CN108288273B - Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network - Google Patents

Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network Download PDF

Info

Publication number
CN108288273B
CN108288273B CN201810132521.0A CN201810132521A CN108288273B CN 108288273 B CN108288273 B CN 108288273B CN 201810132521 A CN201810132521 A CN 201810132521A CN 108288273 B CN108288273 B CN 108288273B
Authority
CN
China
Prior art keywords
candidate region
candidate
final candidate
final
region
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.)
Active
Application number
CN201810132521.0A
Other languages
Chinese (zh)
Other versions
CN108288273A (en
Inventor
吴泽彬
徐洋
石林林
詹天明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing zhiliansen Information Technology Co., Ltd
Original Assignee
Nanjing Zhiliansen Information Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing Zhiliansen Information Technology Co ltd filed Critical Nanjing Zhiliansen Information Technology Co ltd
Priority to CN201810132521.0A priority Critical patent/CN108288273B/en
Publication of CN108288273A publication Critical patent/CN108288273A/en
Application granted granted Critical
Publication of CN108288273B publication Critical patent/CN108288273B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Abstract

The invention discloses a railway contact net abnormal target automatic detection method based on a multi-scale coupling convolution network, and a plurality of final candidate areas of an image are obtained; inputting the final candidate region and the sampling results of the two times of downsampling into a multi-scale coupling convolution network for parameter training and feature extraction; inputting the extracted features into an ELM classifier, and classifying the extracted features; the position of the frame candidate is corrected by using a regressor based on the obtained position of the frame candidate, thereby obtaining the position of the abnormal object in the image information, that is, the position of the frame candidate after correction. The method is suitable for automatic detection of the abnormal target of the railway contact network, can obtain more accurate abnormal detection effect than the observation of human eyes, and has high detection precision and high automation degree.

Description

Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network
Technical Field
The invention relates to the technical field of railway safety guarantee, in particular to a railway contact network abnormal target automatic detection method based on a multi-scale coupling convolution network.
Background
Electrified railways generally adopt a mode of overhead cables for power supply, the safety problem of a contact network directly influences the running safety of railway trains, and abnormal phenomena such as bird nests and the like are an important risk source directly threatening the safe and reliable running of railway power lines. At present, an abnormal target needs to be found and cleared in a manual inspection mode on a railway power line, so that manpower is wasted, and potential safety hazards cannot be eliminated in time. In order to overcome the problems, an abnormal target detection research based on a railway contact network is developed, however, the current research aiming at the aspect of automatic detection of the abnormal target is just started, and both the accuracy and the efficiency of the abnormal detection cannot meet the application requirements.
Target detection is not difficult for humans, and target objects therein are easily located and classified by perception of different color modules in the picture. However, for computers, facing RGB pixel matrices, it is difficult to directly obtain an abstract concept of a dog or cat from an image and locate the position of the abstract concept, and in addition, sometimes a plurality of objects and a cluttered background are mixed together, so that object detection is more difficult.
Aiming at the automatic detection research of abnormal targets, the traditional method has three main problems in the detection of abnormal processes, one is that the region selection strategy of a sliding window has no pertinence; secondly, the manually designed characteristics have no good robustness to the variation of diversity; thirdly, the automatic detection precision of the abnormal target needs to be further optimized, and the efficiency is far from meeting the requirement of real-time application.
How to overcome the above problems is currently needed.
Disclosure of Invention
The invention aims to solve the problems of the existing method for automatically detecting the abnormal target of the overhead line system. The automatic detection method for the abnormal target of the railway contact network based on the multi-scale coupling convolution network is suitable for automatic detection of the abnormal target of the railway contact network, can obtain an abnormal detection effect more accurate than that observed by human eyes, and has the advantages of high detection precision, high automation degree and good application prospect.
In order to achieve the purpose, the invention adopts the technical scheme that:
an automatic detection method for an abnormal target of a railway contact network based on a multi-scale coupling convolution network comprises the following steps,
step (A), acquiring image information of a contact network, respectively acquiring a plurality of groups of candidate regions corresponding to the image information by using a mean shift and normalized segmentation method, and merging the acquired candidate regions by using a shared region merging method to acquire a plurality of final candidate regions of the image;
step (B), normalizing each final candidate area, wherein the sizes are S × S, S represents the length and the width of the normalized final candidate area, performing two times of down-sampling on each normalized final candidate area, inputting the final candidate area and the sampling results of the two times of down-sampling into a multi-scale coupling convolution network, and performing parameter training and feature extraction;
inputting the extracted features into an ELM classifier to classify the extracted features;
and (D) correcting the positions of the candidate frames by using a regressor according to the classified features, so as to obtain the positions of the abnormal targets in the image information, namely the positions of the corrected candidate frames.
The automatic detection method for the abnormal target of the railway contact network based on the multi-scale coupling convolution network, step (a), which is to merge the obtained candidate regions by a shared region merging method to obtain a plurality of final candidate regions of the image, comprises the following steps,
(A1) analyzing the candidate regions generated by the mean shift method one by one, and comparing each candidate region with the corresponding candidate region generated by the normalized segmentation method one by one;
(A2) if the similarity between the candidate region generated by the normalized segmentation method and the candidate region generated by the mean shift method under analysis can be found to be 80% or more, the segmentation of the candidate region is correct, and a final candidate region is obtained by taking a union set of the two coincident candidate regions; and if the similarity between the candidate region generated by the normalized segmentation method and the candidate region generated by the mean shift method under analysis does not reach 80%, the candidate region is considered to be incorrectly divided, and the next candidate region is processed.
The automatic detection method for the abnormal target of the railway contact network based on the multi-scale coupling convolution network comprises the following steps of (A) carrying out two times of downsampling on each normalized final candidate region, inputting the final candidate region and the sampling results of the two times of downsampling into the multi-scale coupling convolution network, and carrying out parameter training and feature extraction,
(B1) performing two times of downsampling on each normalized final candidate region, wherein the final candidate region corresponds to an S/2 final candidate region and an S/4 final candidate region;
(B2) inputting the final candidate region, the S/2 final candidate region and the S/4 final candidate region into a multi-scale coupling convolution network;
(B3) the final candidate region is processed with a convolution kernel of 5 × 5 in the convolution process, the S/2 × S/2 final candidate region is processed with a convolution kernel of 3 × 3 in the convolution process, and the S/4 × S/4 final candidate region is processed with a convolution kernel of 1 × 1 in the convolution process;
(B4) directly combining the extracted features under different scales of the same final candidate region to obtain 7168-dimensional features;
(B5) and sending the 7168-dimensional features into a full-link layer of the multi-scale coupling convolution network for classification, wherein the training of parameters of the multi-scale coupling convolution network is carried out by distinguishing classification results of the full-link layer from actual labels, and the learning rate is 0.01.
In the foregoing automatic detection method for an abnormal target of a railway catenary based on a multi-scale coupled convolutional network, (B4), features extracted under different scales in the same final candidate region are directly combined to obtain 7168-dimensional features, including 4096-dimensional features extracted from the final candidate region, 2048-dimensional features extracted from the S/2 × S/2 final candidate region, and 1024-dimensional features extracted from the S/4 × S/4 final candidate region.
The automatic detection method for the abnormal target of the railway contact network based on the multi-scale coupling convolution network comprises the step (C) of inputting the extracted features into an Extreme Learning Machine (ELM) classifier, classifying the extracted features and obtaining the position of the candidate frame corresponding to the final candidate area, wherein the process comprises the following steps,
there are N different feature samples, X ═ X1，x2，L，xN|xi∈RD，i＝1，2，L，N}，Y＝{y1，y2，L，yN|yi∈RLAnd i is 1,2, L, N }, wherein D is a spectral dimension, L is a class number, X represents a training feature sample set, Y represents a label set corresponding to each feature sample, and X represents a label set corresponding to each feature sampleNRepresents the Nth feature sample, yNLabel, x, representing the Nth characteristic sampleiRepresents the i-th feature sample, yiThe label of the ith characteristic sample is represented, and R represents the value range of the sample value;
let P be the number of hidden nodes, the expression of the extreme learning machine is as follows,
Figure BDA0001575324730000041
where g (x) is the activation function, ωjAnd betajRespectively an input weight and an output weight, bjIs a hidden layer bias.
According to the position of the candidate frame obtained in the step (C), the position of the candidate frame is corrected by using a regressor, so that five parameters of the candidate frame are obtained, wherein the five parameters comprise a category label C and a candidate frame position (x, y, w, h), x is the horizontal coordinate of the upper left vertex of the candidate frame, y is the vertical coordinate of the upper left vertex of the candidate frame, w is the width of the candidate frame, and h is the height of the candidate frame.
The invention has the beneficial effects that: the automatic detection method for the abnormal target of the railway contact network based on the multi-scale coupling convolution network is suitable for the automatic detection of the abnormal target of the railway contact network, can obtain more accurate abnormal detection effect than the observation of human eyes, has high detection precision and high automation degree, and has the following advantages,
(1) extracting candidate regions of the image by adopting a mean shift and normalization segmentation method, and merging the candidate regions obtained by the two methods by using a shared region merging method to obtain a final candidate region, so that the accuracy of extracting the candidate regions is improved;
(2) the final candidate region category is analyzed, and a multi-scale coupling convolution network is used for extracting the features of the candidate region, so that the spatial distribution features of the picture can be reflected to the maximum extent, and the anomaly detection effect of a contact network on a railway is more stable and reliable;
(3) the characteristics are distinguished through the extreme learning machine, the calculation efficiency is high, the generalization capability is strong, and the detection precision is further improved.
Drawings
Fig. 1 is a flow chart of the automatic detection method for the abnormal target of the railway contact network based on the multi-scale coupling convolution network.
Detailed Description
The invention will be further described with reference to the accompanying drawings.
The invention relates to a railway contact net abnormal target automatic detection method based on a multi-scale coupling convolution net, which comprises the following steps,
the method comprises the following steps of (A) obtaining image information of the overhead line system, respectively obtaining a plurality of groups of candidate regions corresponding to the image information by using a mean shift and normalized segmentation method, and combining the obtained candidate regions by using a shared region combining method to obtain a plurality of final candidate regions of the image,
(A1) analyzing the candidate regions generated by the mean shift method one by one, and comparing each candidate region with the corresponding candidate region generated by the normalized segmentation method one by one;
(A2) if the similarity between the candidate region generated by the normalized segmentation method and the candidate region generated by the mean shift method under analysis can be found to be 80% or more, the segmentation of the candidate region is correct, and a final candidate region is obtained by taking a union set of the two coincident candidate regions; if the similarity between the candidate region generated by the normalized segmentation method and the candidate region generated by the mean shift method under analysis does not reach 80%, the candidate region is considered to be divided incorrectly, and the next candidate region is processed;
the basic steps of the mean shift and normalization segmentation method are that the image is segmented into small regions, and then the two regions with the highest merging possibility are repeated until the whole image is merged into a region position, wherein all the regions which are used in the process are the candidate regions determined by the method;
step (B), normalizing each final candidate area, wherein the sizes are S, S represents the length and the width of the normalized final candidate area, performing two times of down-sampling on each normalized final candidate area, inputting the final candidate area and the sampling results of the two times of down-sampling into a multi-scale coupling convolution network, and performing parameter training and feature extraction, comprising the following steps,
(B1) performing two times of downsampling on each normalized final candidate region, wherein the final candidate region corresponds to an S/2 final candidate region and an S/4 final candidate region;
(B2) inputting the final candidate region, the S/2 final candidate region and the S/4 final candidate region into a multi-scale coupling convolution network;
(B3) the final candidate region is processed with a convolution kernel of 5 × 5 in the convolution process, the S/2 × S/2 final candidate region is processed with a convolution kernel of 3 × 3 in the convolution process, and the S/4 × S/4 final candidate region is processed with a convolution kernel of 1 × 1 in the convolution process;
(B4) directly combining the features extracted under different scales of the same final candidate region to obtain 7168-dimensional features, including 4096-dimensional features extracted from the final candidate region, 2048-dimensional features extracted from the S/2 final candidate region, and 1024-dimensional features extracted from the S/4 final candidate region;
(B5) sending 7168 dimensional characteristics into a full-link layer of a multi-scale coupling convolution network for classification, wherein the training of parameters of the multi-scale coupling convolution network is carried out by the difference between classification results of the full-link layer and actual labels, and the learning rate is 0.01;
the multi-scale coupling convolution network comprises a convolution layer, a pooling layer and a full-connection layer, and improves the detection efficiency on the premise of ensuring enough precision;
step (C), inputting the extracted features into an extreme learning machine ELM classifier, classifying the extracted features, wherein the process is as follows,
there are N different feature samples, X ═ X1，x2，L，xN|xi∈RD，i＝1，2，L，N}，Y＝{y1，y2，L，yN|yi∈RLAnd i is 1,2, L, N }, wherein D is a spectral dimension, L is a class number, X represents a training feature sample set, Y represents a label set corresponding to each feature sample, and X represents a label set corresponding to each feature sampleNRepresents the Nth feature sample, yNLabel, x, representing the Nth characteristic sampleiRepresents the i-th feature sample, yiThe label of the ith characteristic sample is represented, and R represents the value range of the sample value;
let P be the number of hidden nodes, the expression of the extreme learning machine is as follows,
Figure BDA0001575324730000071
where g (x) is the activation function, ωjAnd betajRespectively an input weight and an output weight, bjBiasing for a hidden layer;
in the step, each type of target is judged by using an extreme learning machine ELM classifier, and the extreme learning machine ELM classifier is a novel single hidden layer feedforward neural network and is characterized in that input weight and bias are initialized randomly and corresponding output is obtained. Because of high computing efficiency and strong generalization capability, the extreme learning machine is widely applied in the field of image processing in recent years;
step (D) of correcting the position of the candidate frame by using a regressor based on the classified features to obtain the position of the abnormal target in the image information, i.e., the position of the corrected candidate frame,
and correcting the position of the candidate frame by using a regressor to obtain five parameters of the candidate frame, wherein the five parameters comprise a category label C and a candidate frame position (x, y, w, h), wherein x is the horizontal coordinate of the upper left vertex of the candidate frame, y is the vertical coordinate of the upper left vertex of the candidate frame, w is the width of the candidate frame, and h is the height of the candidate frame.
The metric for the anomalous target detection problem is the overlap area: many seemingly accurate detection results often have a small overlap area because the candidate boxes are not accurate enough. Therefore, the position of the candidate frame needs to be corrected, a linear ridge regression is selected for each type of target to be refined, 10000 is taken as a regular term, 7168 dimensional features obtained by a final pooling layer of a multi-scale coupling convolution network are input, and scaling and translation in the x and y directions are output.
In conclusion, the automatic detection method for the abnormal target of the railway contact network based on the multi-scale coupling convolution network is suitable for the automatic detection of the abnormal target of the railway contact network, can obtain more accurate abnormal detection effect than the observation of human eyes, has high detection precision and high automation degree, and has the following advantages,
(1) extracting candidate regions of the image by adopting a mean shift and normalization segmentation method, and merging the candidate regions obtained by the two methods by using a shared region merging method to obtain a final candidate region, so that the accuracy of extracting the candidate regions is improved;
(2) the final candidate region category is analyzed, and a multi-scale coupling convolution network is used for extracting the features of the candidate region, so that the spatial distribution features of the picture can be reflected to the maximum extent, and the anomaly detection effect of a contact network on a railway is more stable and reliable;
(3) the characteristics are distinguished through the extreme learning machine, the calculation efficiency is high, the generalization capability is strong, and the detection precision is further improved.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (5)

1. The automatic detection method for the abnormal targets of the railway contact network based on the multi-scale coupling convolution network is characterized by comprising the following steps of: comprises the following steps of (a) carrying out,
step (A), acquiring image information of a contact network, respectively acquiring a plurality of groups of candidate regions corresponding to the image information by using a mean shift and normalized segmentation method, and merging the acquired candidate regions by using a shared region merging method to acquire a plurality of final candidate regions of the image;
step (B), normalizing each final candidate area, wherein the sizes are S × S, S represents the length and the width of the normalized final candidate area, performing two times of down-sampling on each normalized final candidate area, inputting the final candidate area and the sampling results of the two times of down-sampling into a multi-scale coupling convolution network, and performing parameter training and feature extraction;
inputting the extracted features into an ELM classifier to classify the extracted features;
step (D), according to the classified features, correcting the positions of the candidate frames by using a regressor, thereby obtaining the positions of the abnormal targets in the image information, namely the positions of the corrected candidate frames;
a step (A) of merging the obtained candidate regions by a shared region merging method to obtain a plurality of final candidate regions of the image, comprising the steps of,
(A1) analyzing the candidate regions generated by the mean shift method one by one, and comparing each candidate region with the corresponding candidate region generated by the normalized segmentation method one by one;
(A2) if the similarity between the candidate region generated by the normalized segmentation method and the candidate region generated by the mean shift method under analysis can be found to be 80% or more, the segmentation of the candidate region is correct, and a final candidate region is obtained by taking a union set of the two coincident candidate regions; and if the similarity between the candidate region generated by the normalized segmentation method and the candidate region generated by the mean shift method under analysis does not reach 80%, the candidate region is considered to be incorrectly divided, and the next candidate region is processed.
2. The automatic detection method for the abnormal targets of the railway catenary based on the multi-scale coupling convolutional network as claimed in claim 1, which is characterized in that: step (B), each normalized final candidate area is down-sampled twice, and the final candidate area and the sampling results of the two down-samplings are input into a multi-scale coupling convolution network for parameter training and feature extraction, comprising the following steps,
(B1) performing two times of downsampling on each normalized final candidate region to obtain an S/2 final candidate region and an S/4 final candidate region corresponding to the final candidate region;
(B2) inputting the final candidate region, the S/2 final candidate region and the S/4 final candidate region into a multi-scale coupling convolution network;
(B3) the final candidate region is processed with a convolution kernel of 5 × 5 in the convolution process, the S/2 × S/2 final candidate region is processed with a convolution kernel of 3 × 3 in the convolution process, and the S/4 × S/4 final candidate region is processed with a convolution kernel of 1 × 1 in the convolution process;
(B4) directly combining the extracted features under different scales of the same final candidate region to obtain 7168-dimensional features;
(B5) and sending the 7168-dimensional features into a full-link layer of the multi-scale coupling convolution network for classification, wherein the training of parameters of the multi-scale coupling convolution network is carried out by distinguishing classification results of the full-link layer from actual labels, and the learning rate is 0.01.
3. The automatic detection method for the abnormal targets of the railway catenary based on the multi-scale coupling convolutional network as claimed in claim 2, which is characterized in that: (B4) and directly combining the features extracted under different scales of the same final candidate region to obtain 7168-dimensional features, including 4096-dimensional features extracted from the final candidate region, 2048-dimensional features extracted from the S/2 final candidate region and 1024-dimensional features extracted from the S/4 final candidate region.
4. The automatic detection method for the abnormal targets of the railway catenary based on the multi-scale coupling convolutional network as claimed in claim 1, which is characterized in that: step (C), inputting the extracted features into an extreme learning machine ELM classifier, classifying the extracted features, and obtaining the candidate frame position corresponding to the final candidate region, wherein the process is as follows,
there are N different feature samples, X ═ X1,x2,…,xN|xi∈RD,i＝1,2,…,N}，Y＝{y1,y2,…,yN|yi∈RLAnd i is 1,2, …, N }, wherein D is a spectral dimension, L is a class number, X represents a training feature sample set, Y represents a label set corresponding to each feature sample, and X represents a label set corresponding to each feature sampleNRepresents the Nth feature sample, yNLabel, x, representing the Nth characteristic sampleiRepresents the i-th feature sample, yiThe label of the ith characteristic sample is represented, and R represents the value range of the sample value;
let P be the number of hidden nodes, the expression of the extreme learning machine is as follows,
Figure FDA0003096785860000031
where g (x) is the activation function, ωjAnd betajRespectively an input weight and an output weight, bjIs a hidden layer bias.
5. The automatic detection method for the abnormal targets of the railway catenary based on the multi-scale coupling convolutional network as claimed in claim 1, which is characterized in that: and (D) correcting the position of the candidate frame by using a regressor according to the position of the candidate frame obtained in the step (C) to obtain five parameters of the candidate frame, wherein the five parameters comprise a category label C and the position (x, y, w, h) of the candidate frame, x is the horizontal coordinate of the upper left vertex of the candidate frame, y is the vertical coordinate of the upper left vertex of the candidate frame, w is the width of the candidate frame, and h is the height of the candidate frame.
CN201810132521.0A 2018-02-09 2018-02-09 Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network Active CN108288273B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810132521.0A CN108288273B (en) 2018-02-09 2018-02-09 Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810132521.0A CN108288273B (en) 2018-02-09 2018-02-09 Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network

Publications (2)

Publication Number Publication Date
CN108288273A CN108288273A (en) 2018-07-17
CN108288273B true CN108288273B (en) 2021-07-27

Family

ID=62832826

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810132521.0A Active CN108288273B (en) 2018-02-09 2018-02-09 Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network

Country Status (1)

Country Link
CN (1) CN108288273B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109034366B (en) * 2018-07-18 2021-10-01 北京化工大学 Application of ELM integrated model based on multiple activation functions in chemical engineering modeling
CN110210474B (en) * 2019-04-30 2021-06-01 北京市商汤科技开发有限公司 Target detection method and device, equipment and storage medium
CN110432895B (en) * 2019-08-09 2022-02-11 上海鹰瞳医疗科技有限公司 Training data processing method, electrocardiographic waveform detection method and electronic equipment

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8345976B2 (en) * 2010-08-06 2013-01-01 Sony Corporation Systems and methods for segmenting digital images
CN104156942A (en) * 2014-07-02 2014-11-19 华南理工大学 Detection method for remnants in complex environment
CN105160330A (en) * 2015-09-17 2015-12-16 中国地质大学（武汉） Vehicle logo recognition method and vehicle logo recognition system
CN106611427A (en) * 2015-10-21 2017-05-03 中国人民解放军理工大学 A video saliency detection method based on candidate area merging

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8345976B2 (en) * 2010-08-06 2013-01-01 Sony Corporation Systems and methods for segmenting digital images
CN104156942A (en) * 2014-07-02 2014-11-19 华南理工大学 Detection method for remnants in complex environment
CN105160330A (en) * 2015-09-17 2015-12-16 中国地质大学（武汉） Vehicle logo recognition method and vehicle logo recognition system
CN106611427A (en) * 2015-10-21 2017-05-03 中国人民解放军理工大学 A video saliency detection method based on candidate area merging

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Image segmentation based on genetic algorithm for region growth and region merging;S Angelina et al.;《 2012 International Conference on Computing, Electronics and Electrical Technologies》;20120524;全文 *
基于马尔可夫随机场的图像分割方法综述;徐胜军等;《计算机应用研究》;20130930;全文 *

Also Published As

Publication number Publication date
CN108288273A (en) 2018-07-17

Similar Documents

Publication Publication Date Title
CN107437245B (en) High-speed railway contact net fault diagnosis method based on deep convolutional neural network
US9639748B2 (en) Method for detecting persons using 1D depths and 2D texture
CN111784633B (en) Insulator defect automatic detection algorithm for electric power inspection video
CN109492706B (en) Chromosome classification prediction device based on recurrent neural network
CN108288273B (en) Automatic detection method for abnormal targets of railway contact network based on multi-scale coupling convolution network
Umamaheswari et al. Weed detection in farm crops using parallel image processing
CN105044122A (en) Copper part surface defect visual inspection system and inspection method based on semi-supervised learning model
CN116665011A (en) Coal flow foreign matter identification method for coal mine belt conveyor based on machine vision
Wah et al. Analysis on feature extraction and classification of rice kernels for Myanmar rice using image processing techniques
Tu et al. Instance segmentation based on mask scoring R-CNN for group-housed pigs
Huo et al. Tongue shape classification integrating image preprocessing and convolution neural network
CN114973116A (en) Method and system for detecting foreign matters embedded into airport runway at night by self-attention feature
Li et al. Temporal aggregation network using micromotion features for early lameness recognition in dairy cows
CN113191320A (en) Food safety index monitoring method based on deep learning
EP3971766A1 (en) Dairy cattle nipple detection convolutional neural network model and construction method therefor
CN111814547A (en) Overhead high-voltage wire tracking detection system and method based on visual identification
CN114792300A (en) Multi-scale attention X-ray broken needle detection method
CN106156771B (en) water meter reading area detection algorithm based on multi-feature fusion
Zhang et al. An approach for goose egg recognition for robot picking based on deep learning
Mohanapriya et al. Identification of Fruit Disease Using Instance Segmentation
CN111754547B (en) Multi-target automatic identification tracking positioning method for broiler chickens in dim light environment
Mirra et al. Classification of Fruits using Deep Learning Algorithms
Gurunathan et al. Plant Leaf Diseases Detection Using KNN Classifier
CN117593300B (en) PE pipe crack defect detection method and system
Xu et al. Defect detection of automotive leather based on Nanodet-Plus

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
TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20191210

Address after: 210012 floor 3-012, North Building D2, No. 32, Dazhou Road, Yuhuatai District, Nanjing City, Jiangsu Province

Applicant after: Nanjing zhiliansen Information Technology Co., Ltd

Address before: 210014 9 fresh 506 homes in Guanghua Road, Qinhuai District, Nanjing, Jiangsu

Applicant before: Wu Zebin

GR01 Patent grant
GR01 Patent grant