CN113989541A - Dressing classification method and system based on feature aggregation - Google Patents
Dressing classification method and system based on feature aggregation Download PDFInfo
- Publication number
- CN113989541A CN113989541A CN202111112584.8A CN202111112584A CN113989541A CN 113989541 A CN113989541 A CN 113989541A CN 202111112584 A CN202111112584 A CN 202111112584A CN 113989541 A CN113989541 A CN 113989541A
- Authority
- CN
- China
- Prior art keywords
- layer
- features
- feature
- dressing
- classification model
- 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 48
- 230000002776 aggregation Effects 0.000 title claims abstract description 29
- 238000004220 aggregation Methods 0.000 title claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000013145 classification model Methods 0.000 claims abstract description 29
- 230000009466 transformation Effects 0.000 claims abstract description 22
- 238000000605 extraction Methods 0.000 claims abstract description 18
- 230000008569 process Effects 0.000 claims abstract description 10
- 230000004931 aggregating effect Effects 0.000 claims abstract description 6
- 238000013507 mapping Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 23
- 210000002569 neuron Anatomy 0.000 claims description 6
- 210000004205 output neuron Anatomy 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 6
- 238000003860 storage Methods 0.000 claims description 5
- 230000006835 compression Effects 0.000 claims description 3
- 238000007906 compression Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 3
- 239000010410 layer Substances 0.000 description 77
- 238000001514 detection method Methods 0.000 description 10
- 230000004927 fusion Effects 0.000 description 5
- 239000003086 colorant Substances 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000006116 polymerization reaction Methods 0.000 description 3
- 230000001131 transforming effect Effects 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000002344 surface layer 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a dressing classification method and system based on feature aggregation, which relate to the technical field of image processing and comprise the following steps: carrying out multi-layer feature extraction on the image training set based on a pre-constructed classification model; performing channel dimension transformation on the extracted features of each layer, and compressing the feature dimensions to be consistent with the number of categories in the image training set to obtain compressed features; performing cosine transformation on the compressed features of each layer, mapping the features to an angle space, aggregating the features, and constructing a loss function after the feature aggregation; and training the classification model according to the loss function, and dressing the image to be recognized according to the trained classification model. By means of weighting and fusing coefficients, the color information of a shallow layer and the contour information of a deep layer are fully utilized, meanwhile, feature aggregation operation is added to map feature information to an angle space, a maximized classification boundary is obtained, and the similarity difference in the dressing matching process is increased.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a dressing classification method and system based on feature aggregation.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
At present, in the service industry, the standard dressing of related workers such as service windows, enterprise production, construction sites and the like is restrained, and the standardized dressing not only can improve the service image and enhance the production efficiency of enterprises, but also can reduce the safety production accidents. The existing dressing detection method based on the deep learning algorithm mainly adopts a classification network to classify targets or carry out similarity comparison with standard dressing so as to realize the dressing detection function. The method has the advantages that the classification network training is simple, and the dressing detection process is simple and quick.
The core of the classification model is to find the optimal classification boundary among the final sample characteristics of the convolutional network, so as to achieve the target classification effect. As the distribution of the convolution layer number deepens, the surface layer characteristics of the sample such as color, outline, texture and the like are continuously reduced, and the deep abstract characteristics are continuously increased. Due to the particularity of the tool sample, namely, the dressing types with similar colors exist, the classification boundary is not obvious enough, the similarity difference in the standard dressing matching process is small, and the detection result is influenced, so that the characteristics of the dressing sample need deep abstract characteristics and also need light characteristics such as colors, outlines, textures and the like.
Chinese patent CN114183472A discloses a method for detecting whether a worker wears a work clothes based on improved RetinaNet, which comprises the steps of extracting multi-scale features of an image by using MobileNet with a classification layer removed, obtaining three multi-scale feature maps of the image, and classifying clothing types by using Resnet18 to realize a clothing detection function. The method essentially adopts a classification network to realize the dressing type classification function, and no solution is provided when the dressing discrimination is reduced due to environmental change aiming at a specific scene.
Chinese patent CN110059674A discloses a standard dressing detection method based on deep learning, which adds related categories such as masks and hats in training samples by improving the setting parameters of anchor points in a target detection algorithm, completes target classification while realizing target detection, and further judges whether to dress as required. The method has simple network structure and high detection speed, but does not provide a solution on how to improve the degree of similar dressing areas.
Chinese patent CN111401418A discloses a method for detecting employee dressing specification based on improved Faster r-cnn, which is used for collecting, labeling and enhancing data of sample data sets aiming at different application scenes; establishing an improved Faster r-cnn network model; training the improved network by using the enhanced training sample set; detecting the test sample set by using the trained network model; and analyzing whether the detection result meets the predefined dressing specification. Although the diversity of the samples is increased by using a data enhancement method, no improvement method is proposed on the inter-sample-class difference.
Disclosure of Invention
In order to solve the problems, the invention provides a clothing classification method and a clothing classification system based on feature aggregation.
In order to achieve the purpose, the invention adopts the following technical scheme:
in a first aspect, the present invention provides a clothing classification method based on feature aggregation, including:
carrying out multi-layer feature extraction on the image training set based on a pre-constructed classification model;
performing channel dimension transformation on the extracted features of each layer, and compressing the feature dimensions to be consistent with the number of categories in the image training set to obtain compressed features;
performing cosine transformation on the compressed features of each layer, mapping the features to an angle space, aggregating the features, and constructing a loss function after the feature aggregation;
and training the classification model according to the loss function, and dressing the image to be recognized according to the trained classification model.
As an alternative embodiment, the extracted multi-layer features are embodied as a light color feature, a light texture feature, a deep contour feature and a deep abstract feature.
As an alternative implementation, the multi-layer feature extraction process is that except the last layer of feature layer, the output of each layer is subjected to feature extraction through convolution operation and maximum mean operation in sequence; and taking the output of more than one characteristic layer as input, and directly carrying out maximum mean value operation.
As an alternative embodiment, the channel dimension transformation adds a linear link layer to the classification model, and compresses the feature dimension according to the weight coefficient of the linear link layer.
wherein Linear () represents a Linear chaining operation, WclsIs a weight coefficient of a linear link layer whose dimension is Wcls∈R2048,nN is the number of classes in the image training set, and F is the feature extracted from each layer.
As an alternative embodiment, the cosine transform is performed on the compressed features of each layer to obtain: fcls=||Wcls||||F||cosθ;
Wherein, FclsFor features after compression, WclsThe weight coefficient of the linear link layer is shown, n is the number of categories in the image training set, F is the feature extracted from each layer, and theta is the cosine included angle.
As an alternative embodiment, the loss function for each layer is:
wherein the content of the first and second substances,is a linear link layer weight coefficient WclsThe feature outputs a link weight to the ith output neuron; f. ofiIs the output of the ith neuron in the linear link layer; thetaiIs a cosine included angle; m is an interval parameter; s is a scaling factor.
As an alternative embodiment, in the process of mapping and aggregating the features to the angle space, the features are mapped to a hypersphere on the angle space, the radius of the hypersphere is controlled by a scaling factor s, the feature polymerization degree is controlled by an interval parameter m, and as m increases, the difference between the features is higher, and the feature polymerization degree is higher.
In a second aspect, the present invention provides a clothing classification system based on feature aggregation, including:
the characteristic extraction module is configured to perform multi-layer characteristic extraction on the image training set based on a pre-constructed classification model;
the dimension transformation module is configured to perform channel dimension transformation on the extracted features of each layer, and compress the feature dimensions to be consistent with the number of categories in the image training set to obtain compressed features;
the characteristic aggregation module is configured to perform cosine transform on the compressed characteristics of each layer, map the characteristics to an angle space, aggregate the characteristics, and construct a loss function after the characteristics are aggregated;
and the dressing classification module is configured to train the classification model according to the loss function, and the dressing image to be recognized is subjected to dressing classification according to the trained classification model.
In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored on the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method of the first aspect is performed.
In a fourth aspect, the present invention provides a computer readable storage medium for storing computer instructions which, when executed by a processor, perform the method of the first aspect.
Compared with the prior art, the invention has the beneficial effects that:
according to the dressing classification method and system based on feature aggregation, when feature extraction is carried out on an image training set, features with different dimensionalities are output by each feature extraction layer, when channel conversion is carried out, linear link layers are added, output feature dimensionalities are compressed to the number of categories through weight coefficients of the linear link layers, and through a mode of weighting fusion coefficients, color information output by a shallow layer and profile information of a deep layer are fully utilized on the basis of deep abstract features, the weighting fusion coefficients are continuously optimized, and an optimal fusion proportion is obtained.
According to the dressing classification method and system based on feature aggregation, feature aggregation operation is added, the similarity difference in the dressing matching process is increased, the feature information is mapped to the angle space, the maximum classification boundary is obtained, the difference among different types of samples is improved, and the distinguishing degree among different types of clothes is further improved.
Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.
Fig. 1 is a flowchart of a clothing classification method based on feature aggregation according to embodiment 1 of the present invention.
Detailed Description
The invention is further described with reference to the following figures and examples.
It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention. As used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, and it should be understood that the terms "comprises" and "comprising", and any variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The embodiments and features of the embodiments of the present invention may be combined with each other without conflict.
Example 1
As shown in fig. 1, the present embodiment provides a clothing classification method based on feature aggregation to enhance the difference between different color clothing types, which specifically includes:
s1: carrying out multi-layer feature extraction on the image training set based on a pre-constructed classification model;
s2: performing channel dimension transformation on the extracted features of each layer, and compressing the feature dimensions to be consistent with the number of categories in the image training set to obtain compressed features;
s3: performing cosine transformation on the compressed features of each layer, mapping the features to an angle space, aggregating the features, and constructing a loss function after the feature aggregation;
s4: and training the classification model according to the loss function, and obtaining a dressing classification result of the dressing image to be recognized according to the trained classification model.
In this embodiment, in step S1, the pre-constructed classification model uses Resnet50 as a basic network, and includes 4 layers of feature extraction layers, and totally outputs 4 layers of features with different dimensions, specifically, a light color feature, a light texture feature, a deep contour feature, and a deep abstract feature.
In this embodiment, the images in the image training set are samples with 3 channels, 256 pixels in width and 128 pixels in height, and output features of different dimensions after sequentially passing through 4 layers of feature extraction layers.
In step S2, the process of performing channel dimension transformation on the features extracted from each layer specifically includes:
s21-1: the output of the first Layer1 is L1Dimension of L1∈R1,256,68,32The input of the layer is an image training set, and after convolution operation is performed on the images in the image training set, the following results are obtained:
L1_conv=Conv(L1);
wherein L is1-convTo use a convolution kernel with Conv () of 1 × 1, input 256-dimensional and output 2048-dimensional convolution operations, output L1_conv∈R1,2048,68,32。
S21-2: to L1_convAnd performing an average operation to obtain the output characteristics of the layer:
F1=Avg(L1_conv)
wherein, F1Is the output characteristic of Layer1 and has the dimension of F1∈R1,2048Avg () is the maximum mean operation.
S21-3: the operation of performing channel dimension transformation on the output features of the Layer1 Layer specifically includes:
adding a linear link layer in the classification model, compressing the dimension of the output feature to the number of categories, wherein the compressed feature is expressed as:
wherein Linear () represents a Linear chaining operation, Wcls1Is a weight coefficient of a linear link layer whose dimension is Wcls1∈R2048,nAnd n is the number of classes in the image training set.
S22-1: the output of the second Layer2 is L2Dimension of L2∈R1,512,33,16The input of the Layer is the output L of Layer11After convolution operation, we get:
L2_conv=Conv(L2)
wherein L is2_convAdopting Conv () as 1 × 1 convolution kernel, inputting 512-dimensional convolution operation and outputting 2048-dimensional convolution operation, then outputting L2_conv∈R1,2048,33,16。
S22-2: to L2_convAnd performing an average operation to obtain the output characteristics of the layer:
F2=Avg(L2_conv)
wherein, F2Is the output characteristic of Layer2 and has the dimension of F2∈R1,2048Avg () is the maximum mean operation.
S22-3: the operation of performing channel dimension transformation on the output features of the Layer2 Layer specifically includes:
adding a linear link layer, compressing the dimension of the output feature to the number of categories, wherein the compressed feature is as follows:
wherein Linear () represents a Linear chaining operation, Wcls2Is a weight coefficient of a linear link layer whose dimension is Wcls2∈R2048,n。
S23-1: the output of the third Layer3 is L3Dimension of L3∈R1,1024,17,8The input of the Layer is the output L of Layer22After convolution operation, we get:
L3_conv=Conv(L3)
wherein L is3_convAdopting Conv () as 1 × 1 convolution kernel, inputting 1024-dimension and outputting 2048-dimension convolution operation, then outputting L3_conv∈R1,2048,17,8。
S23-2: to L3_convThe average value operation is carried out, and the average value operation,the output characteristics of the layer are obtained:
F3=Avg(L3_conv)
wherein, F3Is the output characteristic of Layer3 and has the dimension of F3∈R1,2048Avg () is the maximum mean operation.
S23-3: the operation of performing channel dimension transformation on the output features of the Layer3 Layer specifically includes:
adding a linear link layer to the output features, compressing the dimensions of the output features to the number of categories, and then the compressed features are:
wherein Linear () represents a Linear chaining operation, Wcls3Is a weight coefficient of a linear link layer whose dimension is Wcls3∈R2048,n。
S24-1: the output of the fourth Layer4 is L4Dimension of L4∈R1,2048,9,4The input of the Layer is the output L of Layer33Directly carrying out mean value operation on the obtained product to obtain an output characteristic F4Dimension F4∈R1,2048:
F4=Avg(L4)
S24-2: the operation of performing channel dimension transformation on the output features of the Layer3 Layer specifically includes:
adding a linear link layer to the output features, compressing the dimensions of the output features to the number of categories, and then the compressed features are:
wherein Linear () represents a Linear chaining operation, Wcls4Is a weight coefficient of a linear link layer whose dimension is Wcls4∈R2048,n。
In step S3, performing a feature aggregation loss operation on the compressed features of each layer, specifically including:
for the features of Layer1, cosine transform is performed according to the cosine theorem:
the Layer1 Layer loss function is:
wherein N is the number of input samples,is a linear link layer weight coefficient Wcls1The feature outputs a link weight to the ith output neuron;is the output of the ith neuron in the linear link layer;is the cosine included angle; m is a fixed interval parameter, and m is taken in general>0; s is a scaling factor.
Similarly, cosine transform is performed on the Layer2 Layer class output, and the loss function of the Layer is obtained as follows:
wherein the content of the first and second substances,is a linear link layer weight coefficient Wcls2The feature outputs a link weight to the ith output neuron;is the output of the ith neuron in the linear link layer;is the cosine angle.
Cosine transforming the Layer3 Layer class output and obtaining the loss function of the Layer as:
wherein the content of the first and second substances,is a linear link layer weight coefficient Wcls3The feature outputs a link weight to the ith output neuron;is the output of the ith neuron in the linear link layer;is the cosine included angle;
cosine transforming the Layer4 Layer class output and obtaining the loss function of the Layer as:
wherein the content of the first and second substances,is a linear link layer weight coefficient Wcls4The feature outputs a link weight to the ith output neuron;is the output of the ith neuron in the linear link layer;is the cosine angle.
Based on the four types of loss functions, the total loss function of the classification model is as follows:
After cosine transformation, the features are mapped to a hypersphere on an angle space, the radius of the hypersphere is controlled by a scaling factor s, the aggregation degree of the features is controlled by an interval parameter m, and with the increase of m, the difference between the features is more obvious, the polymerization degree of the features is higher, and the feature area is more compact, so that the difference between samples is increased.
In this embodiment, when the classification model is trained, the existing optimization method, such as a gradient correction method, may be used to correct the network model parameters and the feature fusion weights, and obtain the optimal parameters, so as to obtain the optimal classification model.
The embodiment enhances the difference between different colors and different dress types based on a deep learning algorithm, and fully utilizes the light color information and the deep contour information on the basis of deep abstract characteristics by training a weighting fusion coefficient. Meanwhile, the feature aggregation operation is added, the feature information is mapped to an angle space, the maximized classification boundary is obtained, and the similarity difference in the standard dressing matching process is increased.
Example 2
The embodiment provides a dressing classification system based on feature aggregation, which comprises:
the characteristic extraction module is configured to perform multi-layer characteristic extraction on the image training set based on a pre-constructed classification model;
the dimension transformation module is configured to perform channel dimension transformation on the extracted features of each layer, and compress the feature dimensions to be consistent with the number of categories in the image training set to obtain compressed features;
the characteristic aggregation module is configured to perform cosine transform on the compressed characteristics of each layer, map the characteristics to an angle space, aggregate the characteristics, and construct a loss function after the characteristics are aggregated;
and the dressing classification module is configured to train the classification model according to the loss function, and the dressing image to be recognized is subjected to dressing classification according to the trained classification model.
It should be noted that the modules correspond to the steps described in embodiment 1, and the modules are the same as the corresponding steps in the implementation examples and application scenarios, but are not limited to the disclosure in embodiment 1. It should be noted that the modules described above as part of a system may be implemented in a computer system such as a set of computer-executable instructions.
In further embodiments, there is also provided:
an electronic device comprising a memory and a processor and computer instructions stored on the memory and executed on the processor, the computer instructions when executed by the processor performing the method of embodiment 1. For brevity, no further description is provided herein.
It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general purpose processors, digital signal processors DSP, application specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and so on. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory may include both read-only memory and random access memory, and may provide instructions and data to the processor, and a portion of the memory may also include non-volatile random access memory. For example, the memory may also store device type information.
A computer readable storage medium storing computer instructions which, when executed by a processor, perform the method described in embodiment 1.
The method in embodiment 1 may be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The software modules may be located in ram, flash, rom, prom, or eprom, registers, among other storage media as is well known in the art. The storage medium is located in a memory, and a processor reads information in the memory and completes the steps of the method in combination with hardware of the processor. To avoid repetition, it is not described in detail here.
Those of ordinary skill in the art will appreciate that the various illustrative elements, i.e., algorithm steps, described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (10)
1. A clothing classification method based on feature aggregation is characterized by comprising the following steps:
carrying out multi-layer feature extraction on the image training set based on a pre-constructed classification model;
performing channel dimension transformation on the extracted features of each layer, and compressing the feature dimensions to be consistent with the number of categories in the image training set to obtain compressed features;
performing cosine transformation on the compressed features of each layer, mapping the features to an angle space, aggregating the features, and constructing a loss function after the feature aggregation;
and training the classification model according to the loss function, and dressing the image to be recognized according to the trained classification model.
2. The method for classifying dresses based on feature aggregation as claimed in claim 1, wherein the multi-layer feature extraction process comprises the steps of sequentially performing feature extraction on the outputs of each layer except the last layer by convolution operation and maximum mean value operation; and taking the output of more than one characteristic layer as input, and directly carrying out maximum mean value operation.
3. The method of claim 1, wherein the channel dimension transformation is to add a linear link layer to the classification model, and the feature dimension is compressed according to a weight coefficient of the linear link layer.
wherein Linear () represents a Linear chaining operation, WclsIs a weight coefficient of a linear link layer whose dimension is Wcls∈R2048 ,nN is the number of classes in the image training set, and F is the feature extracted from each layer.
5. The method of claim 1, wherein the method is based on feature aggregationThe method for classifying the loads is characterized in that cosine transform is carried out on the compressed features of each layer to obtain: fcls=||Wcls||||F||cosθ;
Wherein, FclsFor features after compression, WclsThe weight coefficient of the linear link layer is shown, n is the number of categories in the image training set, F is the feature extracted from each layer, and theta is the cosine included angle.
6. The method of claim 1, wherein the penalty function for each layer is:
wherein the content of the first and second substances,is a linear link layer weight coefficient WclsThe feature outputs a link weight to the ith output neuron; f. ofiIs the output of the ith neuron in the linear link layer; thetaiIs a cosine included angle; m is an interval parameter; s is a scaling factor.
7. The method of claim 6, wherein in the process of mapping the features to the angle space and aggregating the features, the features are mapped to a hypersphere on the angle space, the hypersphere radius is controlled by a scaling factor s, the degree of aggregation of the features is controlled by an interval parameter m, and as m increases, the higher the difference between the features, the higher the degree of aggregation of the features.
8. A clothing classification system based on feature aggregation, comprising:
the characteristic extraction module is configured to perform multi-layer characteristic extraction on the image training set based on a pre-constructed classification model;
the dimension transformation module is configured to perform channel dimension transformation on the extracted features of each layer, and compress the feature dimensions to be consistent with the number of categories in the image training set to obtain compressed features;
the characteristic aggregation module is configured to perform cosine transform on the compressed characteristics of each layer, map the characteristics to an angle space, aggregate the characteristics, and construct a loss function after the characteristics are aggregated;
and the dressing classification module is configured to train the classification model according to the loss function, and the dressing image to be recognized is subjected to dressing classification according to the trained classification model.
9. An electronic device comprising a memory and a processor and computer instructions stored on the memory and executed on the processor, the computer instructions when executed by the processor performing the method of any of claims 1-7.
10. A computer-readable storage medium storing computer instructions which, when executed by a processor, perform the method of any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111112584.8A CN113989541A (en) | 2021-09-23 | 2021-09-23 | Dressing classification method and system based on feature aggregation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111112584.8A CN113989541A (en) | 2021-09-23 | 2021-09-23 | Dressing classification method and system based on feature aggregation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113989541A true CN113989541A (en) | 2022-01-28 |
Family
ID=79736374
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111112584.8A Pending CN113989541A (en) | 2021-09-23 | 2021-09-23 | Dressing classification method and system based on feature aggregation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113989541A (en) |
Citations (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107463948A (en) * | 2017-07-13 | 2017-12-12 | 西安电子科技大学 | Classification of Multispectral Images method based on binary channels multiple features fusion network |
CN108399421A (en) * | 2018-01-31 | 2018-08-14 | 南京邮电大学 | A kind of zero sample classification method of depth of word-based insertion |
US20180247113A1 (en) * | 2016-10-10 | 2018-08-30 | Gyrfalcon Technology Inc. | Image Classification Systems Based On CNN Based IC and Light-Weight Classifier |
US20180314982A1 (en) * | 2017-04-28 | 2018-11-01 | At&T Intellectual Property I, L.P. | Bridging heterogeneous domains with parallel transport and sparse coding for machine learning models |
CN109522557A (en) * | 2018-11-16 | 2019-03-26 | 中山大学 | Training method, device and the readable storage medium storing program for executing of text Relation extraction model |
CN109858575A (en) * | 2019-03-19 | 2019-06-07 | 苏州市爱生生物技术有限公司 | Data classification method based on convolutional neural networks |
CN110097000A (en) * | 2019-04-29 | 2019-08-06 | 东南大学 | Video behavior recognition methods based on local feature Aggregation Descriptor and sequential relationship network |
CN110215216A (en) * | 2019-06-11 | 2019-09-10 | 中国科学院自动化研究所 | Based on the with different levels Activity recognition method in skeletal joint point subregion, system |
CN110309888A (en) * | 2019-07-11 | 2019-10-08 | 南京邮电大学 | A kind of image classification method and system based on layering multi-task learning |
US20190353703A1 (en) * | 2018-05-16 | 2019-11-21 | Wuhan University | Analog circuit fault feature extraction method based on parameter random distribution neighbor embedding winner-take-all method |
CN110689060A (en) * | 2019-09-16 | 2020-01-14 | 西安电子科技大学 | Heterogeneous image matching method based on aggregation feature difference learning network |
CN111260032A (en) * | 2020-01-14 | 2020-06-09 | 北京迈格威科技有限公司 | Neural network training method, image processing method and device |
CN111340150A (en) * | 2020-05-22 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | Method and device for training first classification model |
CN111832650A (en) * | 2020-07-14 | 2020-10-27 | 西安电子科技大学 | Image classification method based on generation of confrontation network local aggregation coding semi-supervision |
CN112020724A (en) * | 2019-04-01 | 2020-12-01 | 谷歌有限责任公司 | Learning compressible features |
CN112580590A (en) * | 2020-12-29 | 2021-03-30 | 杭州电子科技大学 | Finger vein identification method based on multi-semantic feature fusion network |
CN112819073A (en) * | 2021-02-01 | 2021-05-18 | 上海明略人工智能(集团)有限公司 | Classification network training method, image classification device and electronic equipment |
WO2021115159A1 (en) * | 2019-12-09 | 2021-06-17 | 中兴通讯股份有限公司 | Character recognition network model training method, character recognition method, apparatuses, terminal, and computer storage medium therefor |
CN113033283A (en) * | 2020-12-18 | 2021-06-25 | 神思电子技术股份有限公司 | Improved video classification system |
CN113040772A (en) * | 2021-03-10 | 2021-06-29 | 效隆神思(厦门)科技发展有限公司 | Classification method of syphilis addicts and normal persons based on electroencephalogram signals |
US20210248421A1 (en) * | 2020-02-06 | 2021-08-12 | Shenzhen Malong Technologies Co., Ltd. | Channel interaction networks for image categorization |
-
2021
- 2021-09-23 CN CN202111112584.8A patent/CN113989541A/en active Pending
Patent Citations (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180247113A1 (en) * | 2016-10-10 | 2018-08-30 | Gyrfalcon Technology Inc. | Image Classification Systems Based On CNN Based IC and Light-Weight Classifier |
US20180314982A1 (en) * | 2017-04-28 | 2018-11-01 | At&T Intellectual Property I, L.P. | Bridging heterogeneous domains with parallel transport and sparse coding for machine learning models |
CN107463948A (en) * | 2017-07-13 | 2017-12-12 | 西安电子科技大学 | Classification of Multispectral Images method based on binary channels multiple features fusion network |
CN108399421A (en) * | 2018-01-31 | 2018-08-14 | 南京邮电大学 | A kind of zero sample classification method of depth of word-based insertion |
US20190353703A1 (en) * | 2018-05-16 | 2019-11-21 | Wuhan University | Analog circuit fault feature extraction method based on parameter random distribution neighbor embedding winner-take-all method |
CN109522557A (en) * | 2018-11-16 | 2019-03-26 | 中山大学 | Training method, device and the readable storage medium storing program for executing of text Relation extraction model |
CN109858575A (en) * | 2019-03-19 | 2019-06-07 | 苏州市爱生生物技术有限公司 | Data classification method based on convolutional neural networks |
CN112020724A (en) * | 2019-04-01 | 2020-12-01 | 谷歌有限责任公司 | Learning compressible features |
CN110097000A (en) * | 2019-04-29 | 2019-08-06 | 东南大学 | Video behavior recognition methods based on local feature Aggregation Descriptor and sequential relationship network |
CN110215216A (en) * | 2019-06-11 | 2019-09-10 | 中国科学院自动化研究所 | Based on the with different levels Activity recognition method in skeletal joint point subregion, system |
CN110309888A (en) * | 2019-07-11 | 2019-10-08 | 南京邮电大学 | A kind of image classification method and system based on layering multi-task learning |
CN110689060A (en) * | 2019-09-16 | 2020-01-14 | 西安电子科技大学 | Heterogeneous image matching method based on aggregation feature difference learning network |
WO2021115159A1 (en) * | 2019-12-09 | 2021-06-17 | 中兴通讯股份有限公司 | Character recognition network model training method, character recognition method, apparatuses, terminal, and computer storage medium therefor |
CN111260032A (en) * | 2020-01-14 | 2020-06-09 | 北京迈格威科技有限公司 | Neural network training method, image processing method and device |
US20210248421A1 (en) * | 2020-02-06 | 2021-08-12 | Shenzhen Malong Technologies Co., Ltd. | Channel interaction networks for image categorization |
CN111340150A (en) * | 2020-05-22 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | Method and device for training first classification model |
CN111832650A (en) * | 2020-07-14 | 2020-10-27 | 西安电子科技大学 | Image classification method based on generation of confrontation network local aggregation coding semi-supervision |
CN113033283A (en) * | 2020-12-18 | 2021-06-25 | 神思电子技术股份有限公司 | Improved video classification system |
CN112580590A (en) * | 2020-12-29 | 2021-03-30 | 杭州电子科技大学 | Finger vein identification method based on multi-semantic feature fusion network |
CN112819073A (en) * | 2021-02-01 | 2021-05-18 | 上海明略人工智能(集团)有限公司 | Classification network training method, image classification device and electronic equipment |
CN113040772A (en) * | 2021-03-10 | 2021-06-29 | 效隆神思(厦门)科技发展有限公司 | Classification method of syphilis addicts and normal persons based on electroencephalogram signals |
Non-Patent Citations (6)
Title |
---|
JIANKANG DENG: "ArcFace: Additive Angular Margin Loss for Deep Face Recognition", 《IEEE》, 9 June 2021 (2021-06-09), pages 2 * |
KATARZYNA JANOCHA, WOJCIECH MARIAN CZARNECKI: "On Loss Functions for Deep Neural Networks in Classification", 《COMPUTER SCIENCE 》, 18 February 2017 (2017-02-18) * |
KIM, HJ (KIM, HYO JIN) [1] ; LEE, DH (LEE, DOO HEE) [1] ; NIAZ, A (NIAZ, ASIM) [2] ; KIM, CY (KIM, CHAN YONG) [1] ; MEMON, AA (MEM: "Multiple-Clothing Detection and Fashion Landmark Estimation Using a Single-Stage Detector", 《IEEE ACCESS》, 10 February 2021 (2021-02-10) * |
曹建芳;崔红艳;张琦;: "特征融合AlexNet模型的古代壁画分类", 中国图象图形学报, no. 01, 16 January 2020 (2020-01-16) * |
李振东;钟勇;陈蔓;王理顺;: "角度余量损失和中心损失联合的深度人脸识别", 计算机应用, no. 2, 30 December 2019 (2019-12-30) * |
罗金梅;罗建;李艳梅;赵旭;: "基于多特征融合CNN的人脸识别算法研究", 航空计算技术, no. 03, 25 May 2019 (2019-05-25) * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106599854B (en) | Automatic facial expression recognition method based on multi-feature fusion | |
CN112686331B (en) | Forged image recognition model training method and forged image recognition method | |
CN111428556B (en) | Traffic sign recognition method based on capsule neural network | |
Li et al. | Rank-SIFT: Learning to rank repeatable local interest points | |
CN102194114B (en) | Method for recognizing iris based on edge gradient direction pyramid histogram | |
CN107239759B (en) | High-spatial-resolution remote sensing image transfer learning method based on depth features | |
CN111898621B (en) | Contour shape recognition method | |
JP2017004480A (en) | Conspicuity information acquisition device and conspicuity information acquisition method | |
CN107506765B (en) | License plate inclination correction method based on neural network | |
CN111126240B (en) | Three-channel feature fusion face recognition method | |
JP4098021B2 (en) | Scene identification method, apparatus, and program | |
CN105760858A (en) | Pedestrian detection method and apparatus based on Haar-like intermediate layer filtering features | |
CN112001302B (en) | Face recognition method based on face interesting region segmentation | |
CN112232300B (en) | Global occlusion self-adaptive pedestrian training/identifying method, system, equipment and medium | |
CN104143091B (en) | Based on the single sample face recognition method for improving mLBP | |
CN112836651B (en) | Gesture image feature extraction method based on dynamic fusion mechanism | |
JP6713162B2 (en) | Image recognition device, image recognition method, and image recognition program | |
CN112101467A (en) | Hyperspectral image classification method based on deep learning | |
CN110188646B (en) | Human ear identification method based on fusion of gradient direction histogram and local binary pattern | |
CN109086801A (en) | A kind of image classification method based on improvement LBP feature extraction | |
CN105354547A (en) | Pedestrian detection method in combination of texture and color features | |
CN105809132A (en) | Improved compressed sensing-based face recognition method | |
CN113989541A (en) | Dressing classification method and system based on feature aggregation | |
CN112419278A (en) | Deep learning-based solid wood floor classification method | |
CN112348767A (en) | Wood counting model based on object edge detection and feature matching |
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 |