CN110210493A - Profile testing method and system based on non-classical receptive field modulation neural network - Google Patents
Profile testing method and system based on non-classical receptive field modulation neural network Download PDFInfo
- Publication number
- CN110210493A CN110210493A CN201910361332.5A CN201910361332A CN110210493A CN 110210493 A CN110210493 A CN 110210493A CN 201910361332 A CN201910361332 A CN 201910361332A CN 110210493 A CN110210493 A CN 110210493A
- Authority
- CN
- China
- Prior art keywords
- receptive field
- neural network
- image
- classical
- classical receptive
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses profile testing methods and system based on non-classical receptive field modulation neural network, comprising: constructs one group of multi-resolution image pyramid using multi-scale sampling；The feature representation that sparse self-encoding encoder unsupervised learning simple cell receptive field is utilized to the image collection of each scale obtains the different low-level features mapping of image；The Feature Mapping that respectively different receptive field cells generates constructs a matching modulated environments；Response caused by receptive field is modulated according to the information in non-classical receptive field, and merges the modulation result of different characteristic into unified feature representation；The result of joint different scale inputs classifier calculated profile probability.The present invention generates the modulating action for inhibiting or easily changing to the response in receptive field using the information in non-classical receptive field, meaningless trifling texture edge is reduced in large quantities, and the boundary of outburst area, the conspicuousness structure of image is saved, the contour detecting performance of image is improved.
Description
Technical field
The present invention relates to image procossings, mode identification technology, and in particular to based on non-classical receptive field modulation nerve
The profile testing method and system of network.
Background technique
Profile contains structural information important in image (such as target shape, zone boundary), is widely used in being permitted
Multicomputer visual task, such as object identification, candidate target suggestion, image segmentation.However accurately detect the wheel of natural image
Exterior feature is a difficult task, and Major Difficulties are: the semantic structure of image is often buried in the office as caused by grain background
In portion edge, need to exclude the largely texture edge without practical significance；Some targets may have similar intensity with background, make
Obtain boundary responds vaild evidence that is faint or even lacking part, needs that large-scale information is combined to be judged.
Neuro-physiology research shows that: the periphery of receptive field there is one to cellular response rise modulating action outskirt
Domain, area ratio tradition receptive field is much bigger, and the small luminous point stimulation in this region can not directly cause the anti-of gangliocyte
It answers, but easily can change or inhibit its reaction, adjust original effect as caused by receptive field, this region is known as non-classical impression
It is wild.Non-classical receptive field expands the useful space that visual cortex neuron receives characteristic information input, to allow to appoint in complexity
Large-scale information in the visual field is combined in the early stage processing of business.Thus the contour detecting side of many view-based access control model mechanism has been inspired
Method, it is intended to the shortcomings that by introducing large-scale area information, overcoming Local Edge Detection device.These vision bionics models are one
Determining degree improves the visual effect of contour detecting, but they generally use artificially defined difference of Gaussian (DoG) function or
The DoG function of butterfly describes non-classical receptive field modulating action, the Firing Patterns of a true neuron with neuron and
It is different, all for describing to be occurred cannot be gone with too simple mathematical function.
Summary of the invention
The technical problem to be solved in the present invention is that process for the profile of above-mentioned current detection natural image it is complicated,
The more technological deficiency in numerous and disorderly texture edge, provide based on non-classical receptive field modulation neural network profile testing method and
System solves the above problems.
Profile testing method based on non-classical receptive field modulation neural network, comprising:
Step 1 carries out multi-scale sampling generation multi-resolution image to original image set, to obtain the instruction of different scale
Practice image collection；
Step 2 establishes non-classical receptive field modulation neural network, is arranged the number of plies and number of nodes of network, described in initialization
Non-classical receptive field modulates the parameter of the receptive field layer in neural network, defines the non-classical receptive field modulation neural network
Objective function；
Step 3, the training image set training non-classical receptive field based on the different scale modulate neural network: utilizing
Traditional neural network back-propagation algorithm minimizes the target letter of the non-classical receptive field modulation neural network with having supervision
Number is finally obtained with the parameter of the receptive field layer in the non-classical receptive field modulation neural network after optimize and is completed to optimize
Non-classical receptive field modulate neural network；
Step 4, the non-classical receptive field modulation neural network that images to be recognized is inputted to the optimization, obtain figure to be identified
Each pixel belongs to the probability of profile as in；
Step 5 carries out thresholding processing to the profile probability results of images to be recognized, and probability value is lower than to the picture of preset value
Element is set to 0, and carries out contour thinning by non-maximum suppression method, finally exports the contour detecting result of images to be recognized.
Further, the training neural network specifically includes: by each ruler in the training image set of different scale
The image of degree is inputted, and the first hidden layer of network is receptive field layer, is simulated the characteristic response that classical receptive field generates, is then existed
A non-classical receptive field modulating layer is added after each receptive field Feature Mapping, the original response generated to receptive field is modulated
Effect, and different modulation results is merged, the image of down-sampling is inputted, increases the layer that deconvolutes for restoring
Fusion results optimize combination to original image size, by the result that scale fused layer obtains different scale, finally adopt
Profile probability to the end is obtained with the output layer based on cross entropy loss function.
Further, the training image set training nerve of the different scale is utilized using image-to-image mode
Network, the input layer of network correspond to image, and output layer is that the profile of the image maps, and generate the profile probability of each pixel.
Further, the parameter for initializing the receptive field layer in neural network specifically includes: respectively to the figure of different scale
Image set is sampled to obtain training image blocks, using the mark sheet of the receptive field of sparse self-encoding encoder unsupervised learning neural network
It reaches, different hidden neuron association carries out edge detection in the different location of image and direction, utilizes the feature learnt
Detector initializes the parameter in receptive field layer, the network parameter random initializtion of other layers.
Further, the objective function J of the non-classical receptive field modulation neural network is defined as the true of input data
Cross entropy between label and the label of non-classical receptive field modulation neural network prediction:
Mono- ∑ of J=i[β·1(yi=1) log P (yi=1 | Xi, W) and+(1- β) 1 (yi=0) log P (yi=0 | Xi,
W)],
W indicates parameter set all in the network, the i.e. connection weight of every layer of network, y in formulai∈ { 0,1 } is in image X
The label of pixel i, 1 () are indicator function, P (yi) it is XiBelong to the probability of label,For regulatory factor, balance
Serious classification is unbalanced between profile and non-profile, | Y+| and | Y-| respectively indicate the positive and negative samples collection of image outline.
A kind of outline detection system based on non-classical receptive field modulation neural network, comprising: processor and storage equipment；
The processor loads and executes the instruction in the storage equipment and data for realizing any described in Claims 1 to 5
A kind of profile testing method based on non-classical receptive field modulation neural network.
Compared with prior art, present invention has an advantage that the present invention is embedding by visual cortex non-classical receptive field modulation scheme
Enter neural network, constructs the non-classical receptive field adapted to therewith a modulation for the Feature Mapping that different receptive field cells generates
Environment generates the modulating action for inhibiting or easily changing to the response of receptive field according to periphery contextual information, to efficiently reduce
The important structure feature such as texture skirt response and the profile and the boundary that highlight image.
Detailed description of the invention
Fig. 1 is the flow chart that the profile testing method of neural network is modulated the present invention is based on non-classical receptive field；
Fig. 2 is the block schematic illustration that the profile testing method of neural network is modulated the present invention is based on non-classical receptive field.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to embodiment party of the present invention
Formula is further described.
Profile testing method the present invention is based on non-classical receptive field modulation neural network is by combining depth nerve net
The information processing mechanism of the powerful learning ability of network and human visual system establishes a kind of based on non-classical receptive field tune
Neural network processed provides a kind of computation model of vision bionics for natural image contour detecting.
As shown in Figure 1, the profile testing method based on non-classical receptive field modulation neural network includes:
Step 1 carries out multi-scale sampling generation multi-resolution image to original image set, to obtain the instruction of different scale
Practice image collection；
Step 2 establishes non-classical receptive field modulation neural network, is arranged the number of plies and number of nodes of network, described in initialization
Non-classical receptive field modulates the parameter of the receptive field layer in neural network, defines the non-classical receptive field modulation neural network
Objective function；
Step 3, the training image set training non-classical receptive field based on the different scale modulate neural network: utilizing
Traditional neural network back-propagation algorithm minimizes the target letter of the non-classical receptive field modulation neural network with having supervision
Number is finally obtained with the parameter of the receptive field layer in the non-classical receptive field modulation neural network after optimize and is completed to optimize
Non-classical receptive field modulate neural network；
Step 4, the non-classical receptive field modulation neural network that images to be recognized is inputted to the optimization, obtain figure to be identified
Each pixel belongs to the probability of profile as in；
Step 5 carries out thresholding processing to the profile probability results of images to be recognized, and probability value is lower than to 0.5 pixel
It is set to 0, and contour thinning is carried out by non-maximum suppression method, finally exports the contour detecting result of images to be recognized.
As shown in Fig. 2, the training neural network specifically includes: by each ruler in the training image set of different scale
The image of degree is inputted, and the first hidden layer of network is receptive field layer, and the feature that convolution nuclear mockup classics receptive field generates is rung
It answers.After each receptive field Feature Mapping be added a non-classical receptive field modulating layer, to receptive field generate it is original respond into
Row modulating action establishes a non-classical receptive field modulation areas on the periphery of receptive field, scale is 2-5 times of receptive field
(research " the Extensive integration field being published on Vision Research for 1994 according to Li et al. people
beyond the classical receptive field of cat’s striate cortical neurons-
Classification and tuning properties "), characteristic response of the non-classical receptive field convolution kernel based on receptive field
It calculates and obtains modulated response, will be born in conjunction with the response of receptive field and the modulated response of non-classical receptive field, and by ReLU operation
Value is set to 0, obtains sparse modulation result.Different modulation results is fused into single mapping by next layer network, under
The image of sampling inputs, and it is big to original image for restoring fusion results to be inserted into the layer that deconvolutes that one is made of two-wire interpolation
It is small, combination is optimized by the result that scale fused layer obtains different scale.Finally, one by cross entropy loss function group
At output layer be used for calculating target function cost, export the profile probability of image.
The training image set training neural network of the different scale, network are utilized using image-to-image mode
Input layer correspond to image, output layer be the image profile map, generate the profile probability of each pixel.
The parameter of receptive field layer in initialization neural network specifically includes: adopting respectively to the image set of different scale
Sample obtains training image blocks, different using the feature representation of the receptive field of sparse self-encoding encoder unsupervised learning neural network
Hidden neuron association carries out edge detection in the different location of image and direction, using the property detector learnt to sense
It is initialized by the parameter in wild layer, the network parameter random initializtion of other layers.
The objective function J of the non-classical receptive field modulation neural network is defined as the true tag and non-warp of input data
Allusion quotation receptive field modulates the cross entropy between the label of neural network prediction:
J=- ∑i[β·1(yi=1) log P (yi=1 | Xi, W) and+(1- β) 1 (yi=0) log P (yi=0 | Xi, W)],
W indicates parameter set all in the network, y in formulai∈ { 0,1 } is the label of pixel i in image X, and 1 () referred to
Show function, P (yi) it is XiBelong to the probability of label,It is serious between balance outline and non-profile for regulatory factor
Classification it is unbalanced, | Y+| and | Y-| respectively indicate positive and negative samples collection.
A kind of outline detection system based on non-classical receptive field modulation neural network, comprising: processor and storage equipment；
The processor, which loads and executes instruction and data in the storage equipment, is based on non-warp for realizing any one described
The profile testing method of allusion quotation receptive field modulation neural network.
The information processing of present invention combination deep neural network powerful expression learning ability and human visual system
Mechanism proposes a kind of profile testing method and system based on non-classical receptive field modulation neural network.The present invention will be for not
The Feature Mapping that same receptive field cell generates constructs a matching modulated environments, according to the letter in non-classical receptive field
It ceases and the modulating action for inhibiting or easily changing is generated to the response of receptive field, so as to more effectively reduce numerous and disorderly texture edge and protect
Protect the integrality of the important features information such as profile, boundary.
The embodiment of the present invention is described with above attached drawing, but the invention is not limited to above-mentioned specific
Embodiment, the above mentioned embodiment is only schematical, rather than restrictive, those skilled in the art
Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much
Form, all of these belong to the protection of the present invention.
Claims (6)
1. the profile testing method based on non-classical receptive field modulation neural network characterized by comprising
Step 1 carries out multi-scale sampling generation multi-resolution image to original image set, to obtain the training figure of different scale
Image set closes；
Step 2 establishes non-classical receptive field modulation neural network, and the number of plies and number of nodes of network is arranged, initializes the non-warp
Allusion quotation receptive field modulates the parameter of the receptive field layer in neural network, defines the target of the non-classical receptive field modulation neural network
Function；
Step 3, the training image set training non-classical receptive field based on the different scale modulate neural network: utilizing tradition
Neural network back-propagation algorithm minimize the objective function of non-classical receptive field modulation neural network with having supervision, with
The parameter of the receptive field layer in non-classical receptive field modulation neural network after being optimized, finally obtains the non-warp for completing optimization
Allusion quotation receptive field modulates neural network；
Step 4, the non-classical receptive field modulation neural network that images to be recognized is inputted to the optimization, obtain in images to be recognized
Each pixel belongs to the probability of profile；
Step 5 carries out thresholding processing to the profile probability results of images to be recognized, and the pixel by probability value lower than preset value is set
It is 0, and contour thinning is carried out by non-maximum suppression method, finally exports the contour detecting result of images to be recognized.
2. the profile testing method as described in claim 1 based on non-classical receptive field modulation neural network, which is characterized in that
The training neural network specifically includes: the image of each scale in the training image set of different scale inputted,
The first hidden layer of network is receptive field layer, simulates the characteristic response that classical receptive field generates, then reflects in each receptive field feature
A non-classical receptive field modulating layer is added after penetrating, the original response generated to receptive field is modulated effect, and to different
Modulation result is merged, and the image of down-sampling is inputted, and increases the layer that deconvolutes for restoring fusion results to original
Image size optimizes combination by the result that scale fused layer obtains different scale, finally damages using based on cross entropy
The output layer for losing function obtains profile probability to the end.
3. the profile testing method as described in claim 1 based on non-classical receptive field modulation neural network, which is characterized in that
The training image set training neural network of the different scale, the input layer of network are utilized using image-to-image mode
Corresponding to image, output layer is that the profile of the image maps, and generates the profile probability of each pixel.
4. the profile testing method as described in claim 1 based on non-classical receptive field modulation neural network, which is characterized in that
The parameter of receptive field layer in initialization neural network specifically includes: being sampled instructed to the image set of different scale respectively
Practice image block, using the feature representation of the receptive field of sparse self-encoding encoder unsupervised learning neural network, different hiding nerves
Member has been learned to carry out edge detection in the different location of image and direction, using the property detector learnt in receptive field layer
Parameter initialized, the network parameter random initializtion of other layers.
5. the profile testing method as described in claim 1 based on non-classical receptive field modulation neural network, which is characterized in that
The objective function J of the non-classical receptive field modulation neural network is defined as the true tag and non-classical receptive field of input data
Modulate the cross entropy between the label of neural network prediction:
J=- ∑i[β·1(yi=1) logP (yi=1 | Xi,W)+(1-β)·1(yi=0) logP (yi=0 | Xi, W)],
W indicates parameter set all in the network, the i.e. connection weight of every layer of network, y in formulai∈ { 0,1 } is pixel i in image X
Label, 1 () was indicator function, P (yi) it is XiBelong to the probability of label,For regulatory factor, balance outline
Serious classification is unbalanced between non-profile, | Y+| and | Y-| respectively indicate the positive and negative samples collection of image outline.
6. it is a kind of based on non-classical receptive field modulation neural network outline detection system characterized by comprising processor and
Store equipment；The processor loads and executes the instruction in the storage equipment and data for realizing Claims 1 to 5 institute
Profile testing method of any one stated based on non-classical receptive field modulation neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910361332.5A CN110210493B (en) | 2019-04-30 | 2019-04-30 | Contour detection method and system based on non-classical receptive field modulation neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910361332.5A CN110210493B (en) | 2019-04-30 | 2019-04-30 | Contour detection method and system based on non-classical receptive field modulation neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110210493A true CN110210493A (en) | 2019-09-06 |
| CN110210493B CN110210493B (en) | 2021-03-19 |
Family
ID=67786738
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910361332.5A Expired - Fee Related CN110210493B (en) | 2019-04-30 | 2019-04-30 | Contour detection method and system based on non-classical receptive field modulation neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110210493B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110880035A (en) * | 2019-11-14 | 2020-03-13 | 北京推想科技有限公司 | Convolutional neural network training method and device and nodule sign identification method and device |
| CN111027570A (en) * | 2019-11-20 | 2020-04-17 | 电子科技大学 | Image multi-scale feature extraction method based on cellular neural network |
| CN111062957A (en) * | 2019-10-28 | 2020-04-24 | 广西科技大学鹿山学院 | Non-classical receptive field based contour detection method |
| CN111539969A (en) * | 2020-04-23 | 2020-08-14 | 武汉铁路职业技术学院 | Image edge detection method and device, computer equipment and storage medium |
| CN112926667A (en) * | 2021-03-05 | 2021-06-08 | 中南民族大学 | Method and device for detecting saliency target of depth fusion edge and high-level feature |
| CN113496215A (en) * | 2021-07-07 | 2021-10-12 | 浙江大华技术股份有限公司 | Method and device for detecting human face of living body and electronic equipment |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101930592A (en) * | 2009-09-23 | 2010-12-29 | 电子科技大学 | Image denoising method based on visual non-classical receptive field model |
| CN103903251A (en) * | 2012-12-30 | 2014-07-02 | 南京理工大学 | Night vision image salient contour extracting method based on non-classical receptive field composite modulation |
| JP5971301B2 (en) * | 2013-10-16 | 2016-08-17 | キヤノンマーケティングジャパン株式会社 | Information processing apparatus, information processing apparatus control method, and program |
| CN106033610A (en) * | 2016-03-22 | 2016-10-19 | 广西科技大学 | Contour detection method based on non-classical receptive field space summation modulation |
| CN106355596A (en) * | 2016-08-30 | 2017-01-25 | 刘广海 | Edge detection method fusing uniform color information and compound receptive field model |
| CN106485724A (en) * | 2016-09-20 | 2017-03-08 | 华中科技大学 | A kind of profile testing method that modulates based on combination receptive field and towards feature |
| CN107067407A (en) * | 2017-04-11 | 2017-08-18 | 广西科技大学 | Profile testing method based on non-classical receptive field and linear non-linear modulation |
| CN107273925A (en) * | 2017-06-12 | 2017-10-20 | 太原理工大学 | A kind of Lung neoplasm diagnostic method based on local receptor field and semi-supervised depth own coding |
| CN108898606A (en) * | 2018-06-20 | 2018-11-27 | 中南民族大学 | Automatic division method, system, equipment and the storage medium of medical image |
| CN109060838A (en) * | 2018-07-23 | 2018-12-21 | 三固（厦门）科技有限公司 | A kind of product surface scratch detection method based on machine vision |
| CN109489576A (en) * | 2018-10-19 | 2019-03-19 | 杭州电子科技大学 | A kind of profile testing method based on primary vision access computation model |
-
2019
- 2019-04-30 CN CN201910361332.5A patent/CN110210493B/en not_active Expired - Fee Related
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101930592A (en) * | 2009-09-23 | 2010-12-29 | 电子科技大学 | Image denoising method based on visual non-classical receptive field model |
| CN103903251A (en) * | 2012-12-30 | 2014-07-02 | 南京理工大学 | Night vision image salient contour extracting method based on non-classical receptive field composite modulation |
| JP5971301B2 (en) * | 2013-10-16 | 2016-08-17 | キヤノンマーケティングジャパン株式会社 | Information processing apparatus, information processing apparatus control method, and program |
| CN106033610A (en) * | 2016-03-22 | 2016-10-19 | 广西科技大学 | Contour detection method based on non-classical receptive field space summation modulation |
| CN106355596A (en) * | 2016-08-30 | 2017-01-25 | 刘广海 | Edge detection method fusing uniform color information and compound receptive field model |
| CN106485724A (en) * | 2016-09-20 | 2017-03-08 | 华中科技大学 | A kind of profile testing method that modulates based on combination receptive field and towards feature |
| CN107067407A (en) * | 2017-04-11 | 2017-08-18 | 广西科技大学 | Profile testing method based on non-classical receptive field and linear non-linear modulation |
| CN107273925A (en) * | 2017-06-12 | 2017-10-20 | 太原理工大学 | A kind of Lung neoplasm diagnostic method based on local receptor field and semi-supervised depth own coding |
| CN108898606A (en) * | 2018-06-20 | 2018-11-27 | 中南民族大学 | Automatic division method, system, equipment and the storage medium of medical image |
| CN109060838A (en) * | 2018-07-23 | 2018-12-21 | 三固（厦门）科技有限公司 | A kind of product surface scratch detection method based on machine vision |
| CN109489576A (en) * | 2018-10-19 | 2019-03-19 | 杭州电子科技大学 | A kind of profile testing method based on primary vision access computation model |
Non-Patent Citations (5)
| Title |
|---|
| HUI WEI 等，: "An image representation of infrastructure based on non-classical receptive field", 《SPRINGERLINK》 * |
| MENGCHEN ZHU 等，: "Visual Nonclassical Receptive Field Effects Emerge from Sparse Coding in a Dynamical System", 《PLOS COMPUTATIONAL BIOLOGY》 * |
| SAINING XIE 等，: "Holistically-Nested Edge Detection", 《SPRINGERLINK》 * |
| 张梦楠，: "基于视觉感知的图像轮廓检测技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 郑树泉 等，: "《工业智能技术与应用》", 31 January 2019, 上海科学技术出版社 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111062957A (en) * | 2019-10-28 | 2020-04-24 | 广西科技大学鹿山学院 | Non-classical receptive field based contour detection method |
| CN111062957B (en) * | 2019-10-28 | 2024-02-09 | 广西科技大学鹿山学院 | Non-classical receptive field contour detection method |
| CN110880035A (en) * | 2019-11-14 | 2020-03-13 | 北京推想科技有限公司 | Convolutional neural network training method and device and nodule sign identification method and device |
| CN111027570A (en) * | 2019-11-20 | 2020-04-17 | 电子科技大学 | Image multi-scale feature extraction method based on cellular neural network |
| CN111027570B (en) * | 2019-11-20 | 2022-06-14 | 电子科技大学 | Image multi-scale feature extraction method based on cellular neural network |
| CN111539969A (en) * | 2020-04-23 | 2020-08-14 | 武汉铁路职业技术学院 | Image edge detection method and device, computer equipment and storage medium |
| CN111539969B (en) * | 2020-04-23 | 2023-06-09 | 武汉铁路职业技术学院 | Image edge detection method, device, computer equipment and storage medium |
| CN112926667A (en) * | 2021-03-05 | 2021-06-08 | 中南民族大学 | Method and device for detecting saliency target of depth fusion edge and high-level feature |
| CN112926667B (en) * | 2021-03-05 | 2022-08-30 | 中南民族大学 | Method and device for detecting saliency target of depth fusion edge and high-level feature |
| CN113496215A (en) * | 2021-07-07 | 2021-10-12 | 浙江大华技术股份有限公司 | Method and device for detecting human face of living body and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110210493B (en) | 2021-03-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110210493A (en) | Profile testing method and system based on non-classical receptive field modulation neural network | |
| de Jong et al. | Unsupervised change detection in satellite images using convolutional neural networks | |
| Zou et al. | Ship detection in spaceborne optical image with SVD networks | |
| Gong et al. | Superpixel-based difference representation learning for change detection in multispectral remote sensing images | |
| CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
| Binaghi et al. | A cognitive pyramid for contextual classification of remote sensing images | |
| CN108932455B (en) | Remote sensing image scene recognition method and device | |
| CN113627472A (en) | Intelligent garden defoliating pest identification method based on layered deep learning model | |
| CN108427919B (en) | Unsupervised oil tank target detection method based on shape-guided saliency model | |
| CN111598098A (en) | Water gauge water line detection and effectiveness identification method based on full convolution neural network | |
| CN110991257A (en) | Polarization SAR oil spill detection method based on feature fusion and SVM | |
| Wang et al. | A feature-supervised generative adversarial network for environmental monitoring during hazy days | |
| Kim et al. | Deep learning-based monitoring of overshooting cloud tops from geostationary satellite data | |
| Faisal et al. | Faster R‐CNN Algorithm for Detection of Plastic Garbage in the Ocean: A Case for Turtle Preservation | |
| Hamad | An assessment of artificial neural networks, support vector machines and decision trees for land cover classification using sentinel-2A data | |
| Louw et al. | Object-based land surface segmentation scale optimisation: An ill-structured problem | |
| Zuo et al. | A remote sensing image semantic segmentation method by combining deformable convolution with conditional random fields | |
| CN110135435B (en) | Saliency detection method and device based on breadth learning system | |
| Moschos et al. | Classification of eddy sea surface temperature signatures under cloud coverage | |
| CN114549909A (en) | Pseudo label remote sensing image scene classification method based on self-adaptive threshold | |
| Kathiravan et al. | Ship detection from satellite images using deep learning | |
| EP3671634B1 (en) | Curvilinear object segmentation with geometric priors | |
| CN111209813B (en) | Remote sensing image semantic segmentation method based on transfer learning | |
| CN117671480A (en) | Landslide automatic identification method, system and computer equipment based on visual large model | |
| Nuno et al. | Optimisation of fishing predictions by means of artificial neural networks, anfis, functional networks and remote sensing images |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Tang Qiling Inventor after: Fu Lingling Inventor after: Zheng Fei Inventor after: Liu Haihua Inventor after: Gao Zhiyong Inventor before: Tang Qiling Inventor before: Liu Haihua Inventor before: Gao Zhiyong |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210319 |