtain the finally generated characteristic.
5. The tobacco leaf grading method combining attention mechanism and convolutional neural network model according to claim 4, wherein the channel attention mechanism module compresses the feature map in the spatial dimension to obtain a one-dimensional vector, and then performs the operation, when performing the compression in the spatial dimension, not only the average pooling but also the maximum pooling are considered, the average pooling and the maximum pooling are used for aggregating the spatial information of the feature map, and the spatial information is sent to a shared network, the spatial dimension of the input feature map is compressed, and the element-by-element summation and combination are performed to generate the channel attention map, and the channel attention calculation formula is as shown in formula (7)
In formula 7:
and
respectively representing the average pooling characteristic and the maximum pooling characteristic of the channels; MLP denotes a multilayer perceptron; avg Pool represents average pooling; max Pool denotes maximum pooling; w 1 And W 0 Parameters in the represented multi-layer perceptron; σ denotes the sigmoid activation function.
6. The tobacco leaf grading method integrating the attention mechanism and the convolutional neural network model according to claim 4, wherein the spatial attention mechanism module compresses channels, and performs average pooling and maximum pooling on channel dimensions respectively; the operation of maximum pooling MaxPool is to extract the maximum value on a channel, and the extraction times are height multiplied by width; the operation of average value pooling AvgPool is to extract an average value on a channel, and the extraction times are also height times width; then, combining the extracted feature maps to obtain a 2-channel feature map, which specifically comprises the following steps:
in the formula:
and
respectively representing the average pooling characteristic and the maximum pooling characteristic of the channels; avg Pool represents average pooling; max Pool denotes maximum pooling; 7 × 7 denotes the size of the convolution kernel; σ denotes a sigmoid activation function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210666171.2A CN114943864A (en) | 2022-06-14 | 2022-06-14 | Tobacco leaf grading method integrating attention mechanism and convolutional neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210666171.2A CN114943864A (en) | 2022-06-14 | 2022-06-14 | Tobacco leaf grading method integrating attention mechanism and convolutional neural network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114943864A true CN114943864A (en) | 2022-08-26 |
Family
ID=82908874
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210666171.2A Pending CN114943864A (en) | 2022-06-14 | 2022-06-14 | Tobacco leaf grading method integrating attention mechanism and convolutional neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114943864A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115953384A (en) * | 2023-01-10 | 2023-04-11 | 杭州首域万物互联科技有限公司 | On-line detection and prediction method for tobacco morphological parameters |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111199217A (en) * | 2020-01-09 | 2020-05-26 | 上海应用技术大学 | Traffic sign identification method and system based on convolutional neural network |
| CN111915580A (en) * | 2020-07-27 | 2020-11-10 | 深圳市识农智能科技有限公司 | Tobacco leaf grading method, system, terminal equipment and storage medium |
| CN112767423A (en) * | 2021-02-05 | 2021-05-07 | 吉林师范大学 | Remote sensing image building segmentation method based on improved SegNet |
| CN113177465A (en) * | 2021-04-27 | 2021-07-27 | 江苏科技大学 | SAR image automatic target recognition method based on depth separable convolutional neural network |
| CN113192633A (en) * | 2021-05-24 | 2021-07-30 | 山西大学 | Stomach cancer fine-grained classification method based on attention mechanism |
| CN113344188A (en) * | 2021-06-18 | 2021-09-03 | 东南大学 | Lightweight neural network model based on channel attention module |
| CN113469233A (en) * | 2021-06-23 | 2021-10-01 | 临沂大学 | Tobacco leaf automatic grading method and system based on deep learning |
| CN114266337A (en) * | 2021-11-16 | 2022-04-01 | 中国烟草总公司职工进修学院 | Intelligent tobacco leaf grading model based on residual error network and grading method using model |
-
2022
- 2022-06-14 CN CN202210666171.2A patent/CN114943864A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111199217A (en) * | 2020-01-09 | 2020-05-26 | 上海应用技术大学 | Traffic sign identification method and system based on convolutional neural network |
| CN111915580A (en) * | 2020-07-27 | 2020-11-10 | 深圳市识农智能科技有限公司 | Tobacco leaf grading method, system, terminal equipment and storage medium |
| CN112767423A (en) * | 2021-02-05 | 2021-05-07 | 吉林师范大学 | Remote sensing image building segmentation method based on improved SegNet |
| CN113177465A (en) * | 2021-04-27 | 2021-07-27 | 江苏科技大学 | SAR image automatic target recognition method based on depth separable convolutional neural network |
| CN113192633A (en) * | 2021-05-24 | 2021-07-30 | 山西大学 | Stomach cancer fine-grained classification method based on attention mechanism |
| CN113344188A (en) * | 2021-06-18 | 2021-09-03 | 东南大学 | Lightweight neural network model based on channel attention module |
| CN113469233A (en) * | 2021-06-23 | 2021-10-01 | 临沂大学 | Tobacco leaf automatic grading method and system based on deep learning |
| CN114266337A (en) * | 2021-11-16 | 2022-04-01 | 中国烟草总公司职工进修学院 | Intelligent tobacco leaf grading model based on residual error network and grading method using model |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115953384A (en) * | 2023-01-10 | 2023-04-11 | 杭州首域万物互联科技有限公司 | On-line detection and prediction method for tobacco morphological parameters |
| CN115953384B (en) * | 2023-01-10 | 2024-02-02 | 杭州首域万物互联科技有限公司 | Online detection and prediction method for morphological parameters of tobacco leaves |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113537138A (en) | Traffic sign identification method based on lightweight neural network | |
| CN110826684A (en) | Convolutional neural network compression method, convolutional neural network compression device, electronic device, and medium | |
| CN111950655A (en) | Image aesthetic quality evaluation method based on multi-domain knowledge driving | |
| CN109858575A (en) | Data classification method based on convolutional neural networks | |
| CN113420794B (en) | Binaryzation Faster R-CNN citrus disease and pest identification method based on deep learning | |
| CN110956212A (en) | Threshing quality detection method based on visual feature fusion | |
| CN112767385B (en) | No-reference image quality evaluation method based on significance strategy and feature fusion | |
| CN110909928B (en) | Energy load short-term prediction method and device, computer equipment and storage medium | |
| CN111652307A (en) | Intelligent nondestructive identification method and device for redwood furniture based on convolutional neural network | |
| CN114943864A (en) | Tobacco leaf grading method integrating attention mechanism and convolutional neural network model | |
| CN113221913A (en) | Agriculture and forestry disease and pest fine-grained identification method and device based on Gaussian probability decision-level fusion | |
| CN112529838A (en) | Tobacco leaf maturity online judging method based on image processing technology | |
| CN114241309A (en) | Rice sheath blight identification method and system based on ShuffleNet V2-Unet | |
| CN114139586A (en) | Electric energy quality composite disturbance identification method based on multi-scale analysis and two-dimensional feature extraction | |
| CN113313682A (en) | No-reference video quality evaluation method based on space-time multi-scale analysis | |
| CN113128560A (en) | Attention module enhancement-based CNN regular script style classification method | |
| CN113112482A (en) | PCB defect detection method based on attention mechanism network | |
| CN111652238A (en) | Multi-model integration method and system | |
| CN116758415A (en) | Lightweight pest identification method based on two-dimensional discrete wavelet transformation | |
| CN109101905A (en) | SAR image terrain classification recognition methods based on depth convolutional neural networks | |
| CN115494009A (en) | Intelligent discrimination method and system for tobacco leaf maturity | |
| CN113128525A (en) | Control device and method for desert grassland population patch identification | |
| CN112926450A (en) | Leaf disease feature identification method based on deep belief network | |
| CN112990333A (en) | Deep learning-based weather multi-classification identification method | |
| CN111126364A (en) | Expression recognition method based on packet convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |