CN110349148A - Image target detection method based on weak supervised learning - Google Patents
Image target detection method based on weak supervised learning Download PDFInfo
- Publication number
- CN110349148A CN110349148A CN201910624287.8A CN201910624287A CN110349148A CN 110349148 A CN110349148 A CN 110349148A CN 201910624287 A CN201910624287 A CN 201910624287A CN 110349148 A CN110349148 A CN 110349148A
- Authority
- CN
- China
- Prior art keywords
- classification
- image
- convolutional neural
- neural networks
- thermodynamic chart
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 38
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 38
- 238000004458 analytical method Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 27
- 238000000034 method Methods 0.000 claims abstract description 21
- 238000003062 neural network model Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 24
- 230000004044 response Effects 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 12
- 239000000284 extract Substances 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000013480 data collection Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 230000007423 decrease Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 abstract description 4
- 238000002372 labelling Methods 0.000 abstract 1
- 238000013461 design Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 241000282326 Felis catus Species 0.000 description 1
- 241001494479 Pecora Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000004382 potting Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- 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
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Quality & Reliability (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an image target detection method based on weak supervised learning, and belongs to the technical field of machine vision. Firstly, collecting an image data set, and training a constructed deep convolutional neural network model based on a multi-scale characteristic diagram by adopting a multi-example learning method; then inputting an actual image, and extracting a category thermodynamic diagram of the actual image through a depth convolution neural network model; and finally, outputting the bounding box of the target in the category thermodynamic diagram by adopting a binary image connected region analysis method to obtain a target detection result. The method realizes the image target detection task by using the weak supervision-based learning method, labels in the convolutional neural network model training can finish the target detection task only by using image-level classification label information, and the label information is different from the label information of a target enclosure frame required in the prior art, so that the work of manually labeling targets in images is greatly reduced, and the image target detection task is more economic.
Description
Technical field
The invention belongs to nerual network technique fields, and in particular to a kind of image object detection side based on Weakly supervised study
Method.
Background technique
With deep learning in Computer Vision Task using more and more extensive, instantly deep neural network model is being schemed
As being excellent in the tasks such as classification, target detection, semantic segmentation, but supervised learning formula training pattern need it is largely artificial
Markup information, the relevant mark of these markup informations, especially position, often expends a large amount of manpower and material resources, therefore to mark
Note information, which relies on lower Weakly supervised learning method, becomes research hotspot.Weakly supervised study is a kind of mode of machine learning,
It is different from supervised learning model needs mark to correspond with model output, the markup information that Weakly supervised study relies on only needs portion
Hierarchical mark, therefore Weakly supervised study has a good application prospect in actual computer visual task and economic benefit.
(1) image characteristics extraction
Traditional characteristic extracts the means for mainly using image procossing, hand-designed feature mode and corresponding extraction side
A large amount of artificial trace has been incorporated during method, design and extraction, while labor intensive, has also been unfavorable for from data itself
Angle carry out information excavating.The method of convolutional neural networks then utilizes network to be capable of the characteristic of self study, realizes from design
Artificial excessively intervention is avoided in journey as best one can, while the process of simplification, moreover it is possible to reach recognition effect more better than conventional method,
And the multilayered structure of deep neural network can learn to higher, more abstract expression, using multi-scale feature fusion
Mode can extract the more exact feature of image.
(2) target detection
Object detection task is common a kind of Computer Vision Task in production and living, it is required that model exports target
Encirclement frame target category corresponding with the frame, in subsequent task.The common target based on depth convolutional network
Detection model all uses the encirclement frame manually marked in training as supervision message, usually using neural net regression target
The thinking of frame coordinate is completed.Object detection method based on Weakly supervised study does not depend on the encirclement frame information manually marked, greatly
Reduce the manpower and material resources cost in mark work.
Summary of the invention
It is an object of the invention to: to solve the existing image object detection method based on deep learning for manually marking
The technical issues of depending on unduly, design depth convolutional neural networks model extraction based on Analysis On Multi-scale Features figure and generate image
Classification thermodynamic chart, the classification thermodynamic chart based on model output realize image object Detection task, propose a kind of based on Weakly supervised
The image object detection method of habit.
The technical solution adopted by the invention is as follows:
A kind of image object detection method based on Weakly supervised study, this method comprises the following steps:
Step 1: image data set is collected, using the depth based on Analysis On Multi-scale Features figure of multi-instance learning method training building
Spend convolutional neural networks model;
Step 2: input real image passes through the classification thermodynamic chart of depth convolutional neural networks model extraction real image;
Step 3: using the encirclement frame of target in binary image connected component analysis method output classification thermodynamic chart, obtaining
Object detection results.
Preferably, the step 1 includes the following steps:
Step 1.1: compiling image data, normalized is done to picture size, and more heat coding marks are made to image
Note;
Step 1.2: depth convolutional neural networks model of the building based on Analysis On Multi-scale Features figure, depth convolutional neural networks mould
The core network model of type uses the model through public data collection pre-training;
Step 1.3: the depth convolutional neural networks model of building is trained using multi-instance learning method, wherein labeled data
Use more heat coding marks in step 1.1.
Preferably, step 2.2 includes the following steps:
Step 2.2.1: choosing the different down-sampling stages in the core network model of depth convolutional neural networks model, right
The real image of input extracts Analysis On Multi-scale Features by feature extraction network;
Step 2.2.2: Analysis On Multi-scale Features carry out global multiple dimensioned pond after convolutional layer is converted, and export multi-class point
Class probability value.
Preferably, the step 1.3 includes the following steps:
Step 1.3.1: predicted value is exported as multi-class by more heat coding marks and depth convolutional neural networks model
Probability, to calculate cross entropy loss function;Using bloom loss function to the output feature in depth convolutional neural networks model
Figure is constrained, and forces the response on the output characteristic pattern in convolutional network model to be drawn close toward certainty eminence, and this is handed over
Pitch the overall loss function of the sum of entropy loss function and bloom loss function as model;
Step 1.3.2: gradient decline optimization, training depth convolution are carried out using optimizer to total losses function in training
Neural network model is until convergence, i.e., overall loss function fluctuating range is maintained within 0.1 in 5 wheel training.
Preferably, the step 2 includes the following steps:
Step 2.1: dimension normalization processing is carried out for the real image of input;
Step 2.2: the real image through normalized is obtained into each spy by depth convolutional neural networks model treatment
Levy the classification thermodynamic chart of scale;
Step 2.3: the classification thermodynamic chart after width Fusion Features that the classification thermodynamic chart of each characteristic dimension is permeated.
Preferably, the step 2.2 includes the following steps:
Step 2.2.1: choosing the different down-sampling stages in the core network model of depth convolutional neural networks model, right
The real image of input extracts Analysis On Multi-scale Features by feature extraction network;
Step 2.2.2: Analysis On Multi-scale Features carry out global multiple dimensioned pond after convolutional layer is converted, and export multi-class point
Class probability value.
Preferably, the step 2.3 includes the following steps:
Step 2.3.1: the corresponding classification thermodynamic chart of all characteristic dimensions is sampled to the practical figure after normalized
The size of picture;
Step 2.3.2: in each position, the response to the corresponding classification thermodynamic chart of multiple characteristic dimensions is averaged, and is obtained
Classification thermodynamic chart after Fusion Features.
Preferably, specific step is as follows in the step 3:
Step 3.1: using the classification results that depth convolutional neural networks model exports as the classification foundation of target, selecting class
There are the corresponding characteristic patterns of class in other thermodynamic chart;Wherein, output category probability value is thought more than or equal to 0.5 with the presence of such,
Otherwise it is assumed that such is not present;
Step 3.2: the response mean value in the corresponding characteristic pattern of each class is calculated, using response mean value as binaryzation threshold
Value, and by the corresponding characteristic pattern binaryzation of each class;
Step 3.3: using connected component analysis method, take 8- domain pattern, the characteristic pattern after binaryzation is connected
Logical regional analysis marks target area of each region as corresponding class;
Step 3.4: surrounding frame for the minimum circumscribed rectangle of the target area of each class as corresponding classification target;
Step 3.5: will be all there are the corresponding encirclement frame output of classification in classification thermodynamic chart, complete image object detection.
In conclusion by adopting the above-described technical solution, the beneficial effect that the present invention is different from the prior art is:
1, the present invention is based on Weakly supervised learning method using one kind, and method is mainly by using multi-instance learning
Method trains depth convolutional neural networks model, then real image is input to the depth convolutional neural networks model trained
In, the classification thermodynamic chart of real image is extracted, classification thermodynamic chart is finally exported using binary image connected component analysis method
The encirclement frame of middle target, can be obtained object detection results.The present invention, which is used, realizes image mesh based on Weakly supervised learning method
Detection task is marked, image level classification annotation information, which is used only, in the mark in convolutional neural networks model training can be completed target
Detection task, target needed for being different from the prior art surround frame markup information, greatly reduce target in artificial mark image
Work so that complete image object Detection task have more economic benefit.
It 2,, can be from different rulers by constructing the depth convolutional neural networks model based on Analysis On Multi-scale Features in the present invention
It spends scope and extracts feature, the classification thermodynamic chart of generation can more react the response region of target, be different from existing convolutional network
The single disadvantage of feature extraction scale caused by model structure, the present invention in depth convolutional neural networks model structure guarantee
The accuracy of target detection.
3, in the present invention, propose that the new loss function of one kind is introduced in training convolutional network model is referred to as high light loss
Function, it acts as the output characteristic patterns in constraint convolutional network model, and then the classification thermodynamic chart for generating model finally is more
Stick on conjunction target zone, compared to do not use the loss function model, further increase target detection accuracy and effectively
Property.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow diagrams of the image object detection method of Weakly supervised study;
Fig. 2 is the block schematic illustration of the image object detection method based on Weakly supervised study in the present invention;
Fig. 3 is depth convolutional neural networks model schematic in the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.
Present embodiment is detected for a secondary real scene image, and the training of model uses public data collection VOC
Pascal is carried out, and pre-training core network module is using the sorter network trained on public data collection ImageNet
Inception-ResNet-v2 is carried out, and target category (classification results) is aircraft, bicycle, birds, ship, bottle, public transport
Vehicle, car, cat class, chair, ox class, dining table, dog class, horse class, motorcycle, the mankind, potting, sheep class, sofa, train and display
Device etc..
A kind of image object detection method based on Weakly supervised study, this method comprises the following steps:
Step 1: VOC Pascal data set is used, using image data set and the more classification annotations of correspondence image grade as defeated
Enter, more example training are carried out to the depth convolutional neural networks model based on Analysis On Multi-scale Features figure.Specific step is as follows:
Step 1.1: the picture size concentrated to data does normalized, adopts size normalization operation unanimously for image
Sampling obtains more heat coding marks of the corresponding classification of image to 513 × 513, using the classification information in data set;
Step 1.2: depth convolutional neural networks model of the building based on Analysis On Multi-scale Features figure, core network use
Inception-ResNet-v2 model removes the part of classification layer;
Step 1.3: using the depth convolutional neural networks model of the method training building of multi-instance learning.Including following step
It is rapid:
Step 1.3.1: predicted value is exported as multi-class by more heat coding marks and depth convolutional neural networks model
Probability, to calculate cross entropy loss function, formula such as: L=-ylog (y ')-(1-y) log (1-y '), wherein y is model output
Multi-class probability, y ' be image level classification mark;Using bloom loss function to defeated in depth convolutional neural networks model
Characteristic pattern is constrained out, forces the response on the output characteristic pattern in convolutional network model to be drawn close toward certainty eminence, i.e.,
Toward 1 or 0 approach, specifically in the bloom loss function calculation formula of certain position scale feature figure (i, j) are as follows: H (i, j)=- p (i,
J) logp (i, j)-(1-p (i, j)) log (1-p (i, j)), wherein p indicates the response on this feature figure, and i indicates abscissa, j
Indicate ordinate, and the overall loss function by the sum of this cross entropy loss function and bloom loss function as model;
Step 1.3.2: total losses function is carried out under gradient using optimizer (being not limited to Adam optimizer) in training
Drop optimization, training pattern carry out gradient decline optimization, are divided into two stage-trainings, and the first stage is finely adjusted model, and second
Stage-training overall model, training depth convolutional neural networks model is until convergence, i.e., the overall loss function wave in 5 wheel training
Dynamic amplitude is maintained within 0.1.
Step 2: input real image passes through the model extraction trained and merges its classification thermodynamic chart.Specific steps are such as
Under:
Step 2.1: scale being carried out for the real image of input and is sampled to 513 × 513 (normalizeds);
Step 2.2: processed real image is obtained into the classification thermodynamic chart of each characteristic dimension by model;Depth convolution
The classification layer part for giving up core network in neural network model obtains the defeated of different characteristic scale by the down-sampling stage three times
Out and design conversion layer;The following steps are included:
Step 2.2.1: the different down-sampling stages in the core network model of pre-training are chosen, to the real image of input
Analysis On Multi-scale Features (features of i.e. multiple semantic scales) are extracted by feature extraction network;
Step 2.2.2: Analysis On Multi-scale Features carry out global multiple dimensioned pond, specifically, by each scale after convolutional layer is converted
Under corresponding characteristic pattern convert to port number as the characteristic pattern of M × C by one layer of convolutional layer, wherein C is target classification classification number,
Taking 20, M in this example is different characteristic parameter in every one kind, and M often takes 10 in an implementation, hereafter to each M layers in channel dimension
Average pond is carried out, the classification thermodynamic chart that port number is C is obtained, reuses the global pond top-k, that is, take and taken on classification thermodynamic chart
Maximum output of the k value as this layer, then k takes 20 in implementing, finally by output corresponding under each scale summation and general
The multi-class class probability value of rate final output.
Step 2.3: the classification thermodynamic chart after width Fusion Features that the classification thermodynamic chart of each characteristic dimension is permeated.Including
Following steps:
Step 2.3.1: by the corresponding classification thermodynamic chart use of all characteristic dimensions be resampled to input image size 513 ×
513;
Step 2.3.2: in each position of characteristic pattern that port number is C to the corresponding classification thermodynamic chart of multiple characteristic dimensions
Value is averaged, the classification thermodynamic chart after obtaining Fusion Features.
Step 3: classification thermodynamic chart and class probability based on the output of depth convolutional neural networks model, using binary picture
As the encirclement frame of target in connected component analysis method output classification thermodynamic chart, object detection results are obtained.Specific step is as follows:
Step 3.1: using the classification results that depth convolutional neural networks model exports as the classification foundation of target, selecting class
There are the corresponding characteristic patterns of class in other thermodynamic chart;Wherein, output category probability value is thought more than or equal to 0.5 with the presence of such,
Otherwise it is assumed that such is not present;
Step 3.2: the response mean value in the corresponding characteristic pattern of each class is calculated, using response mean value as binaryzation threshold
Value, by the corresponding characteristic pattern binaryzation of each class, specifically, the position that characteristic pattern response is more than or equal to binarization threshold is set as
1, otherwise it is set as 0;
Step 3.3: using the connected component analysis method in image procossing, 8- domain pattern being taken (to regard a point week
Enclosing 8 direction points is all field), connected component analysis is carried out to the characteristic pattern after binaryzation, marks each region as corresponding class
Target area;
Step 3.4: using the minimum circumscribed rectangle through each target area marked in the corresponding characteristic pattern of each class as
Corresponding classification target surrounds frame;
Step 3.5: surrounding frame by all in classification thermodynamic chart there are the corresponding target of classification and exported together with classification, complete figure
As object detection task.
Claims (7)
1. a kind of image object detection method based on Weakly supervised study, which comprises the steps of:
Step 1: collecting image data set, rolled up using the depth based on Analysis On Multi-scale Features figure of multi-instance learning method training building
Product neural network model;
Step 2: input real image passes through the classification thermodynamic chart of depth convolutional neural networks model extraction real image;
Step 3: using the encirclement frame of target in binary image connected component analysis method output classification thermodynamic chart, obtaining target
Testing result.
2. a kind of image object detection method based on Weakly supervised study according to claim 1, which is characterized in that described
Step 1 includes the following steps:
Step 1.1: compiling image data, normalized is done to picture size, and more heat coding marks are made to image;
Step 1.2: depth convolutional neural networks model of the building based on Analysis On Multi-scale Features figure, depth convolutional neural networks model
Core network model uses the model through public data collection pre-training;
Step 1.3: using the depth convolutional neural networks model of multi-instance learning method training building, wherein labeled data is used
More heat coding marks in step 1.1.
3. a kind of image object detection method based on Weakly supervised study according to claim 2, which is characterized in that described
Step 1.3 includes the following steps:
Step 1.3.1: exporting predicted value as multi-class probability by more heat coding marks and depth convolutional neural networks model,
To calculate cross entropy loss function;The output characteristic pattern in depth convolutional neural networks model is carried out using bloom loss function
Constraint, forces the response on the output characteristic pattern in convolutional network model to be drawn close toward certainty eminence, and this cross entropy is damaged
Lose the overall loss function of the sum of function and bloom loss function as model;
Step 1.3.2: gradient decline optimization, training depth convolutional Neural are carried out using optimizer to total losses function in training
Network model is until convergence.
4. a kind of image object detection method based on Weakly supervised study according to claim 2, it is characterised in that: described
Step 2 includes the following steps:
Step 2.1: dimension normalization processing is carried out for the real image of input;
Step 2.2: the real image through normalized is obtained into each feature ruler by depth convolutional neural networks model treatment
The classification thermodynamic chart of degree;
Step 2.3: the classification thermodynamic chart after width Fusion Features that the classification thermodynamic chart of each characteristic dimension is permeated.
5. a kind of image object detection method based on Weakly supervised study according to claim 4, which is characterized in that described
Step 2.2 includes the following steps:
Step 2.2.1: the different down-sampling stages in the core network model of depth convolutional neural networks model are chosen, to input
Real image by feature extraction network extract Analysis On Multi-scale Features;
Step 2.2.2: Analysis On Multi-scale Features carry out global multiple dimensioned pond, it is general to export multi-class classification after convolutional layer is converted
Rate value.
6. a kind of image object detection method based on Weakly supervised study according to claim 4, it is characterised in that: described
Step 2.3 includes the following steps:
Step 2.3.1: the corresponding classification thermodynamic chart of all characteristic dimensions is sampled to the real image after normalized
Size;
Step 2.3.2: in each position, the response to the corresponding classification thermodynamic chart of multiple characteristic dimensions is averaged, and obtains feature
Fused classification thermodynamic chart.
7. a kind of image object detection method based on Weakly supervised study according to claim 5, it is characterised in that: described
Step 3 includes the following steps:
Step 3.1: using the classification results that depth convolutional neural networks model exports as the classification foundation of target, selecting classification heat
There are the corresponding characteristic patterns of class in trying hard to;Wherein, with the presence of such, otherwise output category probability value is thought more than or equal to 0.5
Think that such is not present;
Step 3.2: the response mean value in the corresponding characteristic pattern of each class is calculated, using response mean value as binarization threshold,
And by the corresponding characteristic pattern binaryzation of each class;
Step 3.3: using connected component analysis method, take 8- domain pattern, connected region is carried out to the characteristic pattern after binaryzation
Domain analysis marks target area of each region as corresponding class;
Step 3.4: surrounding frame for the minimum circumscribed rectangle of the target area of each class as corresponding classification target;
Step 3.5: will be all there are the corresponding encirclement frame output of classification in classification thermodynamic chart, complete image object detection.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910624287.8A CN110349148A (en) | 2019-07-11 | 2019-07-11 | Image target detection method based on weak supervised learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910624287.8A CN110349148A (en) | 2019-07-11 | 2019-07-11 | Image target detection method based on weak supervised learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110349148A true CN110349148A (en) | 2019-10-18 |
Family
ID=68175727
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910624287.8A Pending CN110349148A (en) | 2019-07-11 | 2019-07-11 | Image target detection method based on weak supervised learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110349148A (en) |
Cited By (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110968695A (en) * | 2019-11-18 | 2020-04-07 | 罗彤 | Intelligent labeling method, device and platform based on active learning of weak supervision technology |
CN111046920A (en) * | 2019-11-22 | 2020-04-21 | 中国科学院计算技术研究所 | Method for training food image classification model and image classification method |
CN111091147A (en) * | 2019-12-10 | 2020-05-01 | 东软集团股份有限公司 | Image classification method, device and equipment |
CN111275688A (en) * | 2020-01-19 | 2020-06-12 | 合肥工业大学 | Small target detection method based on context feature fusion screening of attention mechanism |
CN111340126A (en) * | 2020-03-03 | 2020-06-26 | 腾讯云计算(北京)有限责任公司 | Article identification method and device, computer equipment and storage medium |
CN111461182A (en) * | 2020-03-18 | 2020-07-28 | 北京小米松果电子有限公司 | Image processing method, image processing apparatus, and storage medium |
CN111523585A (en) * | 2020-04-16 | 2020-08-11 | 厦门大学 | Weak supervision target detection method based on improved depth residual error network |
CN111523586A (en) * | 2020-04-16 | 2020-08-11 | 厦门大学 | Noise-aware-based full-network supervision target detection method |
CN111738074A (en) * | 2020-05-18 | 2020-10-02 | 上海交通大学 | Pedestrian attribute identification method, system and device based on weak supervised learning |
CN111738275A (en) * | 2020-06-13 | 2020-10-02 | 玖壹叁陆零医学科技南京有限公司 | Generation method and training method of sampling frame and neural network |
CN111832625A (en) * | 2020-06-18 | 2020-10-27 | 中国医学科学院肿瘤医院 | Full-scan image analysis method and system based on weak supervised learning |
CN111882546A (en) * | 2020-07-30 | 2020-11-03 | 中原工学院 | Weak supervised learning-based three-branch convolutional network fabric defect detection method |
CN111915746A (en) * | 2020-07-16 | 2020-11-10 | 北京理工大学 | Weak-labeling-based three-dimensional point cloud target detection method and labeling tool |
CN111967464A (en) * | 2020-06-30 | 2020-11-20 | 西安电子科技大学 | Weak supervision target positioning method based on deep learning |
CN112215303A (en) * | 2020-11-05 | 2021-01-12 | 北京理工大学 | Image understanding method and system based on self-learning attribute |
CN112465057A (en) * | 2020-12-08 | 2021-03-09 | 中国人民解放军空军工程大学 | Target detection and identification method based on deep convolutional neural network |
CN112508816A (en) * | 2020-12-09 | 2021-03-16 | 中国电子科技集团公司第三研究所 | Infrared image sharpening method, sharpening processing system and terminal device |
CN112560999A (en) * | 2021-02-18 | 2021-03-26 | 成都睿沿科技有限公司 | Target detection model training method and device, electronic equipment and storage medium |
CN112613548A (en) * | 2020-12-17 | 2021-04-06 | 中山大学 | User customized target detection method, system and storage medium based on weak supervised learning |
CN112669282A (en) * | 2020-12-29 | 2021-04-16 | 燕山大学 | Spine positioning method based on deep neural network |
CN112686304A (en) * | 2020-12-29 | 2021-04-20 | 山东大学 | Target detection method and device based on attention mechanism and multi-scale feature fusion and storage medium |
CN112784632A (en) * | 2019-11-06 | 2021-05-11 | 济南信通达电气科技有限公司 | Method and device for detecting potential safety hazard of power transmission line |
WO2021098261A1 (en) * | 2019-11-22 | 2021-05-27 | 苏州浪潮智能科技有限公司 | Target detection method and apparatus |
CN113139587A (en) * | 2021-03-31 | 2021-07-20 | 杭州电子科技大学 | Double-quadratic pooling model for self-adaptive interactive structure learning |
CN113255791A (en) * | 2021-05-31 | 2021-08-13 | 西安电子科技大学 | Small sample target detection method and system based on K-combination mean characteristic enhancement |
CN113420827A (en) * | 2021-07-08 | 2021-09-21 | 上海浦东发展银行股份有限公司 | Semantic segmentation network training and image semantic segmentation method, device and equipment |
CN114998840A (en) * | 2022-07-18 | 2022-09-02 | 成都东方天呈智能科技有限公司 | Mouse target detection method based on deep cascade supervised learning |
WO2023060637A1 (en) * | 2021-10-11 | 2023-04-20 | 深圳硅基智能科技有限公司 | Measurement method and measurement apparatus based on deep learning of tight box mark |
CN116310293A (en) * | 2023-02-13 | 2023-06-23 | 中国矿业大学(北京) | Method for detecting target of generating high-quality candidate frame based on weak supervised learning |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107316007A (en) * | 2017-06-07 | 2017-11-03 | 浙江捷尚视觉科技股份有限公司 | A kind of monitoring image multiclass object detection and recognition methods based on deep learning |
CN108090423A (en) * | 2017-12-01 | 2018-05-29 | 上海工程技术大学 | A kind of depth detection method of license plate returned based on thermodynamic chart and key point |
CN108564097A (en) * | 2017-12-05 | 2018-09-21 | 华南理工大学 | A kind of multiscale target detection method based on depth convolutional neural networks |
CN108764292A (en) * | 2018-04-27 | 2018-11-06 | 北京大学 | Deep learning image object mapping based on Weakly supervised information and localization method |
CN109325385A (en) * | 2017-07-31 | 2019-02-12 | 株式会社理光 | Target detection and region segmentation method, device and computer readable storage medium |
US20190180882A1 (en) * | 2017-12-12 | 2019-06-13 | Electronics And Telecommunications Research Institute | Device and method of processing multi-dimensional time series medical data |
CN109948526A (en) * | 2019-03-18 | 2019-06-28 | 北京市商汤科技开发有限公司 | Image processing method and device, detection device and storage medium |
-
2019
- 2019-07-11 CN CN201910624287.8A patent/CN110349148A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107316007A (en) * | 2017-06-07 | 2017-11-03 | 浙江捷尚视觉科技股份有限公司 | A kind of monitoring image multiclass object detection and recognition methods based on deep learning |
CN109325385A (en) * | 2017-07-31 | 2019-02-12 | 株式会社理光 | Target detection and region segmentation method, device and computer readable storage medium |
CN108090423A (en) * | 2017-12-01 | 2018-05-29 | 上海工程技术大学 | A kind of depth detection method of license plate returned based on thermodynamic chart and key point |
CN108564097A (en) * | 2017-12-05 | 2018-09-21 | 华南理工大学 | A kind of multiscale target detection method based on depth convolutional neural networks |
US20190180882A1 (en) * | 2017-12-12 | 2019-06-13 | Electronics And Telecommunications Research Institute | Device and method of processing multi-dimensional time series medical data |
CN108764292A (en) * | 2018-04-27 | 2018-11-06 | 北京大学 | Deep learning image object mapping based on Weakly supervised information and localization method |
CN109948526A (en) * | 2019-03-18 | 2019-06-28 | 北京市商汤科技开发有限公司 | Image processing method and device, detection device and storage medium |
Non-Patent Citations (4)
Title |
---|
王美华: ""基于神经网络的特定目标检测与识别"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
邓凝旖等: "基于类别先验与深度神经网络特征的显著性检测", 《计算机工程》 * |
颜荔: ""基于卷积神经网络的遥感图像飞机目标识别研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
高健杰: ""基于深度神经网络的图像显著性预测方法研究与实现"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (47)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112784632A (en) * | 2019-11-06 | 2021-05-11 | 济南信通达电气科技有限公司 | Method and device for detecting potential safety hazard of power transmission line |
CN112784632B (en) * | 2019-11-06 | 2023-05-30 | 济南信通达电气科技有限公司 | Method and device for detecting potential safety hazards of power transmission line |
CN110968695A (en) * | 2019-11-18 | 2020-04-07 | 罗彤 | Intelligent labeling method, device and platform based on active learning of weak supervision technology |
CN111046920A (en) * | 2019-11-22 | 2020-04-21 | 中国科学院计算技术研究所 | Method for training food image classification model and image classification method |
WO2021098261A1 (en) * | 2019-11-22 | 2021-05-27 | 苏州浪潮智能科技有限公司 | Target detection method and apparatus |
CN111091147A (en) * | 2019-12-10 | 2020-05-01 | 东软集团股份有限公司 | Image classification method, device and equipment |
CN111275688A (en) * | 2020-01-19 | 2020-06-12 | 合肥工业大学 | Small target detection method based on context feature fusion screening of attention mechanism |
CN111275688B (en) * | 2020-01-19 | 2023-12-12 | 合肥工业大学 | Small target detection method based on context feature fusion screening of attention mechanism |
CN111340126A (en) * | 2020-03-03 | 2020-06-26 | 腾讯云计算(北京)有限责任公司 | Article identification method and device, computer equipment and storage medium |
CN111461182A (en) * | 2020-03-18 | 2020-07-28 | 北京小米松果电子有限公司 | Image processing method, image processing apparatus, and storage medium |
CN111461182B (en) * | 2020-03-18 | 2023-04-18 | 北京小米松果电子有限公司 | Image processing method, image processing apparatus, and storage medium |
CN111523586A (en) * | 2020-04-16 | 2020-08-11 | 厦门大学 | Noise-aware-based full-network supervision target detection method |
CN111523586B (en) * | 2020-04-16 | 2022-09-06 | 厦门大学 | Noise-aware-based full-network supervision target detection method |
CN111523585B (en) * | 2020-04-16 | 2022-05-31 | 厦门大学 | Weak supervision target detection method based on improved depth residual error network |
CN111523585A (en) * | 2020-04-16 | 2020-08-11 | 厦门大学 | Weak supervision target detection method based on improved depth residual error network |
CN111738074A (en) * | 2020-05-18 | 2020-10-02 | 上海交通大学 | Pedestrian attribute identification method, system and device based on weak supervised learning |
CN111738275A (en) * | 2020-06-13 | 2020-10-02 | 玖壹叁陆零医学科技南京有限公司 | Generation method and training method of sampling frame and neural network |
CN111832625A (en) * | 2020-06-18 | 2020-10-27 | 中国医学科学院肿瘤医院 | Full-scan image analysis method and system based on weak supervised learning |
CN111967464A (en) * | 2020-06-30 | 2020-11-20 | 西安电子科技大学 | Weak supervision target positioning method based on deep learning |
CN111967464B (en) * | 2020-06-30 | 2024-02-20 | 西安电子科技大学 | Weak supervision target positioning method based on deep learning |
CN111915746A (en) * | 2020-07-16 | 2020-11-10 | 北京理工大学 | Weak-labeling-based three-dimensional point cloud target detection method and labeling tool |
CN111915746B (en) * | 2020-07-16 | 2022-09-13 | 北京理工大学 | Weak-labeling-based three-dimensional point cloud target detection method and labeling tool |
CN111882546A (en) * | 2020-07-30 | 2020-11-03 | 中原工学院 | Weak supervised learning-based three-branch convolutional network fabric defect detection method |
CN111882546B (en) * | 2020-07-30 | 2023-08-01 | 中原工学院 | Three-branch convolution network fabric defect detection method based on weak supervision learning |
CN112215303A (en) * | 2020-11-05 | 2021-01-12 | 北京理工大学 | Image understanding method and system based on self-learning attribute |
CN112215303B (en) * | 2020-11-05 | 2022-02-11 | 北京理工大学 | Image understanding method and system based on self-learning attribute |
CN112465057A (en) * | 2020-12-08 | 2021-03-09 | 中国人民解放军空军工程大学 | Target detection and identification method based on deep convolutional neural network |
CN112465057B (en) * | 2020-12-08 | 2023-05-12 | 中国人民解放军空军工程大学 | Target detection and identification method based on deep convolutional neural network |
CN112508816B (en) * | 2020-12-09 | 2023-09-08 | 中国电子科技集团公司第三研究所 | Infrared image sharpening method, sharpening processing system and terminal equipment |
CN112508816A (en) * | 2020-12-09 | 2021-03-16 | 中国电子科技集团公司第三研究所 | Infrared image sharpening method, sharpening processing system and terminal device |
CN112613548A (en) * | 2020-12-17 | 2021-04-06 | 中山大学 | User customized target detection method, system and storage medium based on weak supervised learning |
CN112613548B (en) * | 2020-12-17 | 2023-06-13 | 中山大学 | User customized target detection method, system and storage medium based on weak supervised learning |
CN112686304A (en) * | 2020-12-29 | 2021-04-20 | 山东大学 | Target detection method and device based on attention mechanism and multi-scale feature fusion and storage medium |
CN112669282A (en) * | 2020-12-29 | 2021-04-16 | 燕山大学 | Spine positioning method based on deep neural network |
CN112669282B (en) * | 2020-12-29 | 2023-02-14 | 燕山大学 | Spine positioning method based on deep neural network |
CN112686304B (en) * | 2020-12-29 | 2023-03-24 | 山东大学 | Target detection method and device based on attention mechanism and multi-scale feature fusion and storage medium |
CN112560999A (en) * | 2021-02-18 | 2021-03-26 | 成都睿沿科技有限公司 | Target detection model training method and device, electronic equipment and storage medium |
CN112560999B (en) * | 2021-02-18 | 2021-06-04 | 成都睿沿科技有限公司 | Target detection model training method and device, electronic equipment and storage medium |
CN113139587B (en) * | 2021-03-31 | 2024-02-06 | 杭州电子科技大学 | Double secondary pooling model for self-adaptive interactive structure learning |
CN113139587A (en) * | 2021-03-31 | 2021-07-20 | 杭州电子科技大学 | Double-quadratic pooling model for self-adaptive interactive structure learning |
CN113255791B (en) * | 2021-05-31 | 2022-11-25 | 西安电子科技大学 | Small sample target detection method and system based on K-combination mean characteristic enhancement |
CN113255791A (en) * | 2021-05-31 | 2021-08-13 | 西安电子科技大学 | Small sample target detection method and system based on K-combination mean characteristic enhancement |
CN113420827A (en) * | 2021-07-08 | 2021-09-21 | 上海浦东发展银行股份有限公司 | Semantic segmentation network training and image semantic segmentation method, device and equipment |
WO2023060637A1 (en) * | 2021-10-11 | 2023-04-20 | 深圳硅基智能科技有限公司 | Measurement method and measurement apparatus based on deep learning of tight box mark |
CN114998840A (en) * | 2022-07-18 | 2022-09-02 | 成都东方天呈智能科技有限公司 | Mouse target detection method based on deep cascade supervised learning |
CN116310293A (en) * | 2023-02-13 | 2023-06-23 | 中国矿业大学(北京) | Method for detecting target of generating high-quality candidate frame based on weak supervised learning |
CN116310293B (en) * | 2023-02-13 | 2023-09-12 | 中国矿业大学(北京) | Method for detecting target of generating high-quality candidate frame based on weak supervised learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110349148A (en) | Image target detection method based on weak supervised learning | |
CN107563381B (en) | Multi-feature fusion target detection method based on full convolution network | |
CN104992223B (en) | Intensive population estimation method based on deep learning | |
CN109949316A (en) | A kind of Weakly supervised example dividing method of grid equipment image based on RGB-T fusion | |
CN112966684A (en) | Cooperative learning character recognition method under attention mechanism | |
Zheng et al. | HLU 2-Net: a residual U-structure embedded U-Net with hybrid loss for tire defect inspection | |
CN111401353B (en) | Method, device and equipment for identifying mathematical formula | |
CN108629367A (en) | A method of clothes Attribute Recognition precision is enhanced based on depth network | |
CN115424282A (en) | Unstructured text table identification method and system | |
CN112613428B (en) | Resnet-3D convolution cattle video target detection method based on balance loss | |
CN104123545A (en) | Real-time expression feature extraction and identification method | |
CN105608454A (en) | Text structure part detection neural network based text detection method and system | |
CN105117707A (en) | Regional image-based facial expression recognition method | |
CN109753853A (en) | One kind being completed at the same time pedestrian detection and pedestrian knows method for distinguishing again | |
CN112818951A (en) | Ticket identification method | |
CN109472226A (en) | A kind of sleep behavioral value method based on deep learning | |
CN111753653B (en) | High-speed rail contact net fastener identification and positioning method based on attention mechanism | |
CN111507353B (en) | Chinese field detection method and system based on character recognition | |
CN114529821A (en) | Offshore wind power safety monitoring and early warning method based on machine vision | |
CN115546670A (en) | Photovoltaic panel infrared image hot spot detection method based on improved BETR model | |
CN104537392B (en) | A kind of method for checking object based on the semantic part study of identification | |
CN117437647B (en) | Oracle character detection method based on deep learning and computer vision | |
CN114529894A (en) | Rapid scene text detection method fusing hole convolution | |
CN113743389A (en) | Facial expression recognition method and device and electronic equipment | |
CN113642500A (en) | Low-illumination target detection method based on multi-stage domain self-adaption |
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: 20191018 |