CN110443172A - A kind of object detection method and system based on super-resolution and model compression - Google Patents
A kind of object detection method and system based on super-resolution and model compression Download PDFInfo
- Publication number
- CN110443172A CN110443172A CN201910678925.4A CN201910678925A CN110443172A CN 110443172 A CN110443172 A CN 110443172A CN 201910678925 A CN201910678925 A CN 201910678925A CN 110443172 A CN110443172 A CN 110443172A
- Authority
- CN
- China
- Prior art keywords
- resolution
- model
- super
- faster
- training
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 84
- 230000006835 compression Effects 0.000 title claims abstract description 47
- 238000007906 compression Methods 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 claims abstract description 56
- 238000013138 pruning Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 description 15
- 230000008569 process Effects 0.000 description 10
- 238000004590 computer program Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 241000196324 Embryophyta Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of object detection method and system based on super-resolution and model compression, this method comprises: the label using the image of high resolution as training, the sample using the low image of corresponding resolution ratio as training, training super-resolution model；Low-resolution image to be processed is handled using trained super-resolution model, generates corresponding high-definition picture；Faster-RCNN network is improved according to presetting method, to the high-definition picture that super-resolution model generates, uses improved Faster-RCNN network training target detection model；Trained target detection model is compressed using preset model compression method, enables it to be deployed on intelligent terminal.The solution of the present invention can realize real-time target detection to the satellite image for the low resolution that satellite is shot, and have good detection accuracy.
Description
Technical field
The present invention relates to target detection technique fields, particularly relate to a kind of based on super-resolution and the inspection of the target of model compression
Survey method and system.
Background technique
Target detection technique is the very fast computer vision field that technology develops iteration in recent years, in this neck
What is had an epoch-marking significance in domain is RCNN list of target detection network.In RCNN series, performance is best, and fastest works as
Belong to Faster-RCNN network, but for the target detection on the satellite image of a large amount of low resolution of satellite shooting, it is existing
The detection detection accuracy of object detection method is unsatisfactory；Although in addition, there is the higher network mould of many detection accuracy at this stage
Type, such as Faster-RCNN, Fast-RCNN etc..But these networks are opposite in the case where not doing model compression
The network speeds such as YOLOv3 are slower, therefore cannot really realize that real-time target detects.
Summary of the invention
Present invention aim to address the target detections on the satellite image of a large amount of low resolution for satellite shooting
Problem, existing object detection method cannot achieve accurate detection in real time；And the higher network model of detection accuracy at this stage exists
The network speeds such as YOLOv3 relatively are slower in the case where not doing model compression, therefore cannot really realize that real-time target detects
The problem of.
In order to solve the above technical problems, the present invention provides a kind of target detection side based on super-resolution and model compression
Method, which comprises
Label using the image of high resolution as training, the sample using the low image of corresponding resolution ratio as training
This, training super-resolution model；
Low-resolution image to be processed is handled using trained super-resolution model, generates corresponding high score
Resolution image；
Faster-RCNN network is improved according to presetting method, the high resolution graphics that super-resolution model is generated
Picture uses improved Faster-RCNN network training target detection model；
Trained target detection model is compressed using preset model compression method, enables it to be deployed to
On intelligent terminal.
It is further, described that Faster-RCNN network is improved according to presetting method, comprising:
The basic network of Faster-RCNN network is changed to ResNeXt101 by VGG16.
It is further, described that Faster-RCNN network is improved according to presetting method, further includes:
Convolutional layer in Faster-RCNN network is replaced with into worm-eaten convolutional layer.
It is further, described that Faster-RCNN network is improved according to presetting method, further includes:
FPN network is used on the basic network.
Further, described that trained target detection model is compressed using preset model compression method, make
Obtaining it can be deployed on intelligent terminal, specifically:
Model beta pruning is carried out using BN layers behind the convolutional layer in improved Faster-RCNN network of gamma parameters,
Trained target detection model is compressed, enables it to be deployed on intelligent terminal.
Correspondingly, in order to solve the above technical problems, the present invention also provides a kind of mesh based on super-resolution and model compression
Detection system is marked, the system comprises:
Super-resolution model training module, the label for the image using high resolution as training, use are corresponding
Sample of the low image of resolution ratio as training, training super-resolution model；
High-definition picture generation module, for utilizing trained super-resolution model to low resolution figure to be processed
As being handled, corresponding high-definition picture is generated；
Target detection model training module, for being improved to Faster-RCNN network according to presetting method, to oversubscription
The high-definition picture that resolution model generates, uses improved Faster-RCNN network training target detection model；
Model compression module, for being pressed using preset model compression method trained target detection model
Contracting, enables it to be deployed on intelligent terminal.
Further, the target detection model training module is specifically used for:
The basic network of Faster-RCNN network is changed to ResNeXt101 by VGG16.
Further, the target detection model training module is also used to:
Convolutional layer in Faster-RCNN network is replaced with into worm-eaten convolutional layer.
Further, the target detection model training module is also used to:
FPN network is used on the basic network.
Further, the model compression module is specifically used for:
Model beta pruning is carried out using BN layers behind the convolutional layer in improved Faster-RCNN network of gamma parameters,
Trained target detection model is compressed, enables it to be deployed on intelligent terminal.
The advantageous effects of the above technical solutions of the present invention are as follows:
Label of the present invention by using the image of high resolution as training, is made using the low image of corresponding resolution ratio
For trained sample, training super-resolution model；Using trained super-resolution model to low-resolution image to be processed
It is handled, generates corresponding high-definition picture；Faster-RCNN network is improved according to presetting method, to oversubscription
The high-definition picture that resolution model generates, uses improved Faster-RCNN network training target detection model；Using
Preset model compression method compresses trained target detection model, can be deployed on intelligent terminal.From
And the effect that real-time target detection can be carried out to the satellite image for the low resolution that satellite is shot is realized, and have good
Detection accuracy.
Detailed description of the invention
Fig. 1 is the flow diagram of the object detection method of the invention based on super-resolution and model compression；
Fig. 2 is RDN network overall architecture schematic diagram；
Fig. 3 is the structural schematic diagram of RDB module；
Fig. 4 is ResNeXt50 and ResNet50 Structure Comparison schematic diagram；
Fig. 5 is the block structure contrast schematic diagram of ResNet and ResNeXt；
Fig. 6 is the structural schematic diagram of FPN network；
Fig. 7 is the step schematic diagram of model beta pruning.
Specific embodiment
To keep the technical problem to be solved in the present invention, technical solution and advantage clearer, below in conjunction with attached drawing and tool
Body embodiment is described in detail.
First embodiment
The present embodiment provides a kind of object detection method based on super-resolution and model compression, this method are related to image
Super-resolution technique, target detection technique and model compression technology.Wherein, high-precision target detection technique is basic, and
Image super-resolution and model compression technology are in order to which the performance for making the method for the present embodiment solve object detection task is more excellent.
Super-resolution technique refers to one kind that corresponding high-definition picture is reconstructed from the low-resolution image observed
Technology.The too low target detection and semantic segmentation that will greatly affect in computer vision of the resolution ratio of picture, because of resolution ratio
Low picture can lose many pictorial informations, therefore the present embodiment uses super-resolution technique first by the figure of low resolution
Piece is converted into high-resolution picture.
Model compression technology be in recent years for deployment depth learning art on intelligent devices and widely used skill
Art, is widely used in smart phone, and the hardware conditions such as intelligent wearable device are not in especially strong terminal.From initial
The methods of MobileNet, SqueezeNet and model beta pruning develop to the automatic beta pruning of model of today, the side such as model retraining
Method enables intelligent terminal to dispose artificial intelligence technology on a large scale.
Specifically, as shown in Figure 1, the object detection method based on super-resolution and model compression of the present embodiment, the base
Include: in the object detection method of super-resolution and model compression
S101, the label using the image of high resolution as training, using the low image of corresponding resolution ratio as instruction
Experienced sample, training super-resolution model；
S102 is handled low-resolution image to be processed using trained super-resolution model, is generated and is corresponded to
High-definition picture；
It should be noted that the effect of above-mentioned S101 and S102 is to use oversubscription to original ambiguous satellite image
Resolution technology makes its sharpening；In S101 training super-resolution model use RDN network, the network architecture as shown in Fig. 2,
The parameter of RDN network is less, is adapted to as the real-time model for generating high-definition picture；The network is in order to as much as possible
The convergence of training is realized using feature and as soon as possible, uses the structure of RDB (Residual dense block).The RDB's
Structure is as shown in Figure 3.
S103 improves Faster-RCNN network according to presetting method, the high-resolution generated to super-resolution model
Rate image uses improved Faster-RCNN network training target detection model；
It should be noted that above-mentioned steps are to use improved Faster-RCNN network to carry out target detection.
Faster-RCNN is to be responsible for original image zooming-out feature, tightly using VGG16 network as basic network at the beginning of proposition
Then the characteristic pattern extracted is input in RPN network, the window of one 3*3 of RPN Web vector graphic slides simultaneously on characteristic pattern
And 9 kinds of anchor are applied in original image, final training remains input of a certain number of ROIs as next step.
It is exactly that basic network has been changed to ResNeXt101 by VGG16 that first of the present embodiment, which improves,. ResNeXt101
Network possesses 101 convolutional layers, and characteristic pattern is divided into 32 parts along channel direction on each convolutional layer, this can be to original
Beginning image carries out better feature extraction；In addition, ResNeXt101 is also multiplexed the feature of low layer, to a certain degree
On solve the problems, such as gradient disappearance；Additional ginseng is not brought for ResNet101 finally, ResNeXt101 compares
Number does not influence trained and derivation process speed.Fig. 4 is the Structure Comparison schematic diagram of ResNeXt50 and ResNet50.
Figure 4, it is seen that although the port number in the output characteristic pattern of each convolutional layer increases twice,
The parameter of ResNeXt network does not increase, and there are also certain reductions instead.Fig. 5 is then the block structure of ResNeXt and ResNet
Contrast schematic diagram.
Second improvement that the present embodiment does Faster-RCNN network is that convolutional layer is replaced with to worm-eaten convolution
(Atrous Convolution) layer, can increase the receptive field of characteristic pattern using this convolution mode.Use the idea of worm-eaten convolution
From InceptionV3 network, the mode of this semantic segmentation Web vector graphic worm-eaten convolution increases special in convolution process
Levy the receptive field of figure.And the receptive field for increasing characteristic pattern is equally important for object detection task, therefore in the present embodiment,
Common convolutional layer has all been replaced by worm-eaten convolutional layer.
It is exactly that FPN has been used on basic network that the present embodiment, which improves the third that Faster-RCNN network is done,
(Feature Pyramid Network) network, network structure as shown in fig. 6, it can be used low layer generation size compared with
Big characteristic pattern, and original network has only used the feature of the last layer.After using FPN, original Faster-RCNN
Network has been thus lifted to 15 for 9 region proposal of an anchor.
S104 compresses trained target detection model using preset model compression method, enable it to by
It is deployed on intelligent terminal.
It should be noted that the model parameter that the target detections network training such as Faster-RCNN comes out is more, speed is slower,
To the more demanding of hardware；And the target detection network model parameter such as YOLOv3 is less, fast speed, is able to achieve in real time
Object detection task, but the precision of model is relatively low.Therefore, carry out model compression be one kind can balance detection precision and
Detect the effective ways of speed.
The theoretical basis of this method is that mankind's neuron not all when pondering a problem all is activated, because
This present invention in deep learning model weeds out some unnecessary connections to reduce the number of parameters of model, accelerates depth
The speed of learning model.
In model compression field, most-often used is that method is model beta pruning, i.e., according to the performance after model beta pruning
Performance connects which to determine to cut.The basic network used in the present embodiment is ResNet101, behind each convolutional layer
It will one BN (Batch Normalization) layer of connection.Therefore BN layers of gamma parameter can be used to cut to carry out model
Branch.In order to use BN layers of gamma parameter, need for gamma parameter plus L1 canonical punishment training model, new loss letter
Number becomes:
L=∑(x,y)l(f(x,W),y)+μ∑γ∈τg(γ)。
All gamma in the network are ranked up, according to beta pruning ratio given by man, remove gamma very little
Channel, be finally finely adjusted, this process is repeated as many times, and is obtained preferably as a result, specific process is as shown in Figure 7.
Label of the present embodiment by using the image of high resolution as training, using the low image of corresponding resolution ratio
As trained sample, training super-resolution model；Using trained super-resolution model to low resolution figure to be processed
As being handled, corresponding high-definition picture is generated；Faster-RCNN network is improved according to presetting method, to super
The high-definition picture that resolution model generates, uses improved Faster-RCNN network training target detection model；Using
Preset model compression method compresses trained target detection model, can be deployed on intelligent terminal.It is real
The effect of real-time target detection can be carried out to the satellite image for the low resolution that satellite is shot by having showed, and have good detection
Precision.
Second embodiment
The present embodiment provides a kind of object detection system based on super-resolution and model compression, should based on super-resolution and
The object detection system of model compression includes:
Super-resolution model training module, the label for the image using high resolution as training, use are corresponding
Sample of the low image of resolution ratio as training, training super-resolution model；
High-definition picture generation module, for utilizing trained super-resolution model to low resolution figure to be processed
As being handled, corresponding high-definition picture is generated；
Target detection model training module, for being improved to Faster-RCNN network according to presetting method, to oversubscription
The high-definition picture that resolution model generates, uses improved Faster-RCNN network training target detection model；
Model compression module, for being pressed using preset model compression method trained target detection model
Contracting, enables it to be deployed on intelligent terminal.
Base in the object detection system based on super-resolution and model compression of the present embodiment and above-mentioned first embodiment
It is corresponded to each other in the object detection method of super-resolution and model compression, wherein the function that each modular unit is realized in the system
It can be corresponded with each process step in the above method；Therefore details are not described herein.
In addition, it should be noted that, it should be understood by those skilled in the art that, the embodiment of the embodiment of the present invention can provide
For method, apparatus or computer program product.Therefore, it is real that complete hardware embodiment, complete software can be used in the embodiment of the present invention
Apply the form of example or embodiment combining software and hardware aspects.Moreover, the embodiment of the present invention can be used it is one or more its
In include computer usable program code computer-usable storage medium (including but not limited to magnetic disk storage, CD-ROM,
Optical memory etc.) on the form of computer program product implemented.
The embodiment of the present invention be referring to according to the method for the embodiment of the present invention, terminal device (system) and computer program
The flowchart and/or the block diagram of product describes.It should be understood that flowchart and/or the block diagram can be realized by computer program instructions
In each flow and/or block and flowchart and/or the block diagram in process and/or box combination.It can provide these
Computer program instructions to general purpose computer, Embedded Processor or other programmable data processing terminal devices processor with
A machine is generated, so that generating by the instruction that computer or the processor of other programmable data processing terminal devices execute
For realizing the function of being specified in one or more flows of the flowchart and/or one or more blocks of the block diagram
Device.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing terminal devices
In computer-readable memory operate in a specific manner, so that instruction stored in the computer readable memory generates packet
The manufacture of command device is included, which realizes in one side of one or more flows of the flowchart and/or block diagram
The function of being specified in frame or multiple boxes.These computer program instructions can also be loaded at computer or other programmable datas
It manages on terminal device, so that executing series of operation steps on computer or other programmable terminal equipments to generate computer
The processing of realization, so that the instruction executed on computer or other programmable terminal equipments is provided for realizing in flow chart one
The step of function of being specified in a process or multiple processes and/or one or more blocks of the block diagram.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications can be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of range of embodiment of the invention.
It should also be noted that, herein, the terms "include", "comprise" or its any other variant are intended to non-
It is exclusive to include, so that process, method, article or terminal device including a series of elements are not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or terminal
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in process, method, article or the terminal device for including the element.
The above is a preferred embodiment of the present invention, it is noted that for those skilled in the art
For, without departing from the principles of the present invention, several improvements and modifications can also be made, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (10)
1. a kind of object detection method based on super-resolution and model compression characterized by comprising
Label using the image of high resolution as training, the sample using the low image of corresponding resolution ratio as training,
Training super-resolution model；
Low-resolution image to be processed is handled using trained super-resolution model, generates corresponding high-resolution
Image；
Faster-RCNN network is improved according to presetting method, to the high-definition picture that super-resolution model generates, is made
With improved Faster-RCNN network training target detection model；
Trained target detection model is compressed using preset model compression method, enables it to be deployed to intelligence
In terminal.
2. the object detection method based on super-resolution and model compression as described in claim 1, which is characterized in that described right
Faster-RCNN network is improved according to presetting method, comprising:
The basic network of Faster-RCNN network is changed to ResNeXt101 by VGG16.
3. the object detection method based on super-resolution and model compression as claimed in claim 2, which is characterized in that described right
Faster-RCNN network is improved according to presetting method, further includes:
Convolutional layer in Faster-RCNN network is replaced with into worm-eaten convolutional layer.
4. the object detection method based on super-resolution and model compression as claimed in claim 3, which is characterized in that described right
Faster-RCNN network is improved according to presetting method, further includes:
FPN network is used on the basic network.
5. the object detection method based on super-resolution and model compression as claimed in claim 4, which is characterized in that described to adopt
Trained target detection model is compressed with preset model compression method, enables it to be deployed to intelligent terminal
On, specifically:
Model beta pruning is carried out using BN layers behind the convolutional layer in improved Faster-RCNN network of gamma parameters, to instruction
The target detection model perfected is compressed, and enables it to be deployed on intelligent terminal.
6. a kind of object detection system based on super-resolution and model compression characterized by comprising
Super-resolution model training module, the label for the image using high resolution as training, using corresponding resolution
Sample of the low image of rate as training, training super-resolution model；
High-definition picture generation module, for using trained super-resolution model to low-resolution image to be processed into
Row processing, generates corresponding high-definition picture；
Target detection model training module, for being improved to Faster-RCNN network according to presetting method, to super-resolution
The high-definition picture that model generates, uses improved Faster-RCNN network training target detection model；
Model compression module is made for being compressed using preset model compression method to trained target detection model
Obtaining it can be deployed on intelligent terminal.
7. the object detection system based on super-resolution and model compression as claimed in claim 6, which is characterized in that the mesh
Mark detection model training module is specifically used for:
The basic network of Faster-RCNN network is changed to ResNeXt101 by VGG16.
8. the object detection system based on super-resolution and model compression as claimed in claim 7, which is characterized in that the mesh
Mark detection model training module is also used to:
Convolutional layer in Faster-RCNN network is replaced with into worm-eaten convolutional layer.
9. the object detection system based on super-resolution and model compression as claimed in claim 8, which is characterized in that the mesh
Mark detection model training module is also used to:
FPN network is used on the basic network.
10. the object detection system based on super-resolution and model compression as claimed in claim 9, which is characterized in that described
Model compression module is specifically used for:
Model beta pruning is carried out using BN layers behind the convolutional layer in improved Faster-RCNN network of gamma parameters, to instruction
The target detection model perfected is compressed, and enables it to be deployed on intelligent terminal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910678925.4A CN110443172A (en) | 2019-07-25 | 2019-07-25 | A kind of object detection method and system based on super-resolution and model compression |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910678925.4A CN110443172A (en) | 2019-07-25 | 2019-07-25 | A kind of object detection method and system based on super-resolution and model compression |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110443172A true CN110443172A (en) | 2019-11-12 |
Family
ID=68431653
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910678925.4A Pending CN110443172A (en) | 2019-07-25 | 2019-07-25 | A kind of object detection method and system based on super-resolution and model compression |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110443172A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111160368A (en) * | 2019-12-24 | 2020-05-15 | 中国建设银行股份有限公司 | Method, device and equipment for detecting target in image and storage medium |
| CN111368696A (en) * | 2020-02-28 | 2020-07-03 | 淮阴工学院 | Dangerous chemical transport vehicle illegal driving behavior detection method and system based on visual cooperation |
| CN111582377A (en) * | 2020-05-09 | 2020-08-25 | 济南浪潮高新科技投资发展有限公司 | Edge end target detection method and system based on model compression |
| CN111652211A (en) * | 2020-05-21 | 2020-09-11 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting foreign matter hanging fault of motor car anti-snaking shock absorber mounting seat |
| CN111833335A (en) * | 2020-07-17 | 2020-10-27 | 燕山大学 | Lattice structure defect detection method and system based on super-resolution reconstruction |
| CN112929665A (en) * | 2021-01-28 | 2021-06-08 | 北京博雅慧视智能技术研究院有限公司 | Target tracking method, device, equipment and medium combining super-resolution and video coding |
| CN113191945A (en) * | 2020-12-03 | 2021-07-30 | 陕西师范大学 | High-energy-efficiency image super-resolution system and method for heterogeneous platform |
| CN113221925A (en) * | 2021-06-18 | 2021-08-06 | 北京理工大学 | Target detection method and device based on multi-scale image |
| CN113435384A (en) * | 2021-07-07 | 2021-09-24 | 中国人民解放军国防科技大学 | Target detection method, device and equipment for medium-low resolution optical remote sensing image |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105389797A (en) * | 2015-10-16 | 2016-03-09 | 西安电子科技大学 | Unmanned aerial vehicle video small-object detecting method based on super-resolution reconstruction |
| CN105741252A (en) * | 2015-11-17 | 2016-07-06 | 西安电子科技大学 | Sparse representation and dictionary learning-based video image layered reconstruction method |
| CN108734143A (en) * | 2018-05-28 | 2018-11-02 | 江苏迪伦智能科技有限公司 | A kind of transmission line of electricity online test method based on binocular vision of crusing robot |
| CN109191369A (en) * | 2018-08-06 | 2019-01-11 | 三星电子（中国）研发中心 | 2D pictures turn method, storage medium and the device of 3D model |
| CN109472801A (en) * | 2018-11-22 | 2019-03-15 | 廖祥 | It is a kind of for multiple dimensioned neuromorphic detection and dividing method |
| CN109784476A (en) * | 2019-01-12 | 2019-05-21 | 福州大学 | A method of improving DSOD network |
-
2019
- 2019-07-25 CN CN201910678925.4A patent/CN110443172A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105389797A (en) * | 2015-10-16 | 2016-03-09 | 西安电子科技大学 | Unmanned aerial vehicle video small-object detecting method based on super-resolution reconstruction |
| CN105741252A (en) * | 2015-11-17 | 2016-07-06 | 西安电子科技大学 | Sparse representation and dictionary learning-based video image layered reconstruction method |
| CN108734143A (en) * | 2018-05-28 | 2018-11-02 | 江苏迪伦智能科技有限公司 | A kind of transmission line of electricity online test method based on binocular vision of crusing robot |
| CN109191369A (en) * | 2018-08-06 | 2019-01-11 | 三星电子（中国）研发中心 | 2D pictures turn method, storage medium and the device of 3D model |
| CN109472801A (en) * | 2018-11-22 | 2019-03-15 | 廖祥 | It is a kind of for multiple dimensioned neuromorphic detection and dividing method |
| CN109784476A (en) * | 2019-01-12 | 2019-05-21 | 福州大学 | A method of improving DSOD network |
Non-Patent Citations (5)
| Title |
|---|
| AI研习社: ""一文教你如何pytorch 构建 faster r-cnn"", 《HTTP：//ZHUANLAN.ZHIHU.COM/P/56710152》 * |
| GEORGE PAPANDREOU 等: ""Towards Accurate Multi-person Pose Estimation in the Wild"", 《ARXIV.ORG》 * |
| JACOB SHERMEYER 等: ""The Effects of Super-Resolution on Object Detection Performance in Satellite Imagery"", 《2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW)》 * |
| YIWU YAO 等: ""CREATING LIGHTWEIGHT OBJECT DETECTORS WITH MODEL COMPRESSION FOR DEPLOYMENT ON EDGE DEVICES"", 《ARXIV.ORG》 * |
| YULUN ZHANG 等: ""Residual Dense Network for Image Super-Resolution"", 《PROCEEDINGS OF THE IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111160368A (en) * | 2019-12-24 | 2020-05-15 | 中国建设银行股份有限公司 | Method, device and equipment for detecting target in image and storage medium |
| CN111368696A (en) * | 2020-02-28 | 2020-07-03 | 淮阴工学院 | Dangerous chemical transport vehicle illegal driving behavior detection method and system based on visual cooperation |
| CN111582377A (en) * | 2020-05-09 | 2020-08-25 | 济南浪潮高新科技投资发展有限公司 | Edge end target detection method and system based on model compression |
| CN111652211A (en) * | 2020-05-21 | 2020-09-11 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting foreign matter hanging fault of motor car anti-snaking shock absorber mounting seat |
| CN111833335A (en) * | 2020-07-17 | 2020-10-27 | 燕山大学 | Lattice structure defect detection method and system based on super-resolution reconstruction |
| CN113191945A (en) * | 2020-12-03 | 2021-07-30 | 陕西师范大学 | High-energy-efficiency image super-resolution system and method for heterogeneous platform |
| CN113191945B (en) * | 2020-12-03 | 2023-10-27 | 陕西师范大学 | Heterogeneous platform-oriented high-energy-efficiency image super-resolution system and method thereof |
| CN112929665A (en) * | 2021-01-28 | 2021-06-08 | 北京博雅慧视智能技术研究院有限公司 | Target tracking method, device, equipment and medium combining super-resolution and video coding |
| CN113221925A (en) * | 2021-06-18 | 2021-08-06 | 北京理工大学 | Target detection method and device based on multi-scale image |
| CN113221925B (en) * | 2021-06-18 | 2022-11-11 | 北京理工大学 | Target detection method and device based on multi-scale image |
| CN113435384A (en) * | 2021-07-07 | 2021-09-24 | 中国人民解放军国防科技大学 | Target detection method, device and equipment for medium-low resolution optical remote sensing image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443172A (en) | A kind of object detection method and system based on super-resolution and model compression | |
| CN111950723B (en) | Neural network model training method, image processing method, device and terminal equipment | |
| US11392822B2 (en) | Image processing method, image processing apparatus, and computer-readable storage medium | |
| CN109191382B (en) | Image processing method, device, electronic equipment and computer readable storage medium | |
| JP2020508531A (en) | Image quality evaluation method and image quality evaluation system | |
| CN108241854B (en) | Depth video saliency detection method based on motion and memory information | |
| CN109377445A (en) | Model training method, the method, apparatus and electronic system for replacing image background | |
| CN111383232A (en) | Matting method, matting device, terminal equipment and computer-readable storage medium | |
| CN116363261B (en) | Training method of image editing model, image editing method and device | |
| CN108875826B (en) | Multi-branch object detection method based on coarse and fine granularity composite convolution | |
| CN110853071A (en) | Image editing method and terminal equipment | |
| CN109447254B (en) | Convolution neural network reasoning hardware acceleration method and device thereof | |
| US11620514B2 (en) | Training artificial neural networks with reduced computational complexity | |
| CN110443764A (en) | Video repairing method, device and server | |
| CN110087097A (en) | It is a kind of that invalid video clipping method is automatically removed based on fujinon electronic video endoscope | |
| CN110223245A (en) | Blurred picture clearness processing method and system based on deep neural network | |
| CN107729885B (en) | Face enhancement method based on multiple residual error learning | |
| CN116205820A (en) | Image enhancement method, target identification method, device and medium | |
| CN108764233B (en) | Scene character recognition method based on continuous convolution activation | |
| KR20010113666A (en) | Color image processing method and apparatus thereof | |
| CN117611821A (en) | Instance segmentation method, device, system and storage medium | |
| CN116342800B (en) | Semantic three-dimensional reconstruction method and system for multi-mode pose optimization | |
| CN110717913B (en) | Image segmentation method and device | |
| CN110163194B (en) | Image processing method, device and storage medium | |
| CN116993978A (en) | Small sample segmentation method, system, readable storage medium and computer device |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191112 |