CN108038502A - Object collaborative detection method based on convolutional neural networks - Google Patents

Object collaborative detection method based on convolutional neural networks Download PDF

Info

Publication number
CN108038502A
CN108038502A CN201711295915.XA CN201711295915A CN108038502A CN 108038502 A CN108038502 A CN 108038502A CN 201711295915 A CN201711295915 A CN 201711295915A CN 108038502 A CN108038502 A CN 108038502A
Authority
CN
China
Prior art keywords
group
fraction
candidate
picture
image
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
Application number
CN201711295915.XA
Other languages
Chinese (zh)
Inventor
孟凡满
郭莉丽
罗堃铭
施雯
李宏亮
吴庆波
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.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
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 University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201711295915.XA priority Critical patent/CN108038502A/en
Publication of CN108038502A publication Critical patent/CN108038502A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Abstract

The present invention provides a kind of object collaborative detection method based on convolutional neural networks, comprises the following steps：Step 1：On one group of associated picture of content, using existing candidate's generation method, the candidate prediction result and objectivity fraction of every width picture in this group of image are obtained；Step 2：The candidate prediction result obtained based on step 1 obtains the repeated matrix of the group picture piece；Step 3：With reference to objectivity fraction and the repeated matrix of the group picture piece, the prediction result of the final consideration object repeatability of this group of image is obtained.The repeatability of object collaborative detection method combination single image target prediction and multiple image content proposed by the present invention based on convolutional neural networks, more relationship informations are added for detection prediction result, enhance the specific aim of traditional detection and the detection result to selecting classification target.

Description

Object collaborative detection method based on convolutional neural networks
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of object cooperation detection based on convolutional neural networks Method.
Background technology
With the fast development of computer vision technique, digital image data increases substantially, and how effectively to handle this A little data and therefrom acquisition key message are receive much concern the problem of.Target detection is image procossing and computer vision One important branch, in the field extensive use such as intelligent monitor system, industrial detection, intelligent transportation system, aerospace.Image Be people obtain external information an important sources, and most of information of image come from it includes destination object, to figure The target detection of piece is one of computer vision field task the most basic.Target detection technique, is exactly detected in picture And the target of certain setting species is positioned, so as to obtain the key message of image, this is very important one in image processing process Step.Therefore the effect of target detection is further improved, is of great significance for obtaining more accurate image information tool.
Had been widely used at present in computer vision application field, deep neural network, it passes through feature extraction With Feature Mapping process, it can preferably learn the invariant feature into image, can adaptively construction feature be retouched under training State, there is higher flexibility and universality.Wherein convolutional neural networks are the god with reference to artificial neural network and Deep Learning It is special through network system, the overall situation training combined with local sensing region, hierarchically structured, feature extraction and assorting process etc. Point, is widely used in target detection, image segmentation, conspicuousness detection etc., to improve the processing for view data.
In recent years, domestic and international association area had carried out the target detection based on deep neural network to study and obtain for many years Many achievements.With the rapid development of multimedia technology, it is gradual that similar or consistent target is found from multiple image As a kind of new demand.Target detection is carried out using deep neural network, existing method is roughly divided into two classes, first, being based on RCNN methods, its main flow are：The extraction of candidate frame is carried out using gathering or sliding window formula method on the basis of artwork, is obtained To a series of candidate frame；The feature of image candidate is extracted using convolutional neural networks, then is classified based on these features Prediction, obtains multiple prediction results and carries out recurrence amendment.Another kind is the method directly predicted based on picture, such as YOLO pin To entire image, 98 prediction blocks and each prediction fraction are drawn from the angle of recurrence.These researchs are based on single at present The method of width image, the information that so only only used single image is not comprehensive.It is mutually related image for one group of content, Cooperation detection is mutually aided in by the association between considering multiple image, similar target is found from multiple image, effectively Using one group of associated advantage of picture material, at the same can to the uncorrelated factor such as the non-targeted object of each image or background into Row suppresses.Therefore, it is worthy of expecting for the cooperation detection effect of multiple image for compared to single image.
The content of the invention
It is an object of the invention to improve the problems of the above-mentioned prior art, propose that a kind of content that is based on is associated picture The method of the object cooperation detection of group and convolutional neural networks, to add several figures in the detection method based on single picture originally The information of piece repeatability, for improving the specific aim and accuracy of detection.We conducted substantial amounts of experiment, it was demonstrated that the detection side Method can combine the prediction result of single image and the repeatability between several, so that improving content is associated the inspection of picture group Effect is surveyed, there is good applicability.
A kind of object collaborative detection method based on convolutional neural networks, comprises the following steps：
Step 1：On one group of associated picture of content, using existing candidate's generation method, obtain every in this group of image The candidate prediction result and objectivity fraction of width picture；
Step 2：The candidate prediction result obtained based on step 1 obtains the repeated matrix of the group picture piece；
Step 3：With reference to objectivity fraction and the repeated matrix of the group picture piece, the final consideration pair of this group of image is obtained As the prediction result of repeatability.
Further, method as described above, the step 2 include：
Step 1：All candidate's picture blocks of a group picture piece obtained for step 1, are obtained often based on similarity measurement network Similitude is matched as a result, obtaining the affinity score of each two between two candidate predictions；
Step 2：The repeated matrix of the group picture piece candidate prediction is generated according to similarity scores.
Further, method as described above, the step 3 include：
Step a：The repeated matrix obtained for step 2, repeated fraction is used as using its maximal eigenvector；
Step b：The objectivity fraction that will be predicted in step 1, and the repeated fraction combination in step a, obtain final Prediction result fraction.
Beneficial effect：
Object collaborative detection method combination single image target prediction proposed by the present invention based on convolutional neural networks and The repeatability of multiple image content, adds more relationship informations for detection prediction result, enhances the pin of traditional detection Detection result to property and to selected classification target, on the basis of single image detection, with reference to the content between image in group Contact so that network uses the relevance of content between image, so as to improve detection network for one group of association picture Detection result.
Brief description of the drawings
Fig. 1 is the object collaborative detection method flow chart of the invention based on convolutional neural networks.
Fig. 2 is the practical operation flow chart of the object collaborative detection method of the invention based on convolutional neural networks.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, below technical solution in the present invention carry out it is clear Chu, be fully described by, it is clear that described embodiment is part of the embodiment of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, those of ordinary skill in the art are obtained every other without creative efforts Embodiment, belongs to the scope of protection of the invention.
Fig. 1 is the object collaborative detection method flow chart of the invention based on convolutional neural networks, and Fig. 2 is the present invention based on volume The practical operation flow chart of the object collaborative detection method of product neutral net, with reference to Fig. 1 and its Fig. 2, method provided by the invention Comprise the following steps：
Step 1: on one group of associated picture of content, using existing candidate's generation method, obtain every in this group of image The candidate prediction result and objectivity fraction of width picture.
Specifically, which includes following 3 steps：
1. being associated the picture set A of picture firstly, for one group of n width content, extracted to obtain each width with YOLO methods Picture aiThe predicting candidate frame of (i=1,2 ..., n).
2. assuming that every width picture generates m candidate prediction, then n × m candidate of this group of n width picture can be obtained altogether FrameAs candidate prediction as a result, wherein i is picture label, p is a width picture Candidate frame label.
3. meanwhile obtain the possibility fraction that each candidate frame includes target That is the objectivity fraction of i-th p-th of candidate frame of width figure.
Step 2: repeatability point is obtained using patch match similarity measurement networks based on the candidate prediction result Matrix number.
Patch match similarity measurement network operating principles are as follows：Selection needs matched two segments, first by its ruler It is very little to normalize to 64 × 64, gray level image is then converted to, a twin-channel image is finally merged into and is used as input, output As a result it is final defeated with it for the matching fraction of two picture blocks of input, the matching degree of two picture blocks, that is, similarity degree It is directly proportional to go out fraction.
1. on the basis of step 1 obtains n × m candidate prediction, by the every m candidate frame of width picture of n width pictures,Matching fraction is respectively obtained between each two, willWith Inputted two-by-two as picture block In patch match similarity measurement networks, each two obtains a similarity scores k in n × m candidatei,j。
2. the similarity scores k between n × m picture blocki,jForm the repeated matrix of this group of image, matrix Scale is (n × m) × (n × m).Wherein i-th of candidate of digital representation of the i-th row jth row of matrix and j-th candidates match Similarity scores ki,j。
Step 3: draw final prediction result with reference to two fractions.
1. after step 2 obtains the repeated matrix of this group of pictures A, be calculated the maximum feature of the matrix to Amount, the maximal eigenvector of matrix represent feature the most obvious in this matrix, after normalized, with this n × m tie up to Measure conduct1,2 ..., m) repeated fraction Corresponding is in this group picture pieceThe repeated fraction of candidate blocks, represents in this group of n width picture, the candidate frame The multiplicity that the similar candidate frame of content occurs.The number that the content that fraction is higher to be represented in the candidate occurs is more, be exactly Multiplicity is higher in different pictures.
2. it is last, with reference in step 1In previous step By this Two fractions are calculated according to the following formula, and comprehensive objectivity fraction and repeated fraction, obtain final prediction fraction Wherein λ is balance pair As property fraction and the constant parameter of repeated fraction, general 0.8≤λ≤1.0.The scoring is higher, illustrates institute in corresponding candidate frame Comprising object be in pictures A share target probability it is bigger.Thus can be by combining other phases in a group picture piece The information of picture is closed, to achieve the purpose that the candidate's sequence for correcting every width picture, so as to more optimize the quality level of candidate.
Finally it should be noted that：The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although The present invention is described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that：It still may be used To modify to the technical solution described in foregoing embodiments, or equivalent substitution is carried out to which part technical characteristic； And these modification or replace, do not make appropriate technical solution essence depart from various embodiments of the present invention technical solution spirit and Scope.

Claims (3)

1. a kind of object collaborative detection method based on convolutional neural networks, it is characterised in that with reference to several content associated images Relevant information, comprise the following steps：
Step 1：On one group of associated picture of content, using existing candidate's generation method, every width figure in this group of image is obtained The candidate prediction result and objectivity fraction of piece；
Step 2：The candidate prediction result obtained based on step 1 obtains the repeated matrix of the group picture piece；
Step 3：With reference to objectivity fraction and the repeated matrix of the group picture piece, the final consideration object weight of this group of image is obtained The prediction result of renaturation.
2. according to the method described in claim 1, it is characterized in that, the step 2 includes：
Step 1：All candidate's picture blocks of a group picture piece obtained for step 1, each two is obtained based on similarity measurement network Similitude is matched as a result, obtaining the similarity scores of each two between candidate prediction；
Step 2：The repeated matrix of the group picture piece candidate prediction is generated according to similarity scores.
3. according to the method described in claim 2, it is characterized in that, the step 3 includes：
Step a：The repeated matrix obtained for step 2, the repeatability of the group picture piece point is used as using its maximal eigenvector Number；
Step b：The objectivity fraction that will be predicted in step 1, and the repeated fraction combination in step a, obtain final prediction Fraction.
CN201711295915.XA 2017-12-08 2017-12-08 Object collaborative detection method based on convolutional neural networks Pending CN108038502A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711295915.XA CN108038502A (en) 2017-12-08 2017-12-08 Object collaborative detection method based on convolutional neural networks

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711295915.XA CN108038502A (en) 2017-12-08 2017-12-08 Object collaborative detection method based on convolutional neural networks

Publications (1)

Publication Number Publication Date
CN108038502A true CN108038502A (en) 2018-05-15

Family

ID=62101574

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711295915.XA Pending CN108038502A (en) 2017-12-08 2017-12-08 Object collaborative detection method based on convolutional neural networks

Country Status (1)

Country Link
CN (1) CN108038502A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111046917A (en) * 2019-11-20 2020-04-21 南京理工大学 Object-based enhanced target detection method based on deep neural network

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102982539A (en) * 2012-11-09 2013-03-20 电子科技大学 Characteristic self-adaption image common segmentation method based on image complexity
CN103268607A (en) * 2013-05-15 2013-08-28 电子科技大学 Common object detection method on weak supervision condition
US8885887B1 (en) * 2012-01-23 2014-11-11 Hrl Laboratories, Llc System for object detection and recognition in videos using stabilization
CN106407891A (en) * 2016-08-26 2017-02-15 东方网力科技股份有限公司 Target matching method based on convolutional neural network and device
CN106504255A (en) * 2016-11-02 2017-03-15 南京大学 A kind of multi-Target Image joint dividing method based on multi-tag multi-instance learning
CN107133955A (en) * 2017-04-14 2017-09-05 大连理工大学 A kind of collaboration conspicuousness detection method combined at many levels
CN107330750A (en) * 2017-05-26 2017-11-07 北京三快在线科技有限公司 A kind of recommended products figure method and device, electronic equipment
US20170330059A1 (en) * 2016-05-11 2017-11-16 Xerox Corporation Joint object and object part detection using web supervision
CN107392251A (en) * 2017-07-26 2017-11-24 成都快眼科技有限公司 A kind of method that target detection network performance is lifted using category images
CN107437246A (en) * 2017-07-05 2017-12-05 浙江大学 A kind of common conspicuousness detection method based on end-to-end full convolutional neural networks

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8885887B1 (en) * 2012-01-23 2014-11-11 Hrl Laboratories, Llc System for object detection and recognition in videos using stabilization
CN102982539A (en) * 2012-11-09 2013-03-20 电子科技大学 Characteristic self-adaption image common segmentation method based on image complexity
CN103268607A (en) * 2013-05-15 2013-08-28 电子科技大学 Common object detection method on weak supervision condition
US20170330059A1 (en) * 2016-05-11 2017-11-16 Xerox Corporation Joint object and object part detection using web supervision
CN106407891A (en) * 2016-08-26 2017-02-15 东方网力科技股份有限公司 Target matching method based on convolutional neural network and device
CN106504255A (en) * 2016-11-02 2017-03-15 南京大学 A kind of multi-Target Image joint dividing method based on multi-tag multi-instance learning
CN107133955A (en) * 2017-04-14 2017-09-05 大连理工大学 A kind of collaboration conspicuousness detection method combined at many levels
CN107330750A (en) * 2017-05-26 2017-11-07 北京三快在线科技有限公司 A kind of recommended products figure method and device, electronic equipment
CN107437246A (en) * 2017-07-05 2017-12-05 浙江大学 A kind of common conspicuousness detection method based on end-to-end full convolutional neural networks
CN107392251A (en) * 2017-07-26 2017-11-24 成都快眼科技有限公司 A kind of method that target detection network performance is lifted using category images

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
DENG-PING FAN 等: "Structure-Measure: A New Way to Evaluate Foreground Maps", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 *
FANMAN MENG 等: "A New Deep Segmentation Quality Assessment Network for Refining Bounding Box Based Segmentation", 《IEEE ACCESS》 *
FANMAN MENG 等: "Weakly Supervised Part Proposal Segmentation From Multiple Images", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 *
HONGLIANG LI 等: "Co-Salient Object Detection From Multiple Images", 《IEEE TRANSACTIONS ON MULTIMEDIA》 *
WEN SHI 等: "Shape based co-segmentation repairing by segment evaluation and object proposals", 《2016 VISUAL COMMUNICATIONS AND IMAGE PROCESSING》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111046917A (en) * 2019-11-20 2020-04-21 南京理工大学 Object-based enhanced target detection method based on deep neural network

Similar Documents

Publication Publication Date Title
CN106547880B (en) Multi-dimensional geographic scene identification method fusing geographic area knowledge
CN104881689B (en) A kind of multi-tag Active Learning sorting technique and system
CN110633708A (en) Deep network significance detection method based on global model and local optimization
AU2017101803A4 (en) Deep learning based image classification of dangerous goods of gun type
CN109558902A (en) A kind of fast target detection method
Zhang et al. An efficient lightweight convolutional neural network for industrial surface defect detection
Cai et al. Improving sampling-based image matting with cooperative coevolution differential evolution algorithm
CN110287952A (en) A kind of recognition methods and system for tieing up sonagram piece character
CN107133854A (en) Information recommendation method and device
Zhu et al. A modified deep neural network enables identification of foliage under complex background
CN112712127A (en) Image emotion polarity classification method combined with graph convolution neural network
Cao et al. 3D convolutional siamese network for few-shot hyperspectral classification
CN111507285A (en) Face attribute recognition method and device, computer equipment and storage medium
CN105608457A (en) Histogram gray moment threshold segmentation method
CN110580339A (en) Method and device for perfecting medical term knowledge base
Jiang et al. Tabcellnet: Deep learning-based tabular cell structure detection
CN108428234B (en) Interactive segmentation performance optimization method based on image segmentation result evaluation
Wang et al. A novel stochastic block model for network-based prediction of protein-protein interactions
CN107452003A (en) A kind of method and device of the image segmentation containing depth information
CN108038502A (en) Object collaborative detection method based on convolutional neural networks
CN112541010B (en) User gender prediction method based on logistic regression
CN107563327B (en) Pedestrian re-identification method and system based on self-walking feedback
CN108845999B (en) Trademark image retrieval method based on multi-scale regional feature comparison
WO2024021321A1 (en) Model generation method and apparatus, electronic device, and storage medium
CN115269816A (en) Core personnel mining method and device based on information processing method and storage medium

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: 20180515