CN112348051A - Method and system for reducing sample imbalance influence in target detection - Google Patents
Method and system for reducing sample imbalance influence in target detection Download PDFInfo
- Publication number
- CN112348051A CN112348051A CN202011058077.6A CN202011058077A CN112348051A CN 112348051 A CN112348051 A CN 112348051A CN 202011058077 A CN202011058077 A CN 202011058077A CN 112348051 A CN112348051 A CN 112348051A
- Authority
- CN
- China
- Prior art keywords
- sample
- samples
- target detection
- transformation
- reducing
- 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
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000001514 detection method Methods 0.000 title claims abstract description 39
- 230000009466 transformation Effects 0.000 claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 34
- 230000000694 effects Effects 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 13
- 230000006870 function Effects 0.000 claims description 19
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000003860 storage Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 238000005520 cutting process Methods 0.000 claims description 2
- 238000001914 filtration Methods 0.000 claims description 2
- 230000002194 synthesizing effect Effects 0.000 claims description 2
- 238000006243 chemical reaction Methods 0.000 abstract description 7
- 230000008569 process Effects 0.000 description 7
- 230000015572 biosynthetic process Effects 0.000 description 6
- 238000003786 synthesis reaction Methods 0.000 description 6
- 238000000844 transformation Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 230000000873 masking effect Effects 0.000 description 5
- 230000001131 transforming effect Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012284 sample analysis method Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a system for reducing sample unbalance influence in target detection, belonging to the technical field of image processing and aiming at providing a sample Y0Comprising M samples Y of no interest1And N samples Y of interest2(ii) a M is greater than N; the method comprises the following steps: firstly, the method comprises the following steps: for sample Y2Carrying out transformation; obtaining a sample Y21(ii) a II, secondly: sample Y1And sample Y21Superposing to obtain a sample Y3(ii) a Thirdly, the method comprises the following steps: sample Y3And sample Y0Inputting the data into a learning machine for training; fourthly, the method comprises the following steps: checking whether the training result meets the requirement, if not, carrying out the transformation mode andor the superposition mode andor the sample Y0Is adjusted. The invention adopts various conversion modes to superpose the existing sample on other images without the sample, so that more samples which are generated intentionally are obtained from the classes with few samples, and the number of the samples is enriched.And then the detection effect is observed, the conversion mode or content is adjusted, and the purpose of reducing the unbalanced influence of the sample is achieved.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a method and a system for reducing sample unbalance influence in target detection.
Background
As is well known, image processing (image processing) refers to: techniques for analyzing images with a computer to achieve a desired result. Also known as image processing. Image processing generally refers to digital image processing. Digital images are large two-dimensional arrays of elements called pixels and values called gray-scale values, which are captured by industrial cameras, video cameras, scanners, etc. Image processing techniques generally include image compression, enhancement and restoration, matching, description and identification of 3 parts.
In recent years, with the rapid development of image processing technology and intelligent technology, people often need to analyze and process a large number of pictures to acquire required information, as shown in fig. 1: in image processing, the above-mentioned large number of pictures are generally defined as original samples; among these original samples, a large number of samples Y of no interest are included1And a small number of samples Y of interest2(ii) a The number of differentiated samples is the sample imbalance; at present, in the existing sample analysis method, due to unbalanced samples, some targets with a large number of samples are easy to detect, while targets with a small number of samples are often unobvious due to less attention obtained in training, and specifically, due to lack of necessary information, targets with a small number of samples are missed in an inspection chart and are falsely reported as targets with a large number of other samples, or an area which should not be a target originally is falsely reported as the target, targets are mixed among a plurality of samples with a small number of samples, and various undesirable phenomena such as errors in position and size when the target is reported are caused. Thus, the detection result is poor; the efficiency of learning training is reduced.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method and a system for reducing sample unbalance influence in target detection, and the method and the system superpose the existing sample on other images without the sample through various conversion modes, so that more samples which are intentionally generated are obtained from the classes with few samples, and the number of the samples is enriched. And then the detection effect is observed, the conversion mode or content is adjusted, and the purpose of reducing the unbalanced influence of the sample is achieved.
One of the objectives of the present invention is to provide a method for reducing the influence of sample imbalance in target detection, comprising the following steps:
sample Y0Comprising M samples Y of no interest1And N samples Y of interest2(ii) a M is greater than N; the method comprises the following steps:
the method comprises the following steps: for sample Y2Carrying out transformation; obtaining a sample Y21；
Step two: sample Y1And sample Y21Superposing to obtain a sample Y3；
Step three: sample Y3And sample Y0Inputting the data into a learning machine for training;
step four: checking whether the training result meets the requirement, if not, carrying out the transformation mode andor the superposition mode andor the sample Y0Is adjusted.
Preferably, during the superposition, the sample Y is subjected to1And (6) carrying out transformation.
Preferably, the transformation comprises one or more of image scaling, image flipping, image warping, image blurring, image coloring, changing transparency, increasing noise, adjusting contrast, adjusting brightness.
Preferably, the superimposing comprises: one or more of covering, synthesizing, color gamut transformation and superposition, superposition after filtering and superposition after cutting.
Preferably, the third step is trained by using a deep artificial neural network.
Preferably, the deep artificial neural network comprises a function f (X), the input of the function f (X) is an image part X of the sample, the output is a label part Y of the sample, each training finishes inputting the images of all the samples and obtains a predicted label
F (X) is optimized, and the optimization aim is to minimize the Loss function;
the Loss function is described by the following formula:
wherein: alpha is alphacIs a factor used to identify the weight of a class of samples, the samples of different classes using different alpha' sc，αcThe larger the class sample number, the smaller the weight needs to be increased;
m is the enumeration number of all target classes to be identified in the target detection task;
c is the target class where the single sample is located;
γcis a pair of (1-p)y,c) The greater γ indicates that the class is of greater interest;
ycis a category label value;
py,cpredicting for F (X)
And then, the probability of belonging to class c.
The deep artificial neural network can be regarded as a function F (X), the input of the function is an image part X of a sample, the output of the function is a label part Y of the sample, and each time X is trained (namely, the image of all samples is input and the predicted label is obtained
) F (X) is optimized, and the optimization aims to minimize the Loss function. (note that the ratio of X, Y,
are all sets rather than single data)
The Loss function is described by the following equation,
wherein: alpha is alphacIs a factor used to identify the sample class weight. Different classes of samples use different alphac，αcLarger samples indicate smaller samples of this type, and more weight needs to be increased.
And m is the enumeration number of all target classes to be identified in the target detection task.
c is the target class in which a single sample is located.
γcIs a pair of (1-p)y,c) A larger γ indicates that the class is more interesting.
ycIs a category label value.
py,cPredicting for F (X)
And then, the probability of belonging to class c.
Another object of the present invention is to provide a system for reducing the influence of sample imbalance in target detection, which at least comprises:
sample Y0Comprising M samples Y of no interest1And N samples Y of interest2(ii) a M is greater than N; the method is characterized in that: at least comprises the following steps:
a transformation module: for sample Y2Carrying out transformation; obtaining a sample Y21；
A superposition module: sample Y1And sample Y21Superposing to obtain a sample Y3；
A learning module: sample Y3And sample Y0Inputting the data into a learning machine for training;
a judging module: checking whether the training result meets the requirement, if not, carrying out the transformation mode andor the superposition mode andor the sample Y0Is adjusted.
It is a further object of the present invention to provide a computer program for implementing a method for reducing the effect of sample imbalance in target detection.
It is a fourth object of the present invention to provide an information data processing terminal that implements a method for reducing the influence of sample imbalance in target detection.
It is a further object of the present invention to provide a computer-readable storage medium, comprising instructions which, when executed on a computer, cause the computer to perform a method for reducing the effect of sample imbalance in target detection.
In summary, the advantages and positive effects of the invention are:
by using the technical scheme of the invention, the targets with few samples are also paid enough attention in the training process, and the effect is obviously better than that of the original scheme. Meanwhile, the scheme adds a step of rechecking and adjusting after training, so that the target with few samples is used for ensuring that the attention obtained in the training is beneficial to the detection effect, the adverse effect is further reduced, and the training efficiency is improved.
According to the invention, a small amount of concerned samples are subjected to multi-form conversion, so that the number of the small amount of concerned samples is increased, the ratio of the concerned samples is improved, and the problem of sample imbalance is further improved.
The invention adopts a mode of superposition output of various conversion modes, and improves the accuracy of the result of target detection.
Drawings
FIG. 1 is a flow chart of a conventional solution;
FIG. 2 is a flow chart of a preferred embodiment of the present invention;
fig. 3 is a block diagram of the preferred embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Referring to FIG. 2, a method for reducing the effect of sample imbalance in target detection is shown for a conventional data set (sample Y)0Comprising M samples Y of no interest1And N samples Y of interest2(ii) a M is greater than N; in general: m is far larger than N), the existing sample is superposed on other images without the sample through various transformation modes, so that more intentionally generated samples are obtained from classes with few samples, and the number of the samples is enriched. And then the detection effect is observed, the conversion mode or content is adjusted, and the purpose of reducing the unbalanced influence of the sample is achieved. The method specifically comprises the following steps:
firstly, transforming a small amount of concerned samples in various forms;
step two, superposing the transformed sample and a large amount of samples which are not concerned in various forms;
thirdly, the superposed samples and original samples (including samples without attention and samples with attention) are taken as input of a target detection training method and device to carry out machine learning or deep learning (deep artificial neural network) training;
the deep artificial neural network can be regarded as a function F (X), the input of the function is an image part X of a sample, the output of the function is a label part Y of the sample, and each time X is trained (namely, the image of all samples is input and the predicted label is obtained
) F (X) is optimized, and the optimization aims to minimize the Loss function. (note that the ratio of X, Y,
are all sets rather than single data)
The Loss function is described by the following equation,
wherein: alpha is alphacIs a factor used to identify the sample class weight. Different classes of samples use different alphac，αcLarger samples indicate smaller samples of this type, and more weight needs to be increased.
And m is the enumeration number of all target classes to be identified in the target detection task.
c is the target class in which a single sample is located.
γcIs a pair of (1-p)y,c) A larger γ indicates that the class is more interesting.
ycIs a category label value.
py,cPredicting for F (X)
And then, the probability of belonging to class c.
Step four, checking whether the result after training meets the requirement, if not, carrying out necessary adjustment on the sample synthesis mode, and the method comprises the following steps: adjusting the transformation form, adjusting the superposition mode and adjusting the category and the number of the superposed samples.
In the preferred embodiment described above: the "multi-form transformation of a small amount of samples of interest" includes, but is not limited to, scaling, flipping, warping, blurring, changing color scheme, changing transparency, increasing noise, adjusting contrast, adjusting brightness, and other ways and combinations thereof for increasing the number of different pictures of the samples of interest, and the transformation rules thereof include: 1) and (4) randomly changing by adopting a random algorithm or in a specified range according to an application scene. 2) Enumerating the traversal in a full application scenario or enumerating the traversal within a specified range.
In the preferred embodiment described above: the "multi-form superposition" includes, but is not limited to, random masking, synthesis, color gamut transform superposition, post-filter superposition, and post-cropping superposition for the purpose of increasing the number of different pictures for a sample of interest, and the transformation of the superposition type, number, and position is also included in the meaning. The transformation rule comprises the following steps: 1) and (4) randomly changing by adopting a random algorithm or in a specified range according to an application scene. 2) Enumerating the traversal in a full application scenario or enumerating the traversal within a specified range.
In the process of overlaying, for the purpose of increasing the number of different pictures for the samples of interest, the samples not of interest are subjected to various transformations, and then overlaid with the samples of interest, in the scheme of the invention, the transformations include, but are not limited to, scaling, flipping, warping, blurring, changing color scheme, changing transparency, increasing noise, adjusting contrast, adjusting brightness, and various ways and methods, and their combined applications. The transformation rule comprises the following steps: 1) and (4) randomly changing by adopting a random algorithm or in a specified range according to an application scene. 2) Enumerating the traversal in a full application scenario or enumerating the traversal within a specified range.
Referring to fig. 3, a system for reducing the effect of sample imbalance in target detection according to a second preferred embodiment includes:
the transformation module is used for carrying out transformation in various forms on a small amount of samples which are concerned;
the superposition module is used for superposing the transformed sample and a large amount of samples which are not concerned with in various forms;
a learning module, which takes the superposed sample and original sample (including sample without attention and sample with attention) as the input of the target detection training method and device to perform machine learning or deep learning (deep artificial neural network) training;
judging whether the result meets the requirements after the module and the test training, if not, carrying out necessary adjustment on the sample synthesis mode, and the method comprises the following steps: adjusting the transformation form, adjusting the superposition mode and adjusting the category and the number of the superposed samples.
In the preferred embodiment described above: the "multi-form transformation of a small number of samples of interest" includes, but is not limited to, scaling, flipping, warping, blurring, changing color schemes, changing transparency, adding noise, adjusting contrast, adjusting brightness, and combinations thereof.
In the preferred embodiment described above: the "multi-form superposition" includes but is not limited to masking, composition, color gamut transform superposition, filtered superposition, and clipped superposition, and the transformation of the superposition type, number, and position is also included in the meaning.
In the process of overlaying, various transformations are performed on the samples not receiving attention originally for the purpose of increasing the number of different pictures of the samples receiving attention, and the transformation is also included in the inventive solution, including but not limited to scaling, flipping, warping, blurring, changing color scheme, changing transparency, increasing noise, adjusting contrast, adjusting brightness, and various ways and methods and their combined applications. The transformation rule comprises the following steps: 1) and (4) randomly changing by adopting a random algorithm or in a specified range according to an application scene. 2) Enumerating the traversal in a full application scenario or enumerating the traversal within a specified range.
In a third preferred embodiment, a computer program for implementing a method for reducing the influence of sample imbalance in target detection includes the following steps:
firstly, transforming a small amount of concerned samples in various forms;
step two, superposing the transformed sample and a large amount of samples which are not concerned in various forms;
thirdly, the superposed samples and original samples (including samples without attention and samples with attention) are taken as input of a target detection training method and device to carry out machine learning or deep learning (deep artificial neural network) training;
step four, checking whether the result after training meets the requirement, if not, carrying out necessary adjustment on the sample synthesis mode, and the method comprises the following steps: adjusting the transformation form, adjusting the superposition mode and adjusting the category and the number of the superposed samples.
In the preferred embodiment described above: the "multi-form transformation of a small number of samples of interest" includes, but is not limited to, scaling, flipping, warping, blurring, changing color schemes, changing transparency, adding noise, adjusting contrast, adjusting brightness, and combinations thereof.
In the preferred embodiment described above: the "multi-form superposition" includes but is not limited to masking, composition, color gamut transform superposition, filtered superposition, and clipped superposition, and the transformation of the superposition type, number, and position is also included in the meaning.
In the process of overlaying, various transformations are performed on the samples not receiving attention originally for the purpose of increasing the number of different pictures of the samples receiving attention, and the transformation is also included in the inventive solution, including but not limited to scaling, flipping, warping, blurring, changing color scheme, changing transparency, increasing noise, adjusting contrast, adjusting brightness, and various ways and methods and their combined applications. The transformation rule comprises the following steps: 1) and (4) randomly changing by adopting a random algorithm or in a specified range according to an application scene. 2) Enumerating the traversal in a full application scenario or enumerating the traversal within a specified range.
In the fourth preferred embodiment, an information data processing terminal for implementing the method for reducing the sample imbalance influence in target detection is provided. The method for reducing the unbalanced influence of the sample in the target detection comprises the following steps:
firstly, transforming a small amount of concerned samples in various forms;
step two, superposing the transformed sample and a large amount of samples which are not concerned in various forms;
thirdly, the superposed samples and original samples (including samples without attention and samples with attention) are taken as input of a target detection training method and device to carry out machine learning or deep learning (deep artificial neural network) training;
step four, checking whether the result after training meets the requirement, if not, carrying out necessary adjustment on the sample synthesis mode, and the method comprises the following steps: adjusting the transformation form, adjusting the superposition mode and adjusting the category and the number of the superposed samples.
In the preferred embodiment described above: the "multi-form transformation of a small number of samples of interest" includes, but is not limited to, scaling, flipping, warping, blurring, changing color schemes, changing transparency, adding noise, adjusting contrast, adjusting brightness, and combinations thereof.
In the preferred embodiment described above: the "multi-form superposition" includes but is not limited to masking, composition, color gamut transform superposition, filtered superposition, and clipped superposition, and the transformation of the superposition type, number, and position is also included in the meaning.
In the process of overlaying, various transformations are performed on the samples not receiving attention originally for the purpose of increasing the number of different pictures of the samples receiving attention, and the transformation is also included in the inventive solution, including but not limited to scaling, flipping, warping, blurring, changing color scheme, changing transparency, increasing noise, adjusting contrast, adjusting brightness, and various ways and methods and their combined applications. The transformation rule comprises the following steps: 1) and (4) randomly changing by adopting a random algorithm or in a specified range according to an application scene. 2) Enumerating the traversal in a full application scenario or enumerating the traversal within a specified range.
A computer-readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the following method of reducing the effect of sample imbalance in target detection:
firstly, transforming a small amount of concerned samples in various forms;
step two, superposing the transformed sample and a large amount of samples which are not concerned in various forms;
thirdly, the superposed samples and original samples (including samples without attention and samples with attention) are taken as input of a target detection training method and device to carry out machine learning or deep learning (deep artificial neural network) training;
step four, checking whether the result after training meets the requirement, if not, carrying out necessary adjustment on the sample synthesis mode, and the method comprises the following steps: adjusting the transformation form, adjusting the superposition mode and adjusting the category and the number of the superposed samples.
In the preferred embodiment described above: the "multi-form transformation of a small number of samples of interest" includes, but is not limited to, scaling, flipping, warping, blurring, changing color schemes, changing transparency, adding noise, adjusting contrast, adjusting brightness, and combinations thereof.
In the preferred embodiment described above: the "multi-form superposition" includes but is not limited to masking, composition, color gamut transform superposition, filtered superposition, and clipped superposition, and the transformation of the superposition type, number, and position is also included in the meaning.
In the process of overlaying, various transformations are performed on the samples not receiving attention originally for the purpose of increasing the number of different pictures of the samples receiving attention, and the transformation is also included in the inventive solution, including but not limited to scaling, flipping, warping, blurring, changing color scheme, changing transparency, increasing noise, adjusting contrast, adjusting brightness, and various ways and methods and their combined applications. The transformation rule comprises the following steps: 1) and (4) randomly changing by adopting a random algorithm or in a specified range according to an application scene. 2) Enumerating the traversal in a full application scenario or enumerating the traversal within a specified range.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When used in whole or in part, can be implemented in a computer program product that includes one or more computer instructions. When loaded or executed on a computer, cause the flow or functions according to embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (10)
1. Method for reducing sample unbalance influence in target detection, sample Y0Comprising M samples Y of no interest1And N samples Y of interest2(ii) a M is greater than N; the method is characterized in that: at least comprises the following steps:
the method comprises the following steps: for sample Y2Carrying out transformation; obtaining a sample Y21；
Step two: sample Y2And sample Y21Superposing to obtain a sample Y3；
Step three: sample Y3And sample Y0Inputting the data into a learning machine for training;
step four: checking whether the training result meets the requirement, if not, carrying out the transformation mode andor the superposition mode andor the sample Y0Is adjusted.
2. The method of claim 1, wherein the sample Y is processed in an overlay process1And (6) carrying out transformation.
3. The method for reducing the sample imbalance effect in target detection according to claim 1 or 2, wherein the transformation includes one or more of image scaling, image flipping, image warping, image blurring, image color matching, transparency change, noise increase, contrast adjustment, and brightness adjustment.
4. The method for reducing the sample imbalance effect in target detection according to claim 1, wherein the superimposing includes: one or more of covering, synthesizing, color gamut transformation and superposition, superposition after filtering and superposition after cutting.
5. The method for reducing the sample imbalance effect in target detection according to claim 1, wherein the third step is training by using a deep artificial neural network.
6. The method as claimed in claim 1, wherein the deep artificial neural network includes a function f (X), the input of the function f (X) is an image portion X of the sample, the output is a label portion Y of the sample, each training completes the image input of all samples and obtains a predicted label
F (X) is optimized, and the optimization aim is to minimize the Loss function;
the Loss function is described by the following formula:
wherein: alpha is alphacIs a factor used to identify the weight of a class of samples, the samples of different classes using different alpha' sc，αcThe larger the class sample number, the smaller the weight needs to be increased;
m is the enumeration number of all target classes to be identified in the target detection task;
c is the target class where the single sample is located;
γcis a pair of (1-p)y,c) The greater γ indicates that the class is of greater interest;
ycis a category label value;
py,cpredicting for F (X)
And then, the probability of belonging to class c.
7. System for reducing sample imbalance influence in target detection, sample Y0Comprising M samples Y of no interest1And N samples Y of interest2(ii) a M is greater than N; the method is characterized in that: at least comprises the following steps:
a transformation module: for sample Y2Carrying out transformation; obtaining a sample Y21；
A superposition module: sample Y2And sample Y21Superposing to obtain a sample Y3；
A learning module: sample Y3And sample Y0Inputting the data into a learning machine for training;
a judging module: checking whether the training result meets the requirement, if not, carrying out the transformation mode andor the superposition mode andor the sample Y0Is adjusted.
8. A computer program for implementing the method of any one of claims 1 to 6 for reducing the effect of sample imbalance in target detection.
9. An information data processing terminal implementing the method for reducing the influence of sample imbalance in target detection according to any one of claims 1 to 6.
10. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method of reducing the effect of sample imbalance in target detection of any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011058077.6A CN112348051A (en) | 2020-09-30 | 2020-09-30 | Method and system for reducing sample imbalance influence in target detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011058077.6A CN112348051A (en) | 2020-09-30 | 2020-09-30 | Method and system for reducing sample imbalance influence in target detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112348051A true CN112348051A (en) | 2021-02-09 |
Family
ID=74360499
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011058077.6A Pending CN112348051A (en) | 2020-09-30 | 2020-09-30 | Method and system for reducing sample imbalance influence in target detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112348051A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113392757A (en) * | 2021-06-11 | 2021-09-14 | 恒睿(重庆)人工智能技术研究院有限公司 | Method, device and medium for training human body detection model by using unbalanced data |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109214413A (en) * | 2018-07-16 | 2019-01-15 | 上海斐讯数据通信技术有限公司 | A kind of determination method and system of data distribution balance |
| CN111079841A (en) * | 2019-12-17 | 2020-04-28 | 深圳奇迹智慧网络有限公司 | Training method and device for target recognition, computer equipment and storage medium |
-
2020
- 2020-09-30 CN CN202011058077.6A patent/CN112348051A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109214413A (en) * | 2018-07-16 | 2019-01-15 | 上海斐讯数据通信技术有限公司 | A kind of determination method and system of data distribution balance |
| CN111079841A (en) * | 2019-12-17 | 2020-04-28 | 深圳奇迹智慧网络有限公司 | Training method and device for target recognition, computer equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| CHASER: "A survey on Image Data Augmentation 数据增强文献综述", pages 1 - 6, Retrieved from the Internet <URL:https://zhuanlan.zhihu.com/p/76044027> * |
| TSUNG-YI LIN ET AL.: "Focal Loss for Dense Object Detection", 《ARXIV》, pages 1 - 10 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113392757A (en) * | 2021-06-11 | 2021-09-14 | 恒睿(重庆)人工智能技术研究院有限公司 | Method, device and medium for training human body detection model by using unbalanced data |
| CN113392757B (en) * | 2021-06-11 | 2023-08-15 | 恒睿(重庆)人工智能技术研究院有限公司 | Method, device and medium for training human body detection model by using unbalanced data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| RU2762144C1 (en) | Apparatus and method for image processing and system for training a neural network | |
| Zhang et al. | Single image defogging based on multi-channel convolutional MSRCR | |
| Park et al. | Double JPEG detection in mixed JPEG quality factors using deep convolutional neural network | |
| Shen et al. | Hybrid no-reference natural image quality assessment of noisy, blurry, JPEG2000, and JPEG images | |
| CN109801215B (en) | Infrared super-resolution imaging method based on countermeasure generation network | |
| CN110675328A (en) | Low-illumination image enhancement method and device based on condition generation countermeasure network | |
| Su et al. | Koniq++: Boosting no-reference image quality assessment in the wild by jointly predicting image quality and defects | |
| WO2023005818A1 (en) | Noise image generation method and apparatus, electronic device, and storage medium | |
| An et al. | HFM: A hybrid fusion method for underwater image enhancement | |
| JP2021189527A (en) | Information processing device, information processing method, and program | |
| CN112348051A (en) | Method and system for reducing sample imbalance influence in target detection | |
| CN117557562B (en) | Image tampering detection method and system based on double-flow network | |
| CN114998261A (en) | Double-current U-Net image tampering detection network system and image tampering detection method thereof | |
| Li et al. | Joint learning of motion deblurring and defocus deblurring networks with a real-world dataset | |
| CN111027573B (en) | Image authenticity identification method based on blind evidence obtaining technology | |
| CN117876233A (en) | Mapping image enhancement method based on unmanned aerial vehicle remote sensing technology | |
| CN116645513A (en) | Watermark extraction method, model training method, device, electronic equipment and medium | |
| US20140193085A1 (en) | Image manipulating system and method | |
| CN116883371A (en) | Flexible printed circuit board appearance defect detection method | |
| Lian et al. | [Retracted] Film and Television Animation Sensing and Visual Image by Computer Digital Image Technology | |
| WO2022113367A1 (en) | Image conversion method, program, and image processing device | |
| Qu et al. | A framework for identifying shifted double JPEG compression artifacts with application to non-intrusive digital image forensics | |
| CN112329497A (en) | Target identification method, device and equipment | |
| CN113947523B (en) | Method and device for replacing background image | |
| Wu et al. | Underwater image restoration with multi-scale shallow feature extraction and detail enhancement network |
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 |