A sock size measurement method, system and electronic device based on neural network

Abstract

The present invention relates to the technical field of footwear products, and specifically to a sock size measurement method, system and electronic device based on a neural network. First, production information, standard sock size information, user information and sock images are obtained; then, the sock feature recognition information output by the sock image feature recognition model is evaluated to obtain a sock deformation degree value; then, the pre-processed sock image is input into a sock size detection model based on multi-information fusion guidance for detection to obtain a sock size detection value; the model uses a multi-information feature fusion guidance module to integrate the standard sock size information and the user information into the detection model to improve the size detection accuracy; finally, the sock size detection value is corrected using a sock size correction formula to obtain a final sock size measurement value.

Description

Technical Field

[0001] The present invention relates to the technical field of footwear products, and in particular to a sock size measurement method, system and electronic equipment based on a neural network. Background Art

[0002] In the production process of socks, sock size measurement is an important link to ensure product quality. It can not only timely detect and eliminate unqualified products and reduce the defective rate, but also help enterprises to adjust production process parameters in time to avoid batch quality problems. Sock size will be affected by factors such as production materials, production process and manual measurement, which will affect the appearance of the product and the user experience. Therefore, it is essential to measure the size of the socks after production.

[0003] At present, there are still some shortcomings in the existing technology for sock size measurement methods; most of the size measurement methods use traditional manual measurement methods, which are time-consuming and labor-intensive, and will cause deviations in size measurement results, thereby reducing the efficiency and accuracy of sock size measurement.

[0004] Therefore, a sock size measurement method, system and electronic device based on neural network are proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a sock size measurement method, system and electronic device based on neural network. First, obtain production information, standard sock size information, user information and sock image; then, evaluate the sock feature recognition information output by the sock image feature recognition model to obtain the sock deformation degree value; then, input the pre-processed sock image into the sock size detection model based on multi-information fusion guidance for detection to obtain the sock size detection value; the model uses the multi-information feature fusion guidance module to integrate the standard sock size information and the user information into the detection model to improve the size detection accuracy; finally, use the sock size correction formula to correct the sock size detection value to obtain the final sock size measurement value.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A sock size measurement method based on neural network, comprising:

[0008] Obtain production information, standard sock size information, user information, and sock images;

[0009] Preprocessing the sock image to obtain a preprocessed sock image;

[0010] Constructing a sock image feature recognition model, inputting the preprocessed sock image into the sock image feature recognition model for recognition, and obtaining sock feature recognition information; evaluating the sock feature recognition information, and calculating a sock deformation degree value;

[0011] Constructing a sock size detection model based on multi-information fusion guidance, inputting the pre-processed sock image into the sock size detection model based on multi-information fusion guidance for detection, and obtaining a sock size detection value;

[0012] Wherein, the multi-information fusion guidance is implemented by a multi-information feature fusion guidance module in the sock size detection model;

[0013] The sock size detection value is corrected by combining the production information, the sock deformation degree value and the sock size detection value using a sock size correction formula to obtain a final sock size measurement value.

[0014] Furthermore, the sock image is preprocessed to obtain a preprocessed sock image. The specific implementation process includes:

[0015] Get the socks image captured by the camera;

[0016] Performing denoising processing on the sock image by using Gaussian filtering to obtain a denoised sock image;

[0017] Using histogram equalization to adjust the contrast and brightness of the denoised sock image to obtain an enhanced sock image;

[0018] Performing edge processing on the enhanced socks image using a Laplace operator to obtain an edge-enhanced socks image;

[0019] A normalization operation is performed on the edge enhanced sock image to obtain a preprocessed sock image.

[0020] Further, a sock image feature recognition model is constructed, and the pre-processed sock image is input into the sock image feature recognition model for recognition to obtain sock feature recognition information; the sock feature recognition information is evaluated, and the specific implementation process of calculating the sock deformation degree value includes:

[0021] Get the preprocessed sock image;

[0022] Inputting the pre-processed sock image into a sock image feature recognition model for recognition, and obtaining sock feature recognition information; wherein the sock feature recognition information includes: size feature, deformation feature and wear feature;

[0023] Obtaining the size deviation, deformation amount and wear area of ​​the socks from the size feature, the deformation feature and the wear feature;

[0024] The size deviation, the deformation amount and the wear area are weightedly evaluated to obtain a sock deformation degree value.

[0025] Furthermore, a sock size detection model based on multi-information fusion guidance is constructed, and the pre-processed sock image is input into the sock size detection model based on multi-information fusion guidance for detection. The specific implementation process of obtaining the sock size detection value includes:

[0026] Obtain pre-processed sock images, standard sock size information, and user information;

[0027] Performing feature extraction on the standard sock size information and the user information to obtain standard sock size features and user features;

[0028] Constructing a sock size detection model based on multi-information fusion guidance, inputting the pre-processed sock image, the standard sock size features and the user features into the sock size detection model for detection, and obtaining a sock size detection value;

[0029] The sock size detection model implements the multi-information fusion guidance operation through a multi-information feature fusion guidance module; the specific formula of the multi-information feature fusion guidance module is:

[0030] ;

[0031] in, M Represented as multi-information fusion features; Conv () represents a convolution operation with a convolution kernel size of 3×3; Concat () represents the channel merging operation; S Expressed as the standard sock size characteristics; U represented as said user features; F () represents the merging function of the multi-information feature fusion guidance module; X is represented as the encoded features in the sock size detection model; Sigmoid () represents a nonlinear activation function.

[0032] Furthermore, the sock size detection value is corrected by combining the production information, the sock deformation degree value and the sock size detection value using the sock size correction formula, and the calculation formula for the final sock size measurement value is obtained as follows:

[0033] ;

[0034] Among them, CMCL is the final sock size measurement value, CMJC is the sock size detection value, and BXCD is the sock deformation degree value. is the temperature coefficient, is the actual temperature, is the reference temperature, is the humidity coefficient, is the actual humidity, For reference humidity.

[0035] A sock size measurement system based on neural network, comprising: a system control module, a data acquisition module, a data processing module, a sock deformation detection module, a sock size detection module and an output module;

[0036] Wherein, the system control module is used to control the start, pause and stop of the system;

[0037] The data acquisition module is used to obtain the collected production information, standard sock size information, user information and sock images;

[0038] The data processing module is used to preprocess the sock image to obtain a preprocessed sock image;

[0039] The sock deformation detection module is used to evaluate the sock features output by the sock image feature recognition model to obtain a sock deformation degree value;

[0040] The sock size detection module detects the pre-processed sock image using a sock size detection model guided by multi-information fusion to obtain a sock size detection value;

[0041] The output module includes a correction unit and an output unit; the correction unit corrects the sock size detection value using a sock size correction formula to obtain a final sock size measurement value; the output unit is used to output the sock size measurement result to the staff.

[0042] Furthermore, the sock deformation detection module is used to evaluate the sock features output by the sock image feature recognition model, and the specific implementation process of obtaining the sock deformation degree value includes:

[0043] Get the preprocessed sock image;

[0044] Inputting the pre-processed sock image into a sock image feature recognition model for recognition, and obtaining sock feature recognition information; wherein the sock feature recognition information includes: size feature, deformation feature and wear feature;

[0045] Obtaining the size deviation, deformation amount and wear area of ​​the socks from the size feature, the deformation feature and the wear feature;

[0046] The size deviation, the deformation amount and the wear area are weightedly evaluated to obtain a sock deformation degree value.

[0047] Furthermore, the sock size detection module detects the pre-processed sock image using a sock size detection model guided by multi-information fusion, and the specific implementation process of obtaining the sock size detection value includes:

[0048] Obtain pre-processed sock images, standard sock size information, and user information;

[0049] Performing feature extraction on the standard sock size information and the user information to obtain standard sock size features and user features;

[0050] Constructing a sock size detection model based on multi-information fusion guidance, inputting the pre-processed sock image, the standard sock size features and the user features into the sock size detection model for detection, and obtaining a sock size detection value;

[0051] The sock size detection model implements the multi-information fusion guidance operation through a multi-information feature fusion guidance module; the specific formula of the multi-information feature fusion guidance module is:

[0052] ;

[0053] in, M Represented as multi-information fusion features; Conv () represents a convolution operation with a convolution kernel size of 3×3; Concat () represents the channel merging operation; S Expressed as the standard sock size characteristics; U represented as said user features; F () represents the merging function of the multi-information feature fusion guidance module; X is represented as the encoded features in the sock size detection model; Sigmoid () represents a nonlinear activation function.

[0054] Furthermore, the correction unit corrects the sock size detection value using a sock size correction formula, and obtains a calculation formula for a final sock size measurement value as follows:

[0055] ;

[0056] Among them, CMCL is the final sock size measurement value, CMJC is the sock size detection value, and BXCD is the sock deformation degree value. is the temperature coefficient, is the actual temperature, is the reference temperature, is the humidity coefficient, is the actual humidity, For reference humidity.

[0057] An electronic device comprises: a memory and a processor; the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the sock size measurement method based on a neural network provided by the present invention.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. The present invention proposes a sock deformation detection function for detecting the deformation degree of socks by using an image recognition method; the function first recognizes the sock image by using a sock image feature recognition model to obtain sock feature recognition information; then, the sock feature recognition information is comprehensively evaluated to calculate the sock deformation degree value; the function uses the sock deformation degree value to judge the deformation degree of the socks, and provides accurate numerical support for subsequent sock size correction, thereby effectively improving the accuracy of sock size measurement.

[0060] 2. The present invention proposes a sock size detection model based on multi-information fusion guidance for obtaining sock size detection values; the model uses a multi-information feature fusion guidance module to implement the multi-information fusion guidance operation; the multi-information feature fusion guidance module guides and enhances the coding features by using user information features and standard size information features, which is beneficial to improving the detection accuracy output by the sock size detection model, thereby effectively improving the efficiency and accuracy of sock size measurement.

[0061] 3. The present invention proposes a sock size correction function for improving the accuracy of size measurement by using a sock size correction formula; the function corrects the sock size detection value output by the sock size detection model by using the sock size correction formula; the sock size correction formula is obtained by comprehensive calculation based on the sock size detection value, the sock deformation degree value, and the production temperature and humidity; the function further improves the accuracy of size measurement by combining the actual production environment and the sock production deformation degree, thereby effectively improving the efficiency and accuracy of sock size measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of a process of a sock size measurement method based on a neural network according to the present invention;

[0063] Figure 2 It is a structural schematic diagram of the socks image feature recognition model of the present invention;

[0064] Figure 3 It is a structural schematic diagram of the sock size detection model based on multi-information fusion guidance of the present invention;

[0065] Figure 4It is a structural schematic diagram of a sock size measurement system based on a neural network of the present invention;

[0066] Figure 5 The figure is a structural diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] In the production process of socks, sock size measurement is an important link to ensure product quality. It can not only timely detect and eliminate unqualified products and reduce the defective rate, but also help enterprises to adjust production process parameters in time to avoid batch quality problems. Sock size will be affected by factors such as production materials, production process and manual measurement, which will affect the appearance of the product and the user experience. Therefore, it is essential to measure the size of the socks after production.

[0069] At present, there are still some shortcomings in the existing technology for sock size measurement methods; most of the size measurement methods use traditional manual measurement methods, which are time-consuming and labor-intensive, and will cause deviations in size measurement results, thereby reducing the efficiency and accuracy of sock size measurement.

[0070] Embodiment 1

[0071] The specific implementation process in the embodiment of the present application is implemented by a sock size measurement method based on a neural network of the present invention, see Figure 1 The process of the method proposed in the present invention is described in the following; the sock size measurement method based on neural network includes:

[0072] S10. Obtaining production information, standard sock size information, user information, and sock images;

[0073] S20. performing a preprocessing operation on the sock image to obtain a preprocessed sock image;

[0074] S30. Inputting the pre-processed sock image into a sock image feature recognition model for recognition to obtain sock feature recognition information; and obtaining a sock deformation degree value according to the sock feature recognition information;

[0075] S40. Inputting the pre-processed sock image into the sock size detection model guided by multi-information fusion for detection, and obtaining a sock size detection value;

[0076] S50. Obtain a sock size correction formula according to the production information, the sock deformation degree value and the sock size detection value; and use the sock size correction formula to correct the sock size detection value.

[0077] Furthermore, the specific implementation process of a sock size measurement method based on a neural network is as follows:

[0078] Obtain production information, standard sock size information, user information, and sock images;

[0079] Preprocessing the sock image to obtain a preprocessed sock image;

[0080] Constructing a sock image feature recognition model, inputting the preprocessed sock image into the sock image feature recognition model for recognition, and obtaining sock feature recognition information; evaluating the sock feature recognition information, and calculating a sock deformation degree value;

[0081] Constructing a sock size detection model based on multi-information fusion guidance, inputting the pre-processed sock image into the sock size detection model based on multi-information fusion guidance for detection, and obtaining a sock size detection value;

[0082] Wherein, the multi-information fusion guidance is implemented by a multi-information feature fusion guidance module in the sock size detection model;

[0083] The sock size detection value is corrected by combining the production information, the sock deformation degree value and the sock size detection value using a sock size correction formula to obtain a final sock size measurement value.

[0084] In this embodiment, a sock size measurement method based on a neural network is proposed; first, production information, standard sock size information, user information and sock images are obtained; then, the sock feature recognition information output by the sock image feature recognition model is evaluated to obtain a sock deformation degree value; then, the pre-processed sock image is input into a sock size detection model based on multi-information fusion guidance for detection to obtain a sock size detection value; the model uses a multi-information feature fusion guidance module to integrate the standard sock size information and the user information into the detection model to improve the size detection accuracy; finally, the sock size detection value is corrected using a sock size correction formula to obtain a final sock size measurement value; this method can effectively improve the efficiency and accuracy of sock size measurement.

[0085] For the purpose of specific explanation, the following embodiments are described as follows:

[0086] Obtain production information, standard sock size information, user information, and sock images;

[0087] In this embodiment, the temperature and humidity in the production information can be obtained through sensors; the socks image is collected through a camera; and the user information includes the user's age group and the user's gender.

[0088] Furthermore, the sock image is preprocessed to obtain a preprocessed sock image. The specific implementation process includes:

[0089] Get the socks image captured by the camera;

[0090] Performing denoising processing on the sock image by using Gaussian filtering to obtain a denoised sock image;

[0091] Using histogram equalization to adjust the contrast and brightness of the denoised sock image to obtain an enhanced sock image;

[0092] Performing edge processing on the enhanced socks image using a Laplace operator to obtain an edge-enhanced socks image;

[0093] A normalization operation is performed on the edge enhanced sock image to obtain a preprocessed sock image.

[0094] In this embodiment, the preprocessing process includes Gaussian denoising, histogram equalization, edge enhancement and normalization operations; wherein the Gaussian denoising is used to suppress Gaussian noise to improve the recognition accuracy of the model; the histogram equalization is used to adjust the image contrast using the image histogram to improve the image quality; the edge enhancement is used to enhance the high-frequency details of the image to improve the accuracy of the model in identifying high-frequency features; the edge enhancement algorithm can use the Sobel operator, Prewitt operator and Canny operator in addition to the Laplace operator; the normalization operation is used to compress the data into the [0,1] interval to improve the calculation efficiency of the model; this process can effectively improve the quality of the sock image, thereby improving the accuracy of the sock image feature recognition, and further improving the accuracy of the sock size measurement.

[0095] Further, a sock image feature recognition model is constructed, and the pre-processed sock image is input into the sock image feature recognition model for recognition to obtain sock feature recognition information; the sock feature recognition information is evaluated, and the specific implementation process of calculating the sock deformation degree value includes:

[0096] Get the preprocessed sock image;

[0097] Inputting the pre-processed sock image into a sock image feature recognition model for recognition, and obtaining sock feature recognition information; the sock feature recognition information includes: size feature, deformation feature and wear feature;

[0098] The structure of the socks image feature recognition model is as follows: Figure 2 As shown, it includes: input layer, sock feature extraction layer, sock feature marking layer, sock feature enhancement layer, sock feature recognition layer and output layer; the specific implementation process of the model includes:

[0099] Inputting the preprocessed sock image into the input layer to obtain an initial feature map;

[0100] Further, the initial feature map is input into the sock feature extraction layer to obtain a sock feature map; wherein the sock feature extraction layer comprises three cascaded residual blocks for deep extraction of sock features;

[0101] Further, the sock feature map is input into the sock feature marking layer to obtain an anchor frame; wherein the anchor frame is set to five different sizes;

[0102] Further, extracting features from the anchor frame and inputting them into the sock feature enhancement layer, using channel attention and spatial attention mechanisms to improve model recognition performance, and obtaining a recognition result;

[0103] Furthermore, the recognition result is input into the output layer to obtain sock feature recognition information.

[0104] Obtaining the size deviation, deformation amount and wear area of ​​the socks from the size feature, the deformation feature and the wear feature;

[0105] The size deviation, the deformation amount and the wear area are weightedly evaluated to obtain a sock deformation degree value; the calculation formula for the sock deformation degree value is:

[0106] ;

[0107] Wherein, BXCD represents the deformation degree of the socks, and the larger the value, the greater the deformation degree of the socks; Expressed as a size weight factor; expressed as a measured dimension taken from said dimension feature; Indicated as standard size; It is expressed as the size deviation threshold; Expressed as deformation weight factor; is expressed as the deformation amount; It is expressed as the deformation threshold; Expressed as wear weight factor; Expressed as the wear area; Expressed as the total area of ​​the sock.

[0108] In this embodiment , and They are set to 0.35, 0.35 and 0.3 respectively; the setting of the size deviation threshold and the deformation threshold is related to factors such as the sock material, sock type and production environment, and requires relevant staff to flexibly set them based on work experience and is not unique.

[0109] In this embodiment, a sock deformation detection function is proposed for detecting the deformation degree of socks by using an image recognition method; the function first recognizes the sock image by using a sock image feature recognition model to obtain sock feature recognition information; then, the sock feature recognition information is comprehensively evaluated to calculate the sock deformation degree value; the function uses the sock deformation degree value to judge the deformation degree of the socks, and provides accurate numerical support for subsequent sock size correction, thereby effectively improving the accuracy of sock size measurement.

[0110] Furthermore, a sock size detection model based on multi-information fusion guidance is constructed, and the pre-processed sock image is input into the sock size detection model based on multi-information fusion guidance for detection. The specific implementation process of obtaining the sock size detection value includes:

[0111] Obtain pre-processed sock images, standard sock size information, and user information;

[0112] Performing feature extraction on the standard sock size information and the user information to obtain standard sock size features and user features;

[0113] Constructing a sock size detection model based on multi-information fusion guidance, inputting the pre-processed sock image, the standard sock size features and the user features into the sock size detection model for detection, and obtaining a sock size detection value;

[0114] The structure of the sock size detection model based on multi-information fusion guidance can be referred to Figure 3 , including: input layer, encoding layer, 3 multi-information fusion guidance layers, 3 residual layers, feature fusion layer, decoding layer and output layer. Its specific implementation process includes:

[0115] Inputting the preprocessed sock image into the input layer to obtain sock features;

[0116] Further, the sock features are input into the encoding layer to obtain sock encoding features; wherein the encoding layer implements a downsampling operation by using a convolution kernel with a step size of 2 to reduce feature resolution;

[0117] Further, the sock coding feature is input into the multi-information fusion guidance layer to obtain the sock guidance feature; at the same time, the sock coding feature is input into the residual layer to obtain the sock residual feature; wherein the multi-information fusion guidance layer fuses the standard sock size feature and the user feature with the sock coding feature through a multi-information feature fusion guidance module;

[0118] Further, the sock guide feature and the sock residual feature are input into the feature fusion layer to obtain a sock fusion feature;

[0119] Further, the sock fusion feature is input into the decoding layer to obtain the sock decoding feature;

[0120] Furthermore, the sock decoding feature is input into the output layer to obtain the sock size detection value.

[0121] The sock size detection model implements the multi-information fusion guidance operation through a multi-information feature fusion guidance module; the specific formula of the multi-information feature fusion guidance module is:

[0122] ;

[0123] in, M Represented as multi-information fusion features; Conv () represents a convolution operation with a convolution kernel size of 3×3; Concat () represents the channel merging operation; S Expressed as the standard sock size characteristics; U represented as said user features; F () represents the merging function of the multi-information feature fusion guidance module; X is represented as the encoded features in the sock size detection model; Sigmoid () represents a nonlinear activation function.

[0124] In this embodiment, a sock size detection model based on multi-information fusion guidance is proposed to obtain sock size detection values; the model uses a multi-information feature fusion guidance module to implement the multi-information fusion guidance operation; the multi-information feature fusion guidance module guides and enhances the coding features by using user information features and standard size information features, which can help improve the detection accuracy output by the sock size detection model, thereby effectively improving the efficiency and accuracy of sock size measurement.

[0125] Furthermore, the sock size detection value is corrected by combining the production information, the sock deformation degree value and the sock size detection value using the sock size correction formula, and the calculation formula for the final sock size measurement value is obtained as follows:

[0126] ;

[0127] Among them, CMCL is the final sock size measurement value, CMJC is the sock size detection value, and BXCD is the sock deformation degree value. is the temperature coefficient, is the actual temperature, is the reference temperature, is the humidity coefficient, is the actual humidity, For reference humidity.

[0128] In this embodiment, a sock size correction function is proposed to improve the accuracy of size measurement by using a sock size correction formula; this function corrects the sock size detection value output by the sock size detection model by using the sock size correction formula; the sock size correction formula is obtained by comprehensive calculation based on the sock size detection value, the sock deformation degree value, and the production temperature and humidity; this function further improves the accuracy of size measurement by combining the actual production environment and the sock production deformation degree, thereby effectively improving the efficiency and accuracy of sock size measurement.

[0129] The present invention collects 1000 sock images from sock production factory A; wherein the sock images are obtained by using a camera to shoot socks of the same type under different production environments, and the socks are from the same production batch, so that the standard size of the produced socks can be guaranteed to be consistent with user information; and the preprocessed sock images are obtained by performing Gaussian denoising, contrast enhancement, edge enhancement and normalization on the sock images.

[0130] In order to verify the actual effect of the sock size measurement method based on neural network proposed in the present invention, multiple groups of comparative experiments were designed; among them, method one applied the sock size measurement method based on neural network proposed in the present invention, which included sock size detection and sock size correction based on the sock size detection model guided by multi-information fusion; method two only used the sock size detection model guided by multi-information fusion to perform sock size detection, and omitted the correction of sock size; method three adopted the traditional manual measurement method.

[0131] Table 1. Comparison of the effects of different measurement schemes

[0132]

[0133] As shown in Table 1, the comprehensive method combining the sock size detection model guided by multi-information fusion and the sock size correction (method 1) performed best, indicating that the method proposed in the present invention is the most accurate in sock size measurement.

[0134] Embodiment 2

[0135] As an embodiment of the present invention, refer to Figure 4 , a sock size measurement system based on neural network, comprising: a system control module, a data acquisition module, a data processing module, a sock deformation detection module, a sock size detection module and an output module;

[0136] Wherein, the system control module is used to control the start, pause and stop of the system;

[0137] The data acquisition module is used to obtain the collected production information, standard sock size information, user information and sock images;

[0138] The data processing module is used to preprocess the sock image to obtain a preprocessed sock image;

[0139] Furthermore, the process of obtaining the pre-processed sock image includes:

[0140] Get the socks image captured by the camera;

[0141] Further, the sock image is denoised by using Gaussian filtering to obtain a denoised sock image;

[0142] Further, the contrast and brightness of the denoised sock image are adjusted using histogram equalization to obtain an enhanced sock image;

[0143] Further, edge processing is performed on the enhanced sock image using a Laplace operator to obtain an edge-enhanced sock image;

[0144] Furthermore, a normalization operation is performed on the edge enhanced sock image to obtain a preprocessed sock image.

[0145] The sock deformation detection module is used to evaluate the sock features output by the sock image feature recognition model to obtain a sock deformation degree value;

[0146] Furthermore, the sock deformation detection module is used to evaluate the sock features output by the sock image feature recognition model, and the specific implementation process of obtaining the sock deformation degree value includes:

[0147] Get the preprocessed sock image;

[0148] Further, the pre-processed sock image is input into a sock image feature recognition model for recognition, to obtain sock feature recognition information; wherein the sock feature recognition information includes: size feature, deformation feature and wear feature;

[0149] Further, the size deviation, deformation and wear area of ​​the socks are obtained from the size feature, the deformation feature and the wear feature;

[0150] Furthermore, a weighted evaluation is performed on the size deviation, the deformation amount and the wear area to obtain a sock deformation degree value.

[0151] The sock size detection module detects the pre-processed sock image using a sock size detection model guided by multi-information fusion to obtain a sock size detection value;

[0152] Furthermore, the sock size detection module detects the pre-processed sock image using a sock size detection model guided by multi-information fusion, and the specific implementation process of obtaining the sock size detection value includes:

[0153] Obtain pre-processed sock images, standard sock size information, and user information;

[0154] Further, feature extraction is performed on the standard sock size information and the user information to obtain standard sock size features and user features;

[0155] Furthermore, a sock size detection model based on multi-information fusion guidance is constructed, and the pre-processed sock image, the standard sock size feature and the user feature are input into the sock size detection model for detection to obtain a sock size detection value; the sock size detection model realizes the multi-information fusion guidance operation through a multi-information feature fusion guidance module; the specific formula of the multi-information feature fusion guidance module is:

[0156] ;

[0157] in, M Represented as multi-information fusion features; Conv () represents a convolution operation with a convolution kernel size of 3×3; Concat () represents the channel merging operation; S Expressed as the standard sock size characteristics; U represented as said user features; F () represents the merging function of the multi-information feature fusion guidance module; X is represented as the encoded features in the sock size detection model; Sigmoid () represents a nonlinear activation function.

[0158] The present invention collects 3000 sock images of a sock production factory B; wherein the sock images are obtained by using a camera to shoot socks of the same type under different production environments, and the socks are from the same production batch, so that the standard sizes of the produced socks are consistent with user information; by performing a preprocessing operation on the sock images, a preprocessed sock image is obtained; the preprocessed sock image is divided into three sock size detection data sets, which are recorded as a first data set, a second data set and a third data set; wherein the first data set and the third data set contain defective sock image data; the defective sock image data includes blurred sock image data and worn sock image data.

[0159] In order to further verify the accuracy of the sock size detection model based on multi-information fusion guidance proposed in the present invention, multiple groups of comparative experiments were designed. The sock size detection model based on multi-information fusion guidance proposed in the present invention and the traditional convolutional neural network model were trained using the training sets in the three data sets respectively, and then the test sets in the data sets were input into the respective trained models to obtain the sock size detection values; wherein, the data sets were divided into training sets and test sets in a ratio of 8:2; finally, the sock size detection values ​​obtained by each output were compared.

[0160] Table 2. Comparison of detection effects of different data sets

[0161]

[0162] As shown in Table 2, compared with the traditional convolutional neural network model, the sock size detection model proposed in the present invention has the highest accuracy in detecting sock sizes under different data sets, and even in the third data set containing defective sock image data, it still has a high detection accuracy, indicating that the network model proposed in the present invention can effectively improve the accuracy of sock size detection.

[0163] The output module includes a correction unit and an output unit; the correction unit corrects the sock size detection value using a sock size correction formula to obtain a final sock size measurement value; the output unit is used to output the sock size measurement result to the staff;

[0164] Furthermore, the correction unit corrects the sock size detection value using a sock size correction formula, and obtains a calculation formula for a final sock size measurement value as follows:

[0165] ;

[0166] Among them, CMCL is the final sock size measurement value, CMJC is the sock size detection value, and BXCD is the sock deformation degree value. is the temperature coefficient, is the actual temperature, is the reference temperature, is the humidity coefficient, is the actual humidity, For reference humidity.

[0167] Embodiment 3

[0168] In order to implement the above-mentioned embodiments, an electronic device proposed by the present invention will be described in the embodiments of the present application; Figure 5 , the electronic device 1 comprises:

[0169] At least one memory 2, a sock size measurement program stored in the memory 2, and a processor 5; the processor 5 implements when executing the sock size measurement program Figure 1 The sock size measurement method described in steps S10-S50; wherein the processor 5 is connected to the sensor 3 and the camera 4, and is used to receive sock image data and production information; the memory 2 also contains standard size information, user information, sock deformation degree value and sock size detection value.

[0170] Specifically, the processor executes Figure 1 The program described in step S10 obtains the sock image data and production information from the camera and the sensor, and reads the standard size information and user information from the memory; then, the processor executes Figure 1 The program in step S20 pre-processes the sock image data to obtain a pre-processed sock image; then, the processor executes Figure 1 Steps S30-S40 are used to obtain the sock deformation degree value and the sock size detection value respectively, and store them in the memory; finally, the processor executes Figure 1 The program described in step S50 reads the sock deformation degree value and the sock size detection value from the memory, obtains the sock size correction formula according to the production information, the sock deformation degree value and the sock size detection value, and corrects the sock size detection value to obtain the final sock size measurement value.

[0171] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A sock size measurement method based on neural network, characterized in that: include: Acquire production information, standard sock size information, user information and sock images; wherein the production information includes production temperature and production humidity; the user information includes user age group and user gender; Preprocessing the sock image to obtain a preprocessed sock image; Constructing a sock image feature recognition model, inputting the preprocessed sock image into the sock image feature recognition model for recognition, and obtaining sock feature recognition information; evaluating the sock feature recognition information, and calculating a sock deformation degree value; A sock size detection model based on multi-information fusion guidance is constructed, and the pre-processed sock image is input into the sock size detection model based on multi-information fusion guidance for detection to obtain a sock size detection value; wherein the multi-information fusion guidance is realized by a multi-information feature fusion guidance module in the sock size detection model; the realization process of the sock size detection value includes: Acquire the pre-processed sock image, the standard sock size information and the user information; Performing feature extraction on the standard sock size information and the user information to obtain standard sock size features and user features; Constructing the sock size detection model guided by multi-information fusion, inputting the pre-processed sock image, the standard sock size features and the user features into the sock size detection model for detection, and obtaining the sock size detection value; The sock size detection model implements the multi-information fusion guidance operation through the multi-information feature fusion guidance module; the specific formula of the multi-information feature fusion guidance module is: in, M Represented as multi-information fusion features; Conv () represents a convolution operation with a convolution kernel size of 3×3; Concat () represents the channel merging operation; S Expressed as the standard sock size characteristics; U represented as said user features; F () represents the merging function of the multi-information feature fusion guidance module; X is represented as the encoded features in the sock size detection model; Sigmoid () represents a nonlinear activation function; Combining the production information, the sock deformation degree value and the sock size detection value, the sock size detection value is corrected using the sock size correction formula to obtain a final sock size measurement value; the calculation formula for the final sock size measurement value is: Among them, CMCL is the final sock size measurement value, CMJC is the sock size detection value, BXCD is the sock deformation degree value, is the temperature coefficient, is the actual temperature, is the reference temperature, is the humidity coefficient, is the actual humidity, For reference humidity.

2. A method for measuring sock size based on neural network according to claim 1, characterized in that: The specific implementation process of preprocessing the sock image to obtain the preprocessed sock image includes: Get the socks image captured by the camera; Performing denoising processing on the sock image by using Gaussian filtering to obtain a denoised sock image; Using histogram equalization to adjust the contrast and brightness of the denoised sock image to obtain an enhanced sock image; Performing edge processing on the enhanced socks image using a Laplace operator to obtain an edge-enhanced socks image; A normalization operation is performed on the edge enhanced sock image to obtain a preprocessed sock image.

3. The method for measuring sock size based on neural network according to claim 1, characterized in that: Constructing a socks image feature recognition model, inputting the preprocessed socks image into the socks image feature recognition model for recognition, and obtaining socks feature recognition information; The specific implementation process of evaluating the sock feature recognition information and calculating the sock deformation degree value includes: Get the preprocessed sock image; Inputting the pre-processed sock image into a sock image feature recognition model for recognition, and obtaining sock feature recognition information; wherein the sock feature recognition information includes: size feature, deformation feature and wear feature; Obtaining the size deviation, deformation amount and wear area of ​​the socks from the size feature, the deformation feature and the wear feature; The size deviation, the deformation amount and the wear area are weightedly evaluated to obtain a sock deformation degree value.

4. A sock size measurement system based on neural network, characterized in that: include: System control module, data acquisition module, data processing module, socks deformation detection module, socks size detection module and output module; Wherein, the system control module is used to control the start, pause and stop of the system; The data acquisition module is used to obtain the collected production information, standard sock size information, user information and sock images; wherein the production information includes production temperature and production humidity; the user information includes user age group and user gender; The data processing module is used to preprocess the sock image to obtain a preprocessed sock image; The sock deformation detection module is used to evaluate the sock features output by the sock image feature recognition model to obtain a sock deformation degree value; The sock size detection module detects the pre-processed sock image using a sock size detection model guided by multi-information fusion to obtain a sock size detection value; the process of realizing the sock size detection value includes: Acquire the pre-processed sock image, the standard sock size information and the user information; Performing feature extraction on the standard sock size information and the user information to obtain standard sock size features and user features; Constructing the sock size detection model guided by multi-information fusion, inputting the pre-processed sock image, the standard sock size features and the user features into the sock size detection model for detection, and obtaining the sock size detection value; The sock size detection model implements the multi-information fusion guidance operation through a multi-information feature fusion guidance module; the specific formula of the multi-information feature fusion guidance module is: in, M Represented as multi-information fusion features; Conv () represents a convolution operation with a convolution kernel size of 3×3; Concat () represents the channel merging operation; S Expressed as the standard sock size characteristics; U represented as said user features; F () represents the merging function of the multi-information feature fusion guidance module; X is represented as the encoded features in the sock size detection model; Sigmoid () represents a nonlinear activation function; The output module includes a correction unit and an output unit; the correction unit corrects the sock size detection value using a sock size correction formula to obtain a final sock size measurement value; the output unit is used to output the sock size measurement result to the staff; the calculation formula of the final sock size measurement value is: Among them, CMCL is the final sock size measurement value, CMJC is the sock size detection value, BXCD is the sock deformation degree value, is the temperature coefficient, is the actual temperature, is the reference temperature, is the humidity coefficient, is the actual humidity, For reference humidity.

5. The sock size measurement system based on neural network according to claim 4, characterized in that: The sock deformation detection module is used to evaluate the sock features output by the sock image feature recognition model, and the specific implementation process of obtaining the sock deformation degree value includes: Get the preprocessed sock image; Inputting the pre-processed sock image into a sock image feature recognition model for recognition, and obtaining sock feature recognition information; wherein the sock feature recognition information includes: size feature, deformation feature and wear feature; Obtaining the size deviation, deformation amount and wear area of ​​the socks from the size feature, the deformation feature and the wear feature; The size deviation, the deformation amount and the wear area are weightedly evaluated to obtain a sock deformation degree value.

6. An electronic device, characterized in that: The electronic device includes a memory and a processor; the memory stores a computer program, and when the computer program is executed by the processor, the processor executes a sock size measurement method based on a neural network as described in any one of claims 1 to 3.

Images