Wool fineness determination method and system based on deep learning

Abstract

The invention discloses a wool fineness determination method and system based on deep learning, and belongs to the field of wool fineness determination, and the method comprises the following steps: S1, collecting a wool fiber image, and obtaining a data set; s3, obtaining fineness characteristics of wool; s4, constructing a convolutional neural network model, inputting the training set into the convolutional neural network model for training, inputting the verification set into the trained convolutional neural network model, and verifying and evaluating the convolutional neural network model by using cross validation; s5, inputting the test set into the verified convolutional neural network model, predicting the fineness of the wool fibers by using the convolutional neural network model, and outputting a fineness evaluation report; and S6, displaying and storing a result. By adopting the wool fineness determination method and system based on deep learning, wool fineness determination can be efficiently and accurately completed, the advantages of simple operation, non-contact measurement, reliable result and the like are achieved, and the automation level of wool detection is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wool fineness measurement, and in particular to a wool fineness measurement method and system based on deep learning. Background Art

[0002] Wool fineness is one of the important indicators for evaluating wool quality, which directly affects the feel, luster and market value of wool. Traditional methods for measuring wool fineness mainly rely on microscope measurement and air flow method (such as AFIS system), but these methods have the disadvantages of complex operation, long time consumption and measurement accuracy greatly affected by human factors. Summary of the invention

[0003] The purpose of the present invention is to provide a wool fineness measurement method and system based on deep learning to solve the above technical problems.

[0004] To achieve the above object, the present invention provides a wool fineness determination method based on deep learning, comprising the following steps:

[0005] S1. Use a high-definition camera with a resolution of not less than 8 million pixels to collect wool fiber images to obtain a data set, and divide the data set into a training set, a validation set, and a test set according to the proportion, and manually annotate the wool fiber images in the training set;

[0006] S2, preprocessing the wool fiber image collected in step S1;

[0007] S3, based on SIFT feature extraction algorithm combined with Canny algorithm edge detection algorithm and gray-level co-occurrence matrix, feature extraction is performed on the pre-processed wool fiber image to obtain the fineness characteristics of the wool;

[0008] S4, constructing a convolutional neural network model, inputting the training set into the convolutional neural network model for training, inputting the validation set into the trained convolutional neural network model, and using cross-validation to validate and evaluate the convolutional neural network model;

[0009] S5, inputting the test set into the convolutional neural network model verified in step S4, using the convolutional neural network model to predict the fineness of the wool fiber, and outputting a fineness evaluation report;

[0010] S6. Result display and storage: The fineness evaluation report is displayed to the user through a visual interface, and the fineness evaluation report and the corresponding wool fiber image are stored in a database.

[0011] Preferably, the preprocessing described in step S2 includes denoising, color correction, enhancement processing, and cropping and scaling.

[0012] Preferably, step S3 specifically includes the following steps:

[0013] S31. Feature extraction using SIFT feature extraction algorithm:

[0014] Descriptor=SIFT(I)Descriptor=SIFT(I)

[0015] Where Descriptor represents the feature descriptor set extracted from the wool fiber image I; SIFT(I) represents the feature extraction of the wool fiber image I using the SIFT feature extraction algorithm;

[0016] S32, using the Canny algorithm edge detection algorithm to detect the edges in the wool fiber image I, extracting the contour and boundary information of the wool fiber to determine the shape and size of the wool fiber;

[0017] S33, extracting the texture characteristics of the wool fiber using the gray level co-occurrence matrix to analyze the surface roughness and smoothness of the wool fiber, thereby obtaining the structural and morphological information of the wool fiber;

[0018] S34, normalize the features extracted in step S31, step S32 and step S33.

[0019] Preferably, step S31 specifically includes the following steps:

[0020] S311. Calculate the scale space of wool fiber image:

[0021] L(x,y,σ)=G(x,y,σ)×I(x,y)

[0022] Where L(x, y, σ) represents the scale space image at scale σ; G(z, y, σ) is the Gaussian kernel function; (x, y) represents the spatial coordinates of the wool fiber image I, that is, the position of the pixel;

[0023] S312, search for key points by differential Gaussian:

[0024] D(x,y,σ)=L(x,y,kσ)-L(x,y,σ)

[0025] Where D(x,y,σ) represents the difference Gaussian image; L(x,y,kσ) represents the scale space image at scale kσ, and k is a constant;

[0026] S313: Determine the direction and descriptor of the key point.

[0027] Preferably, step S32 specifically includes the following steps:

[0028] S321, calculating the gradient amplitude and direction of the wool fiber image I to determine the strength and direction of the edge:

[0029]

[0030] Where Gx and Gy represent the first-order derivatives in the x-direction and y-direction respectively;

[0031] in,

[0032]

[0033] S322, calculate edge strength G:

[0034]

[0035] S323, non-maximum suppression: at each pixel, compare its gradient amplitude with the gradient amplitudes of its two adjacent pixel points in the gradient direction, and if the gradient amplitude of the current pixel is not the maximum, set it to 0 to remove non-maximum values;

[0036] S324, dual threshold processing: set two thresholds: T low and T high , and T low <T high , the gradient amplitude is greater than T high The pixel points with gradient amplitude between T low and T high The pixels between are marked as weak edges, and the weak edges are connected to the strong edges to form continuous edges, thereby obtaining the contour and boundary information of the wool fiber.

[0037] Preferably, step S33 specifically includes the following steps:

[0038] S331, set the selected relative position to (dx, dy) to construct a gray-level co-occurrence matrix (for example, (dx, dy) = (1, 0), indicating a horizontally adjacent pixel pair, that is, the pixel to the right of the current pixel):

[0039] P(i,j)=∑(x,y)∈I

[0040] δ(I(x,y)=i,I(x+dx,y+dy)=j)

[0041] Wherein, P(i,j) represents the element in the gray-level co-occurrence matrix, that is, the number of occurrences of the pixel with gray value i and the pixel with gray value j at its relative position (dx,dy); I(x,y) represents the gray value of the wool fiber image I at the coordinate (x,y), that is, the gray value of the current pixel; I(x+dx,y+dy) represents the gray value of the pixel at the relative position (dx,dy) to the current pixel; δ(I(x,y)=i,I(x+dx,y+dy)=j) represents the indicator function, which is used to judge whether the gray value of the current pixel is i and whether the gray value of the pixel at its relative position (dx,dy) is j;

[0042] S332, calculating texture features:

[0043]

[0044] Entropy=-∑i,j P(i,j)log(P(i,j))

[0045] Wherein, Contrast represents contrast, which is used to reflect the intensity of local grayscale changes in wool fiber images; Entropy represents entropy, which is used to reflect the uncertainty of grayscale distribution in wool fiber images.

[0046] Preferably, the normalization processing expression described in step S34 is as follows:

[0047]

[0048] In the formula, X′ represents the normalized eigenvalue; X represents the original eigenvalue; μ represents the mean value of the feature; σ1 represents the standard deviation of the feature;

[0049] in,

[0050]

[0051] Where N is the number of eigenvalues; x i Represents the eigenvalue.

[0052] Preferably, in step S4, the convolutional neural network model includes three convolutional layers, two pooling layers and one fully connected layer arranged in sequence, so as to perform the following convolution operation on the input wool fiber image I and the convolution kernel K:

[0053]

[0054] In the formula, (I*K)(x,y) represents the result of the convolution operation, that is, the convolution output value at the coordinate (x,y); (m,n) represents the index of the convolution kernel; I(m,n) represents the pixel value of the input wool fiber image I at the position (m,n); K(xm,yn) represents the weight value of the convolution kernel K at the position (xm,yn);

[0055] The convolutional neural network model uses a cross entropy loss function, and the cross entropy loss function L is expressed as follows:

[0056]

[0057] In the formula, yi represents the true label; Represents the predicted value.

[0058] Preferably, the convolutional neural network model uses the Adam optimizer optimization algorithm to adjust the model parameters, and the specific steps are as follows:

[0059] First calculate the initial momentum and secondary momentum:

[0060] m0=0,v0=0

[0061] In the formula, m0 represents the initialization momentum; v0 represents the initialization secondary momentum;

[0062] Then calculate the gradient:

[0063]

[0064] In the formula, g t represents the gradient of the current time step t; θ t-1 represents the model parameters at the previous time step;

[0065] Update the momentum and quadratic momentum again:

[0066] m t =β1m t -1+(1-β1)g t

[0067] v t =β2v t -1+(1-β2)g t 2

[0068] In the formula, m t represents the updated momentum; β1 represents the momentum decay rate; v t represents the updated secondary momentum; β2 represents the secondary momentum decay rate;

[0069] Then the deviation correction of momentum and secondary momentum is performed:

[0070]

[0071] In the formula, represents the corrected momentum; represents the kth power of β1, where k represents the current number of iterations; represents the modified secondary momentum; represents β2 to the power of k;

[0072] Finally update the model parameters:

[0073]

[0074] In the formula, θ t represents the updated model parameters; α represents the learning rate; ε represents a constant;

[0075] A wool fineness measuring system based on deep learning is used to execute a wool fineness measuring method based on deep learning, which comprises a high-definition camera, a preprocessing module, a feature extraction module, a convolutional neural network model and a display storage module connected in sequence, wherein the resolution of the high-definition camera is not less than 8 million pixels and is used to collect wool fiber images; the preprocessing module is used to preprocess the wool fiber images; the feature extraction module is used to extract the fineness features of the wool fiber images; the convolutional neural network model is used to predict the fineness of the wool fibers; and the display storage module is used to display and store the results.

[0076] Therefore, the present invention adopts the above-mentioned wool fineness determination method and system based on deep learning, which has the following beneficial effects:

[0077] 1. High precision: Using high-resolution cameras to collect wool fiber images ensures the quality and accuracy of the data;

[0078] Combining SIFT feature extraction algorithm, Canny edge detection algorithm and gray-level co-occurrence matrix, it is possible to extract multiple features of wool fibers and improve the accuracy of fineness measurement.

[0079] 2. High degree of automation: The entire process from image acquisition to feature extraction, model training, verification, prediction and result display is fully automated, which greatly reduces manual intervention and improves efficiency;

[0080] 3. High data utilization: The data set is reasonably divided into training set, validation set and test set, ensuring the generalization ability and reliability of the model;

[0081] By manually labeling the training set, the accuracy and reliability of the training data are guaranteed;

[0082] 4. Strong model robustness: The convolutional neural network (CNN) model can effectively capture the complex features in wool fiber images and has strong robustness and adaptability;

[0083] The model was verified and evaluated through cross-validation to ensure the stability and reliability of the model;

[0084] 5. Results visualization and storage:

[0085] The fineness evaluation report is displayed through a visual interface, allowing users to intuitively understand the fineness of wool fibers;

[0086] The fineness assessment report and the corresponding wool fiber image are stored in the database to facilitate subsequent data management and analysis.

[0087] In summary, compared with the experimental method, the present invention can reduce human errors, reduce measurement costs, and improve measurement efficiency. It is suitable for large-scale wool fineness detection scenarios and has broad application prospects in wool processing, textile manufacturing and other fields, thereby providing strong support for quality control and production management in related industries.

[0088] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 This is a flow chart of a wool fineness determination method based on deep learning of the present invention. DETAILED DESCRIPTION

[0090] In the description of the present invention, it should be noted that the terms "upper", "lower", "inside", "outside", etc. indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, or the positions or positional relationships in which the invented product is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In the description of the present invention, it should also be noted that, unless otherwise clearly specified and limited, the terms "setting", "installation", and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0091] The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0092] like Figure 1As shown, a wool fineness determination method based on deep learning comprises the following steps:

[0093] S1. Use a high-definition camera with a resolution of not less than 8 million pixels to collect wool fiber images to obtain a data set, and divide the data set into a training set, a validation set, and a test set according to the proportion, and manually annotate the wool fiber images in the training set;

[0094] S2, preprocessing the wool fiber image collected in step S1;

[0095] The preprocessing described in step S2 includes denoising (using a Gaussian filter to denoise and reduce interference information in the image), color correction (ensuring image color consistency), enhancement processing (highlighting the detailed features of wool), and cropping and scaling (standardizing image size).

[0096] S3, based on SIFT feature extraction algorithm combined with Canny algorithm edge detection algorithm and gray-level co-occurrence matrix, feature extraction is performed on the pre-processed wool fiber image to obtain the fineness characteristics of the wool;

[0097] Step S3 specifically includes the following steps:

[0098] S31. Feature extraction using SIFT feature extraction algorithm:

[0099] Descriptor=SIFT(I)Descriptor=SIFT(I)

[0100] Where Descriptor represents the feature descriptor set extracted from the wool fiber image I; SIFT(I) represents the feature extraction of the wool fiber image I using the SIFT feature extraction algorithm;

[0101] Step S31 specifically includes the following steps:

[0102] S311. Calculate the scale space of wool fiber image:

[0103] L(x,y,σ)=G(x,y,σ)×I(x,y)

[0104] Where L(x, y, σ) represents the scale space image at scale σ; G(x, y, σ) is the Gaussian kernel function; (x, y) represents the spatial coordinates of the wool fiber image I, that is, the position of the pixel;

[0105] S312, search for key points by differential Gaussian:

[0106] D(x,y,σ)=L(x,y,kσ)-L(x,y,σ)

[0107] Where D(x,y,σ) represents the difference Gaussian image; L(x,y,kσ) represents the scale space image at scale kσ, and k is a constant;

[0108] S313: Determine the direction and descriptor of the key point.

[0109] S32, using the Canny algorithm edge detection algorithm to detect the edges in the wool fiber image I, extracting the contour and boundary information of the wool fiber to determine the shape and size of the wool fiber;

[0110] Step S32 specifically includes the following steps:

[0111] S321, calculating the gradient amplitude and direction of the wool fiber image I to determine the strength and direction of the edge:

[0112]

[0113] Where Gx and Gy represent the first-order derivatives in the x-direction and y-direction respectively;

[0114] in,

[0115]

[0116] S322, calculate edge strength G:

[0117]

[0118] S323, non-maximum suppression: at each pixel, compare its gradient amplitude with the gradient amplitudes of its two adjacent pixel points in the gradient direction, and if the gradient amplitude of the current pixel is not the maximum, set it to 0 to remove non-maximum values;

[0119] S324, dual threshold processing: set two thresholds: T low and T high , and T low <T high , the gradient amplitude is greater than T high The pixel points with gradient amplitude between T low and T high The pixels between are marked as weak edges, and the weak edges are connected to the strong edges to form continuous edges, thereby obtaining the contour and boundary information of the wool fiber.

[0120] S33, extracting the texture characteristics of the wool fiber using the gray level co-occurrence matrix to analyze the surface roughness and smoothness of the wool fiber, thereby obtaining the structural and morphological information of the wool fiber;

[0121] Step S33 specifically includes the following steps:

[0122] S331, set the selected relative position to (dx, dy) to construct a gray-level co-occurrence matrix (for example, (dx, dy) = (1, 0), indicating a horizontally adjacent pixel pair, that is, the pixel to the right of the current pixel):

[0123] P(i,j)=∑(x,y)∈I

[0124] δ(I(x,y)=i,I(x+dx,y+dy)=j)

[0125] Wherein, P(i,j) represents the element in the gray-level co-occurrence matrix, that is, the number of occurrences of the pixel with gray value i and the pixel with gray value j at its relative position (dx,dy); I(x,y) represents the gray value of the wool fiber image I at the coordinate (x,y), that is, the gray value of the current pixel; I(x+dx,y+dy) represents the gray value of the pixel at the relative position (dx,dy) to the current pixel; δ(I(x,y)=i,I(x+dx,y+dy)=j) represents the indicator function, which is used to judge whether the gray value of the current pixel is i and whether the gray value of the pixel at its relative position (dx,dy) is j;

[0126] S332, calculating texture features:

[0127]

[0128] Entropy=-∑i,j P(i,j)log(P(i,j))

[0129] Wherein, Contrast represents contrast, which is used to reflect the intensity of local grayscale changes in wool fiber images; Entropy represents entropy, which is used to reflect the uncertainty of grayscale distribution in wool fiber images.

[0130] S34, normalize the features extracted in step S31, step S32 and step S33.

[0131] The normalization processing expression described in step S34 is as follows:

[0132]

[0133] In the formula, X′ represents the normalized eigenvalue; X represents the original eigenvalue; μ represents the mean value of the feature; σ1 represents the standard deviation of the feature;

[0134] in,

[0135]

[0136] Where N is the number of eigenvalues; x i Represents the eigenvalue.

[0137] S4, construct a convolutional neural network model, input the training set into the convolutional neural network model for training, and input the validation set into the trained convolutional neural network model, and use cross-validation to verify and evaluate the convolutional neural network model; use cross-validation and validation sets to verify and evaluate the model to ensure the generalization ability and stability of the model. And the cross-validation described in this embodiment is: divide the data set into K folds, train the model on each fold and calculate the error, and finally take the average as the cross-validation result.

[0138] In step S4, the convolutional neural network model includes three convolutional layers, two pooling layers and one fully connected layer arranged in sequence to perform the following convolution operation on the input wool fiber image I and the convolution kernel K:

[0139]

[0140] In the formula, (I*K)(x,y) represents the result of the convolution operation, that is, the convolution output value at the coordinate (x,y); (m,n) represents the index of the convolution kernel; I(m,n) represents the pixel value of the input wool fiber image I at the position (m,n); K(xm,yn) represents the weight value of the convolution kernel K at the position (xm,yn);

[0141] Assume that the size of the wool fiber image I is 5×5, K is a 3×3 convolution kernel, and the output size after convolution is 3×3.

[0142] The convolutional neural network model uses a cross entropy loss function, and the cross entropy loss function L is expressed as follows:

[0143]

[0144] In the formula, yi represents the true label; Represents the predicted value.

[0145] Preferably, the convolutional neural network model uses the Adam optimizer optimization algorithm to adjust the model parameters, and the specific steps are as follows:

[0146] First calculate the initial momentum and secondary momentum:

[0147] m0=0,v0=0

[0148] In the formula, m0 represents the initialization momentum; v0 represents the initialization secondary momentum;

[0149] Then calculate the gradient:

[0150]

[0151] In the formula, g trepresents the gradient of the current time step t; θ t-1 represents the model parameters at the previous time step;

[0152] Update the momentum and quadratic momentum again:

[0153] m t =β1m t -1+(1-β1)g t

[0154] v t =β2v t -1+(1-β2)g t 2

[0155] In the formula, m t represents the updated momentum; β1 represents the momentum decay rate; v t represents the updated secondary momentum; β2 represents the secondary momentum decay rate;

[0156] Then the deviation correction of momentum and secondary momentum is performed:

[0157]

[0158] In the formula, represents the corrected momentum; represents the kth power of β1, where k represents the current number of iterations; represents the modified secondary momentum; represents β2 to the power of k;

[0159] Finally update the model parameters:

[0160]

[0161] In the formula, θ t represents the updated model parameters; α represents the learning rate; ε represents a constant;

[0162] S5, inputting the test set into the convolutional neural network model verified in step S4, using the convolutional neural network model to predict the fineness of the wool fiber, and outputting a fineness evaluation report;

[0163] S6. Result display and storage: The fineness evaluation report is displayed to the user through a visual interface (charts and text descriptions), and the fineness evaluation report and the corresponding wool fiber image are stored in a database (the stored data includes the wool fiber image ID, measurement date, measurement results and original wool fiber image file) for subsequent query and analysis.

[0164] A wool fineness measuring system based on deep learning is used to execute a wool fineness measuring method based on deep learning, which comprises a high-definition camera, a preprocessing module, a feature extraction module, a convolutional neural network model and a display storage module connected in sequence, wherein the resolution of the high-definition camera is not less than 8 million pixels and is used to collect wool fiber images; the preprocessing module is used to preprocess the wool fiber images; the feature extraction module is used to extract the fineness features of the wool fiber images; the convolutional neural network model is used to predict the fineness of the wool fibers; and the display storage module is used to display and store the results.

[0165] Simulation experiment

[0166] Wool fibers were collected from the 10 cm rear edge of the scapula of a living sheep or from wool samples cut at 0.5 cm from the skin, and impurities such as feces on the wool fibers were removed. The wool fiber images were captured using a high-definition camera with a built-in LED light source and a magnification of 200 times. More than 2000 wool fiber images from different wool samples were collected, and the wool fiber images were processed using the method of the present invention to obtain a predicted value of fineness. At the same time, a traditional wool fineness determination method was used, and the fineness value of each wool sample was measured using a microscope after being pressed into a sheet. The prediction results of the present invention were compared with the values ​​determined by the experimental method, and the root mean square error (RMSE) and the mean absolute percentage error (MAPE) were used as prediction error indicators, and the Pearson correlation coefficient and the Spearman correlation coefficient were used as accuracy indicators. The results obtained are shown in Table 1.

[0167] Table 1 Statistical indicators

[0168] RMSE(%) MAPE(%) Pearson correlation Spearman correlation 0.3812 10.08 0.9215 0.9296

[0169] As can be seen from Table 1, the root mean square error (RMSE) and the mean absolute percentage error (MAPE) of the determination method of the present invention and the test method are both within an acceptable range, and the accuracy of the present invention is proved by the Pearson correlation coefficient (Pearson) and the Spearman correlation coefficient (Spearman).

[0170] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A wool fineness determination method based on deep learning, characterized in that: The following steps are involved: S1. Use a high-definition camera with a resolution of not less than 8 million pixels to collect wool fiber images to obtain a data set, and divide the data set into a training set, a validation set, and a test set according to the proportion, and manually annotate the wool fiber images in the training set; S2, preprocessing the wool fiber image collected in step S1; S3, based on SIFT feature extraction algorithm combined with Canny algorithm edge detection algorithm and gray-level co-occurrence matrix, feature extraction is performed on the pre-processed wool fiber image to obtain the fineness characteristics of the wool; S4, constructing a convolutional neural network model, inputting the training set into the convolutional neural network model for training, inputting the validation set into the trained convolutional neural network model, and using cross-validation to validate and evaluate the convolutional neural network model; S5, inputting the test set into the convolutional neural network model verified in step S4, using the convolutional neural network model to predict the fineness of the wool fiber, and outputting a fineness evaluation report; S6. Result display and storage: The fineness evaluation report is displayed to the user through a visual interface, and the fineness evaluation report and the corresponding wool fiber image are stored in a database.

2. A wool fineness determination method based on deep learning according to claim 1, characterized in that: The preprocessing described in step S2 includes denoising, color correction, enhancement processing, and cropping and scaling.

3. A wool fineness determination method based on deep learning according to claim 2, characterized in that: Step S3 specifically includes the following steps: S31. Feature extraction using SIFT feature extraction algorithm: Descriptor=SIFT(I)Descriptor=SIFT(I) Where Descriptor represents the feature descriptor set extracted from the wool fiber image I; SIFT(I) represents the feature extraction of the wool fiber image I using the SIFT feature extraction algorithm; S32, using the Canny algorithm edge detection algorithm to detect the edges in the wool fiber image I, extracting the contour and boundary information of the wool fiber to determine the shape and size of the wool fiber; S33, extracting the texture characteristics of the wool fiber using the gray level co-occurrence matrix to analyze the surface roughness and smoothness of the wool fiber, thereby obtaining the structural and morphological information of the wool fiber; S34, normalize the features extracted in step S31, step S32 and step S33.

4. The wool fineness determination method based on deep learning according to claim 3, characterized in that: Step S31 specifically includes the following steps: S311. Calculate the scale space of wool fiber image: L(x,y,σ)=G(x,y,σ)×I(x,y) Where L(x, y, σ) represents the scale space image at scale σ; G(z, y, σ) is the Gaussian kernel function; (x, y) represents the spatial coordinates of the wool fiber image I, that is, the position of the pixel; S312, search for key points by differential Gaussian: D(x,y,σ)=L(x,y,kσ)-L(x,y,σ) Where D(x,y,σ) represents the difference Gaussian image; L(x,y,kσ) represents the scale space image at scale kσ, and k is a constant; S313: Determine the direction and descriptor of the key point.

5. The wool fineness determination method based on deep learning according to claim 4, characterized in that: Step S32 specifically includes the following steps: S321, calculating the gradient amplitude and direction of the wool fiber image I to determine the strength and direction of the edge: Where Gx and Gy represent the first-order derivatives in the x-direction and y-direction respectively; in, S322, calculate edge strength G: S323, non-maximum suppression: at each pixel, compare its gradient amplitude with the gradient amplitudes of its two adjacent pixel points in the gradient direction, and if the gradient amplitude of the current pixel is not the maximum, set it to 0 to remove non-maximum values; S324, dual threshold processing: set two thresholds: T low and T high , and T low <T high , the gradient amplitude is greater than T high The pixel points with gradient amplitude between T low and T high The pixels between are marked as weak edges, and the weak edges are connected to the strong edges to form continuous edges, thereby obtaining the contour and boundary information of the wool fiber.

6. The method for measuring wool fineness based on deep learning according to claim 5, characterized in that: Step S33 specifically includes the following steps: S331, set the selected relative position to (dx, dy) to construct the gray-level co-occurrence matrix: P(i,j)=∑(x,y)∈I δ(I(x,y)=i,I(x+dx,y+dy)=j) Wherein, P(i,j) represents the element in the gray-level co-occurrence matrix, that is, the number of occurrences of the pixel with gray value i and the pixel with gray value j at its relative position (dx,dy); I(x,y) represents the gray value of the wool fiber image I at the coordinate (x,y), that is, the gray value of the current pixel; I(x+dx,y+dy) represents the gray value of the pixel at the relative position (dx,dy) to the current pixel; δ(I(x,y)=i,I(x+dx,y+dy)=j) represents the indicator function, which is used to judge whether the gray value of the current pixel is i and whether the gray value of the pixel at its relative position (dx,dy) is j; S332, calculating texture features: Entropy=-∑i,jP(i,j)log(P(i,j)) Wherein, Contrast represents contrast, which is used to reflect the intensity of local grayscale changes in wool fiber images; Entropy represents entropy, which is used to reflect the uncertainty of grayscale distribution in wool fiber images.

7. The method for measuring wool fineness based on deep learning according to claim 6, characterized in that: The normalization processing expression described in step S34 is as follows: In the formula, X′ represents the normalized eigenvalue; X represents the original eigenvalue; μ represents the mean value of the feature; σ1 represents the standard deviation of the feature; in, Where N is the number of eigenvalues; x i Represents the eigenvalue.

8. The method for measuring wool fineness based on deep learning according to claim 7, characterized in that: In step S4, the convolutional neural network model includes three convolutional layers, two pooling layers and one fully connected layer arranged in sequence to perform the following convolution operation on the input wool fiber image I and the convolution kernel K: Where, (I*K)(x,y) represents the result of the convolution operation, that is, the convolution output value at the coordinate (x,y); (m,n) represents the index of the convolution kernel; I(m,n) represents the pixel value of the input wool fiber image I at the position (m,n); K(xm,yn) represents the weight value of the convolution kernel K at the position (xm,yn); The convolutional neural network model uses a cross entropy loss function, and the cross entropy loss function L is expressed as follows: In the formula, yi represents the true label; Represents the predicted value.

9. A wool fineness determination method based on deep learning according to claim 8, characterized in that: The convolutional neural network model uses the Adam optimizer optimization algorithm to adjust the model parameters. The specific steps are as follows: First calculate the initial momentum and secondary momentum: m0=0,v0=0 In the formula, m0 represents the initialization momentum; v0 represents the initialization secondary momentum; Then calculate the gradient: In the formula, g t represents the gradient of the current time step t; θ t-1 represents the model parameters at the previous time step; Update the momentum and quadratic momentum again: m t =β1m t -1+(1-β1)g t v t =β2v t -1+(1-β2)g t 2 In the formula, m t represents the updated momentum; β1 represents the momentum decay rate; v t represents the updated secondary momentum; β2 represents the secondary momentum decay rate; Then the deviation correction of momentum and secondary momentum is performed: In the formula, represents the corrected momentum; represents the kth power of β1, where k represents the current number of iterations; represents the modified secondary momentum; represents β2 to the power of k; Finally update the model parameters: In the formula, θ t represents the updated model parameters; α represents the learning rate; ε represents a constant.

10. A wool fineness measurement system based on deep learning, characterized in that: A method for measuring wool fineness based on deep learning, used to execute any one of claims 1 to 9, comprising a high-definition camera, a preprocessing module, a feature extraction module, a convolutional neural network model and a display storage module connected in sequence, wherein the resolution of the high-definition camera is not less than 8 million pixels and is used to collect wool fiber images; the preprocessing module is used to preprocess the wool fiber images; the feature extraction module is used to extract the fineness features of the wool fiber images; the convolutional neural network model is used to predict the fineness of the wool fibers; and the display storage module is used to display and store the results.

Images