Sand and gravel image segmentation method, device, equipment and medium based on semantic segmentation

Through the semantic segmentation method, the semantic segmentation model is trained to perform pixel-level segmentation on sand and gravel images, solving the problem of low accuracy in sand and gravel images in the prior art, and achieving efficient segmentation effect in complex situations.

CN115035291BActive Publication Date: 2024-12-03JIANGHAN UNIVERSITY
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210383550.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-12
Publication Date
2024-12-03
Estimated Expiration
2042-04-12

AI Technical Summary

Technical Problem

The prior art cannot effectively perform pixel-level segmentation of concrete sand and gravel images, especially due to discontinuous changes in grayscale values, complex texture characteristics and inconspicuous grayscale contrast, it is difficult for traditional methods to accurately segment sand and gravel with extremely high similarity.

Method used

Using a semantic segmentation method, by acquiring and preprocessing sand and gravel images, dividing them into training sets and test sets, annotating the images in the training sets, and using feature extraction modules and semantic mapping modules to train the semantic segmentation model to realize the segmentation Pixel-level segmentation of the image.

Benefits of technology

提高了砂石图像的分割精确度,能够在灰度值变化不连续和纹理复杂的情况下实现像素级分割,解决了传统方法无法精确分割的问题,具有鲁棒性和高分割粒度。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115035291B_ABST
    Figure CN115035291B_ABST
Patent Text Reader

Abstract

The embodiment of the present application discloses a method, device, equipment and medium for sand and gravel image segmentation based on semantic segmentation. The method includes: obtaining multiple sand and gravel images; preprocessing the sand and gravel images to obtain processed sand and gravel images, the preprocessing includes image enhancement processing and filtering denoising processing; dividing the processed sand and gravel images into a training set and a test set; respectively annotating the sand images and stone images in the sand and gravel images in the training set to obtain an annotated training set; training the model parameters of a preset semantic segmentation model according to the annotated training set to obtain a trained semantic segmentation model; performing simulation tests on the trained semantic segmentation model according to the test set to obtain a training effect; if the segmentation accuracy in the training effect is greater than a preset value, performing sand and gravel segmentation processing on the sand and gravel images to be segmented according to the trained semantic segmentation model. In this solution, the accuracy of sand and gravel image segmentation can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a sand and gravel image segmentation method, device, equipment and medium based on semantic segmentation. Background Art

[0002] For the discontinuous grayscale value changes of sand and gravel mixed images, complex texture features, and unclear grayscale contrast, whether it is threshold segmentation based on color features (Hue Saturation Value, HSV), segmentation based on local binary pattern (Local Binary Pattern, LBP) features, or feature segmentation based on gray-level co-occurrence matrix (Gray-level co-occurrence matrix), it is difficult to accurately segment concrete sand and gravel with extremely high similarity at the pixel level.

[0003] In order to improve the accuracy of sand and gravel image segmentation, a method is urgently needed to accurately segment concrete sand and gravel images. Summary of the invention

[0004] The embodiments of the present application provide a method, device, equipment and medium for sand and gravel image segmentation based on semantic segmentation, which can perform pixel-level segmentation on concrete sand and gravel images and improve the accuracy of sand and gravel image segmentation.

[0005] In a first aspect, an embodiment of the present application provides a sand and gravel image segmentation method based on semantic segmentation, which includes: acquiring a plurality of sand and gravel images;

[0006] Preprocessing the sandstone image to obtain a plurality of processed sandstone images, wherein the preprocessing includes image enhancement processing and filtering and denoising processing;

[0007] Dividing the processed sandstone images into a training set and a test set;

[0008] The sand images and the stone images in the sand and gravel images in the training set are respectively labeled to obtain a labeled training set;

[0009] Training model parameters of a preset semantic segmentation model according to the labeled training set to obtain a trained semantic segmentation model;

[0010] Performing a simulation test on the trained semantic segmentation model according to the test set to obtain a training effect;

[0011] If the segmentation accuracy in the training result is greater than a preset value, the gravel image to be segmented is processed by gravel segmentation according to the trained semantic segmentation model.

[0012] In some embodiments, the preset semantic segmentation model includes a feature extraction module and a semantic mapping module, the feature extraction module includes multiple feature extraction units, each feature extraction unit is composed of a first convolutional layer, a pooling layer and a regularization layer, and the semantic mapping module is composed of an upsampling layer, a second convolutional layer and a SoftMax classification layer.

[0013] In some embodiments, training model parameters of a preset semantic segmentation model according to the labeled training set to obtain a trained semantic segmentation model includes:

[0014] Inputting the labeled sandstone sample images in the training set into the preset semantic segmentation model;

[0015] The first convolution layer, the pooling layer and the regularization layer sequentially perform linear filtering processing, maximum pooling processing and batch regularization processing on the sandstone sample image to obtain semantic information of the sandstone sample image;

[0016] The semantic information is sequentially subjected to upsampling processing, deconvolution processing, and normalized probability calculation processing through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain a prediction result of the sandstone sample image;

[0017] Determine a cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image;

[0018] The preset semantic segmentation model is back-propagated according to the cross entropy loss value to obtain a trained semantic segmentation model.

[0019] In some embodiments, the first convolution layer, the pooling layer, and the regularization layer sequentially perform linear filtering, maximum pooling, and batch regularization on the sandstone sample image to obtain semantic information of the sandstone sample image, including:

[0020] The sandstone sample image is sequentially subjected to linear filtering processing, maximum pooling processing and batch regularization processing by using the linear filtering formula preset in the first convolutional layer, the maximum pooling formula preset in the pooling layer and the batch regularization formula preset in the regularization layer to obtain semantic information of the sandstone sample image, wherein:

[0021] The linear filtering formula is:

[0022]

[0023] Among them, f(x+s,y+t) is the image matrix for filtering, w(s,t) is the convolution kernel, and g(x,y) is the output image matrix;

[0024] The maximum pooling formula is:

[0025]

[0026] Among them, m, n are the dimensions of the maximum pooling matrix, is the feature matrix after maximum pooling;

[0027] In the regularization layer, the mean μ and variance σ of each batch are calculated along the channel. 2 , normalize, and add the scaling and translation amounts γ and β, and batch regularization can be calculated as:

[0028]

[0029]

[0030]

[0031]

[0032] Among them, m is the number of batches, z i Represents the propagation value after batch normalization.

[0033] In some embodiments, the semantic information is sequentially upsampled, deconvolved, and processed by the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain a prediction result of the sandstone sample image, including:

[0034] Upsampling the feature image corresponding to the semantic information through the upsampling layer to obtain an upsampled feature image;

[0035] Performing deconvolution processing on the upsampled feature image by using the linear filtering formula in the second convolution layer to obtain a deconvolved feature image;

[0036] Performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result;

[0037] The result corresponding to the probability with the largest value in the probability results is determined as the predicted result.

[0038] In some embodiments, performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result includes:

[0039] The probability calculation result is obtained by performing normalized probability calculation on the deconvolved feature image through the probability calculation formula in the SoftMax classification layer, wherein the probability calculation formula is:

[0040]

[0041] Among them, k represents the number of categories output by the neural network, v is the output vector, vj is the jth output value in v, i represents the category that needs to be calculated currently, the calculation result is between 0 and 1, and the sum of the softmax values ​​of all categories is 1.

[0042] In some embodiments, determining the cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image includes:

[0043] The cross entropy loss value is determined according to a preset cross entropy loss calculation formula, the prediction result, and the annotation corresponding to the sandstone sample image, wherein the cross entropy loss calculation formula is:

[0044]

[0045] Among them, x represents the sample and n represents the total number of samples.

[0046] In a second aspect, the embodiment of the present application further provides a sandstone image segmentation device based on semantic segmentation, which includes: an acquisition unit and a processing unit, wherein:

[0047] The acquisition unit is used to acquire a plurality of sandstone images;

[0048] The processing unit is used to preprocess the sand and gravel images to obtain multiple processed sand and gravel images, wherein the preprocessing includes image enhancement processing and filtering denoising processing; divide the multiple processed sand and gravel images into a training set and a test set; respectively annotate the sand images and stone images in the sand and gravel images in the training set to obtain an annotated training set; train the model parameters of a preset semantic segmentation model according to the annotated training set to obtain a trained semantic segmentation model; simulate and test the trained semantic segmentation model according to the test set to obtain a training effect; if the segmentation accuracy in the training effect is greater than a preset value, perform sand and gravel segmentation processing on the sand and gravel images to be segmented according to the trained semantic segmentation model.

[0049] In a third aspect, an embodiment of the present application further provides a computer device, which includes a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0050] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the storage medium stores a computer program, wherein the computer program includes program instructions, and wherein the program instructions can implement the above method when executed by a processor.

[0051] The embodiment of the present application provides a method, device, equipment and medium for sand and gravel image segmentation based on semantic segmentation. The method includes: obtaining multiple sand and gravel images; preprocessing the sand and gravel images to obtain multiple processed sand and gravel images, wherein the preprocessing includes image enhancement processing and filtering denoising processing; dividing the multiple processed sand and gravel images into a training set and a test set; respectively annotating the sand images and stone images in the sand and gravel images in the training set to obtain an annotated training set; training the model parameters of a preset semantic segmentation model according to the annotated training set to obtain a trained semantic segmentation model; performing a simulation test on the trained semantic segmentation model according to the test set to obtain a training effect; if the segmentation accuracy in the training effect is greater than a preset value, performing sand and gravel segmentation processing on the sand and gravel images to be segmented according to the trained semantic segmentation model. In this solution, the sand and gravel images can be segmented at the pixel level by the trained semantic segmentation model, thereby improving the accuracy of sand and gravel image segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A schematic diagram of an application scenario of a sandstone image segmentation method based on semantic segmentation provided in an embodiment of the present application;

[0054] Figure 2 A schematic diagram of a flow chart of a sandstone image segmentation method based on semantic segmentation provided in an embodiment of the present application;

[0055] Figure 3 A structural diagram of a feature extraction module in a semantic segmentation model of an embodiment of the present application;

[0056] Figure 4 A schematic diagram of a structure of a semantic segmentation model according to an embodiment of the present application;

[0057] Figure 5 A schematic diagram of the algorithm fitting effect of the sand and gravel image segmentation method based on semantic segmentation provided in an embodiment of the present application;

[0058] Figure 6A schematic block diagram of a sandstone image segmentation device based on semantic segmentation provided in an embodiment of the present application;

[0059] Figure 7 A schematic block diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0061] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0062] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0063] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0064] The embodiments of the present application provide a sand and gravel image segmentation method, device, equipment and medium based on semantic segmentation.

[0065] The executor of the sand and gravel image segmentation method based on semantic segmentation can be the sand and gravel image segmentation device based on semantic segmentation provided in the embodiment of the present application, or a computer device integrated with the sand and gravel image segmentation device based on semantic segmentation, wherein the sand and gravel image segmentation device based on semantic segmentation can be implemented in hardware or software, the computer device can be a terminal or a server, and the terminal can be a smart phone, a tablet computer, a PDA, or a laptop computer, etc.

[0066] See also Figure 1 , Figure 1 Schematic diagram of the application scenario of the sand and gravel image segmentation method based on semantic segmentation provided in the embodiment of the present application. The sand and gravel image segmentation method based on semantic segmentation is applied to Figure 1 In the computer device 10 in the method, the computer device 10 obtains a plurality of sand and gravel images; pre-processes the sand and gravel images to obtain a plurality of processed sand and gravel images, wherein the pre-processing includes image enhancement processing and filtering denoising processing; divides the plurality of processed sand and gravel images into a training set and a test set; respectively annotates the sand images and stone images in the sand and gravel images in the training set to obtain an annotated training set; trains the model parameters of a preset semantic segmentation model according to the annotated training set to obtain a trained semantic segmentation model; performs a simulation test on the trained semantic segmentation model according to the test set to obtain a training effect; if the segmentation accuracy in the training effect is greater than a preset value, sand and gravel segmentation processing is performed on the sand and gravel images to be segmented according to the trained semantic segmentation model.

[0067] Figure 2 is a flow chart of a sandstone image segmentation method based on semantic segmentation provided in an embodiment of the present application. Figure 2 As shown, the method includes the following steps S110-150.

[0068] S110, acquiring a plurality of sandstone images.

[0069] Specifically, the sandstone image may be an image collected in advance and stored in a local memory.

[0070] S120, preprocessing the sandstone image to obtain a plurality of processed sandstone images, wherein the preprocessing includes image enhancement processing and filtering and denoising processing.

[0071] Specifically, due to the randomness of the collected images, it is difficult to ensure that the acquired sandstone images are all good samples, so it is necessary to perform image enhancement processing on the sandstone images with weak signals. In some embodiments, the enhancement can be performed by gamma correction defined by the following formula:

[0072] s=cr γ ; (1)

[0073] Where γ is the correction scale, c is the constant coefficient, r is the input pixel value, and s is the output pixel value.

[0074] For the collected images, reasonable denoising is also a way to improve the model fit. The linear spatial filtering defined by the following formula is used for filtering and denoising:

[0075]

[0076] Where f(x+s,y+t) is the image matrix for filtering, w(s,t) is the convolution kernel, and g(x,y) is the output image matrix.

[0077] S130, dividing the plurality of processed sandstone images into a training set and a test set.

[0078] In this embodiment, after the processed sandstone image is obtained, the sandstone image is divided into two parts, one is a training set and the other is a test set.

[0079] S140 , labeling the sand images and the stone images in the sand and gravel images in the training set respectively to obtain a labeled training set.

[0080] After the sand and gravel images are segmented, the preset image annotation software is used to annotate the sand pixel images in each sand and gravel image in the training set as sand and the stone pixel images in the sand and gravel images as stone.

[0081] S150 , training model parameters of a preset semantic segmentation model according to the annotated training set to obtain a trained semantic segmentation model.

[0082] Specifically, step S150 includes the following steps:

[0083] S151, inputting the labeled sandstone sample images in the training set into the preset semantic segmentation model.

[0084] Among them, the preset semantic segmentation model includes a feature extraction module encoder and a semantic mapping module decoder. The encoder part uses VGG-16 as the backbone network model to perform high-level semantic extraction on the features of the sandstone image. The decoder part will parse the semantics to map them to the corresponding image pixels, that is, each pixel is represented by the color (or label) corresponding to its object. The feature extraction module includes multiple feature extraction units, each of which is composed of a first convolutional layer, a pooling layer, and a regularization layer. The semantic mapping module is composed of an upsampling layer, a second convolutional layer, and a SoftMax classification layer.

[0085] In some embodiments, Figure 3 As shown, Figure 3 is the structural diagram of the feature extraction module. The feature extraction module includes 5 feature extraction units, such as Figure 4 As shown, Figure 4 This is the structural diagram of the semantic segmentation model. The semantic mapping module consists of 5 semantic mapping units. The first 4 semantic mapping units include an upsampling layer and a second convolutional layer. The fifth semantic mapping unit includes an upsampling layer, a second convolutional layer, and a SoftMax classification layer.

[0086] S152. Perform linear filtering, maximum pooling, and batch regularization processing on the sand and gravel sample image in sequence through the first convolution layer, the pooling layer, and the regularization layer to obtain semantic information of the sand and gravel sample image.

[0087] Each convolution layer is a back propagation parameter to be optimized in the model. The convolution is performed according to the linear filtering formula (2). The pooling layer uses maximum pooling to enhance the robustness of the model and prevent overfitting. The maximum pooling is obtained according to the maximum pooling formula (3).

[0088] Specifically, step S152 includes: performing linear filtering processing, maximum pooling processing, and batch regularization processing on the sandstone sample image in sequence through the linear filtering formula preset in the first convolutional layer, the maximum pooling formula preset in the pooling layer, and the batch regularization formula preset in the regularization layer to obtain the semantic information of the sandstone sample image, wherein:

[0089] The maximum pooling formula is:

[0090]

[0091] Among them, m, n are the dimensions of the maximum pooling matrix, is the feature matrix after maximum pooling;

[0092] In the regularization layer, the mean μ and variance σ of each batch are calculated along the channel. 2 , normalize, and add the scaling and translation amounts γ and β, and batch regularization can be calculated as:

[0093]

[0094]

[0095]

[0096]

[0097] Among them, m is the number of batches, z i Represents the propagation value after batch normalization.

[0098] S153, performing upsampling processing, deconvolution processing, and normalized probability calculation processing on the semantic information in sequence through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain a prediction result of the sandstone sample image.

[0099] After obtaining the semantic information obtained by the encoder, it is necessary to map this semantic information to the corresponding pixel points. First, the decoder upsamples the reduced feature image through the upsampling layer, and then the second convolution layer applies formula (2) to perform deconvolution processing to improve the geometric shape and compensate for the detail loss of the object in the pooling layer. The mapped semantics are then sent to the SoftMax classification layer to calculate the normalized probability.

[0100] Specifically, step S153 includes: upsampling the feature image corresponding to the semantic information through the upsampling layer to obtain the upsampled feature image; deconvolving the upsampled feature image through the linear filtering formula in the second convolution layer to obtain the deconvolved feature image; performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result; and determining the result corresponding to the probability with the largest median value of the probability results as the prediction result.

[0101] Among them, the step of performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain the probability calculation result specifically includes: performing normalized probability calculation on the deconvolved feature image through the probability calculation formula in the SoftMax classification layer to obtain the probability calculation result, wherein the probability calculation formula is:

[0102]

[0103] Among them, k represents the number of categories output by the neural network, v is the output vector, vj is the jth output value in v, i represents the category that needs to be calculated currently, the calculation result is between 0 and 1, and the sum of the softmax values ​​of all categories is 1.

[0104] As shown in the following formula, the maximum value of the calculated probability is taken as the prediction result:

[0105] s=max(p i ); (9)

[0106] S154. Determine a cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image.

[0107] Compare the predicted results with the pre-marked label values ​​and optimize the cross entropy loss function. The cross entropy loss value is obtained by the following formula:

[0108]

[0109] Among them, x represents the sample and n represents the total number of samples.

[0110] S155. Back-propagating the preset semantic segmentation model according to the cross entropy loss value to obtain a trained semantic segmentation model.

[0111] After obtaining the cross entropy loss value, back propagation is performed according to the gradient calculated by the back propagation formula, where the back propagation formula is:

[0112]

[0113] S160: Perform a simulation test on the trained semantic segmentation model according to the test set to obtain a training effect.

[0114] Specifically, the trained model parameters in the trained semantic segmentation model are loaded, and the model is tested on the test set to obtain the training effect.

[0115] S170: If the segmentation accuracy in the training result is greater than a preset value, performing sand and gravel segmentation processing on the sand and gravel image to be segmented according to the trained semantic segmentation model.

[0116] That is, whether the semantic segmentation model is trained is determined based on the training effect. If the segmentation accuracy in the training effect is greater than the preset value, it means that the training is completed. Subsequently, the trained semantic segmentation model is used to perform sand and gravel segmentation processing on the gravel image to be segmented.

[0117] If the segmentation accuracy in the training effect is not greater than the preset value, it means that the training is not completed, and the process returns to execute steps S150-S170 until the segmentation accuracy in the training effect is greater than the preset value.

[0118] Among them, the final algorithm fitting effect Figure 5 shown.

[0119] Semantic segmentation in this application is one of the basic research fields of computer vision. Different from the problem of target detection, the problem of semantic segmentation research is more sophisticated. It is necessary not only to detect the object category, but also to segment the image category at the pixel level. Therefore, it is suitable for particle-level detection in concrete aggregates and sand and gravel. This application proposes to use the method of semantic segmentation to solve the problem of sand and gravel image segmentation. The whole is composed of a feature extractor encoder and a decoder decoder that extract the features of sand and gravel images. The encoder part uses VGG-16 as the backbone network model to extract high-level semantics of the features of the sand and gravel images. The decoder part will parse the semantics to map to the corresponding image pixels, that is, each pixel is represented by the color (or label) corresponding to its object. The algorithm downsamples a large number of pre-trained sand and gravel image samples, classifies and analyzes the low-level local pixel information of the image, and obtains the high-order semantic information of sand and stone, thereby achieving the effect of pixel-level segmentation.

[0120] In summary, the present application acquires a plurality of sand and gravel images; pre-processes the sand and gravel images to obtain a plurality of processed sand and gravel images, wherein the pre-processing includes image enhancement processing and filtering and denoising processing; divides the plurality of processed sand and gravel images into a training set and a test set; respectively labels the sand images and stone images in the sand and gravel images in the training set to obtain a labeled training set; trains the model parameters of a preset semantic segmentation model according to the labeled training set to obtain a trained semantic segmentation model; performs simulation tests on the trained semantic segmentation model according to the test set to obtain a training effect; if the segmentation accuracy in the training effect is greater than a preset value, sand and gravel segmentation processing is performed on the sand and gravel images to be segmented according to the trained semantic segmentation model. In this solution, the sand and gravel images can be segmented at the pixel level by the trained semantic segmentation model, thereby improving the accuracy of sand and gravel image segmentation.

[0121] In addition, the sand and gravel image segmentation method based on semantic segmentation proposed in this application has the advantages of strong robustness, fine segmentation granularity, and good segmentation effect. It solves the problem that the grayscale value of the mixed sand and gravel image is discontinuous, the texture features are complex, and the grayscale contrast is not obvious. The traditional threshold segmentation based on color features, the segmentation based on local binary pattern features, and the feature segmentation based on grayscale co-occurrence matrix are all unable to accurately segment the concrete sand and gravel with extremely high similarity at the pixel level. It is of great significance to the detection and segmentation of concrete particle size in the construction industry.

[0122] Figure 6 is a schematic block diagram of a sandstone image segmentation device based on semantic segmentation provided in an embodiment of the present application. Figure 6 As shown, corresponding to the above sand and gravel image segmentation method based on semantic segmentation, the present application also provides a sand and gravel image segmentation device based on semantic segmentation. The sand and gravel image segmentation device based on semantic segmentation includes a unit for executing the above sand and gravel image segmentation method based on semantic segmentation, and the device can be configured in a terminal such as a desktop computer, a tablet computer, a laptop computer or a sand and gravel image acquisition device. Specifically, please refer to Figure 6 The sandstone image segmentation device based on semantic segmentation includes an acquisition unit 601 and a processing unit 602, wherein:

[0123] The acquisition unit 601 is used to acquire multiple sandstone images;

[0124] The processing unit 602 is used to preprocess the sand and gravel images to obtain multiple processed sand and gravel images, wherein the preprocessing includes image enhancement processing and filtering denoising processing; divide the multiple processed sand and gravel images into a training set and a test set; respectively annotate the sand images and stone images in the sand and gravel images in the training set to obtain an annotated training set; train the model parameters of a preset semantic segmentation model according to the annotated training set to obtain a trained semantic segmentation model; simulate and test the trained semantic segmentation model according to the test set to obtain a training effect; if the segmentation accuracy in the training effect is greater than a preset value, perform sand and gravel segmentation processing on the sand and gravel images to be segmented according to the trained semantic segmentation model.

[0125] In some embodiments, the preset semantic segmentation model includes a feature extraction module and a semantic mapping module, the feature extraction module includes multiple feature extraction units, each feature extraction unit is composed of a first convolutional layer, a pooling layer and a regularization layer, and the semantic mapping module is composed of an upsampling layer, a second convolutional layer and a SoftMax classification layer.

[0126] In some embodiments, when the processing unit 602 performs the step of training the model parameters of the preset semantic segmentation model according to the annotated training set to obtain the trained semantic segmentation model, it is specifically used to:

[0127] Inputting the labeled sandstone sample images in the training set into the preset semantic segmentation model;

[0128] The first convolution layer, the pooling layer and the regularization layer sequentially perform linear filtering processing, maximum pooling processing and batch regularization processing on the sandstone sample image to obtain semantic information of the sandstone sample image;

[0129] The semantic information is sequentially subjected to upsampling processing, deconvolution processing, and normalized probability calculation processing through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain a prediction result of the sandstone sample image;

[0130] Determine a cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image;

[0131] The preset semantic segmentation model is back-propagated according to the cross entropy loss value to obtain a trained semantic segmentation model.

[0132] In some embodiments, when the processing unit 602 performs the step of performing linear filtering, maximum pooling, and batch regularization processing on the sandstone sample image through the first convolution layer, the pooling layer, and the regularization layer in sequence to obtain the semantic information of the sandstone sample image, it is specifically used to:

[0133] The sandstone sample image is sequentially subjected to linear filtering processing, maximum pooling processing and batch regularization processing by using the linear filtering formula preset in the first convolutional layer, the maximum pooling formula preset in the pooling layer and the batch regularization formula preset in the regularization layer to obtain semantic information of the sandstone sample image, wherein:

[0134] The linear filtering formula is:

[0135]

[0136] Among them, f(x+s,y+t) is the image matrix for filtering, w(s,t) is the convolution kernel, and g(x,y) is the output image matrix;

[0137] The maximum pooling formula is:

[0138]

[0139] Among them, m, n are the dimensions of the maximum pooling matrix, is the feature matrix after maximum pooling;

[0140] In the regularization layer, the mean μ and variance σ of each batch are calculated along the channel. 2 , normalize, and add the scaling and translation amounts γ and β, and batch regularization can be calculated as:

[0141]

[0142]

[0143]

[0144]

[0145] Among them, m is the number of batches, z i Represents the propagation value after batch normalization.

[0146] In some embodiments, when the processing unit 602 performs the step of performing upsampling processing, deconvolution processing, and normalized probability calculation processing on the semantic information in sequence through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain the prediction result of the sandstone sample image, it is specifically used to:

[0147] Upsampling the feature image corresponding to the semantic information through the upsampling layer to obtain an upsampled feature image;

[0148] Performing deconvolution processing on the upsampled feature image by using the linear filtering formula in the second convolution layer to obtain a deconvolved feature image;

[0149] Performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result;

[0150] The result corresponding to the probability with the largest value in the probability results is determined as the predicted result.

[0151] In some embodiments, when the processing unit 602 performs the step of performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result, it is specifically used to:

[0152] The probability calculation result is obtained by performing normalized probability calculation on the deconvolved feature image through the probability calculation formula in the SoftMax classification layer, wherein the probability calculation formula is:

[0153]

[0154] Among them, k represents the number of categories output by the neural network, v is the output vector, vj is the jth output value in v, i represents the category that needs to be calculated currently, the calculation result is between 0 and 1, and the sum of the softmax values ​​of all categories is 1.

[0155] In some embodiments, when the processing unit 602 performs the step of determining the cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image, it is specifically used to:

[0156] The cross entropy loss value is determined according to a preset cross entropy loss calculation formula, the prediction result, and the annotation corresponding to the sandstone sample image, wherein the cross entropy loss calculation formula is:

[0157]

[0158] Among them, x represents the sample and n represents the total number of samples.

[0159] It should be noted that technicians in the relevant field can clearly understand that the specific implementation process of the above-mentioned sand and gravel image segmentation device based on semantic segmentation and each unit can refer to the corresponding description in the aforementioned method embodiment. For the convenience and conciseness of the description, it will not be repeated here.

[0160] The above-mentioned sandstone image segmentation device based on semantic segmentation can be implemented in the form of a computer program. The computer program can be used in Figure 7 Runs on the computer device shown.

[0161] See also Figure 7 , Figure 7 700 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 700 may be a terminal or a server, wherein the terminal may be an electronic device with communication functions such as a smart phone, a tablet computer, a laptop computer, a desktop computer, a personal digital assistant, and a wearable device. The server may be an independent server or a server cluster composed of multiple servers.

[0162] See also Figure 7 The computer device 700 includes a processor 702 , a memory and a network interface 705 connected via a system bus 701 , wherein the memory may include a non-volatile storage medium 703 and an internal memory 704 .

[0163] The non-volatile storage medium 703 can store an operating system 7031 and a computer program 7032. The computer program 7032 includes program instructions, and when the program instructions are executed, the processor 702 can execute a sandstone image segmentation method based on semantic segmentation.

[0164] The processor 702 is used to provide computing and control capabilities to support the operation of the entire computer device 700 .

[0165] The internal memory 704 provides an environment for the operation of the computer program 7032 in the non-volatile storage medium 703. When the computer program 7032 is executed by the processor 702, the processor 702 can execute a sand and gravel image segmentation method based on semantic segmentation.

[0166] The network interface 705 is used to communicate with other devices over the network. Figure 7 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 700 to which the solution of the present application is applied. The specific computer device 700 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0167] The processor 702 is used to run the computer program 7032 stored in the memory to implement the following steps:

[0168] Acquire multiple sandstone images;

[0169] Preprocessing the sandstone image to obtain a plurality of processed sandstone images, wherein the preprocessing includes image enhancement processing and filtering and denoising processing;

[0170] Dividing the processed sandstone images into a training set and a test set;

[0171] The sand images and the stone images in the sand and gravel images in the training set are respectively labeled to obtain a labeled training set;

[0172] Training model parameters of a preset semantic segmentation model according to the labeled training set to obtain a trained semantic segmentation model;

[0173] Performing a simulation test on the trained semantic segmentation model according to the test set to obtain a training effect;

[0174] If the segmentation accuracy in the training result is greater than a preset value, the gravel image to be segmented is processed by gravel segmentation according to the trained semantic segmentation model.

[0175] In some embodiments, the preset semantic segmentation model includes a feature extraction module and a semantic mapping module, the feature extraction module includes multiple feature extraction units, each feature extraction unit is composed of a first convolutional layer, a pooling layer and a regularization layer, and the semantic mapping module is composed of an upsampling layer, a second convolutional layer and a SoftMax classification layer.

[0176] In some embodiments, when the processor 702 implements the step of training the model parameters of the preset semantic segmentation model according to the labeled training set to obtain the trained semantic segmentation model, the processor 702 specifically implements the following steps:

[0177] Inputting the labeled sandstone sample images in the training set into the preset semantic segmentation model;

[0178] The first convolution layer, the pooling layer and the regularization layer sequentially perform linear filtering processing, maximum pooling processing and batch regularization processing on the sandstone sample image to obtain semantic information of the sandstone sample image;

[0179] The semantic information is sequentially subjected to upsampling processing, deconvolution processing, and normalized probability calculation processing through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain a prediction result of the sandstone sample image;

[0180] Determine a cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image;

[0181] The preset semantic segmentation model is back-propagated according to the cross entropy loss value to obtain a trained semantic segmentation model.

[0182] In some embodiments, when the processor 702 implements the step of performing linear filtering, maximum pooling, and batch regularization processing on the sandstone sample image through the first convolution layer, the pooling layer, and the regularization layer in sequence to obtain the semantic information of the sandstone sample image, the following steps are specifically implemented:

[0183] The sandstone sample image is sequentially subjected to linear filtering processing, maximum pooling processing and batch regularization processing by using the linear filtering formula preset in the first convolutional layer, the maximum pooling formula preset in the pooling layer and the batch regularization formula preset in the regularization layer to obtain semantic information of the sandstone sample image, wherein:

[0184] The linear filtering formula is:

[0185]

[0186] Among them, f(x+s,y+t) is the image matrix for filtering, w(s,t) is the convolution kernel, and g(x,y) is the output image matrix;

[0187] The maximum pooling formula is:

[0188]

[0189] Among them, m, n are the dimensions of the maximum pooling matrix, is the feature matrix after maximum pooling;

[0190] In the regularization layer, the mean μ and variance σ of each batch are calculated along the channel. 2 , normalize, and add the scaling and translation amounts γ and β, and batch regularization can be calculated as:

[0191]

[0192]

[0193]

[0194]

[0195] Among them, m is the number of batches, z i Represents the propagation value after batch normalization.

[0196] In some embodiments, when the processor 702 implements the step of performing upsampling processing, deconvolution processing, and normalized probability calculation processing on the semantic information in sequence through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain the prediction result of the sandstone sample image, the following steps are specifically implemented:

[0197] Upsampling the feature image corresponding to the semantic information through the upsampling layer to obtain an upsampled feature image;

[0198] Performing deconvolution processing on the upsampled feature image by using the linear filtering formula in the second convolution layer to obtain a deconvolved feature image;

[0199] Performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result;

[0200] The result corresponding to the probability with the largest value in the probability results is determined as the predicted result.

[0201] In some embodiments, when the processor 702 performs normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result, the following steps are specifically implemented:

[0202] The probability calculation result is obtained by performing normalized probability calculation on the deconvolved feature image through the probability calculation formula in the SoftMax classification layer, wherein the probability calculation formula is:

[0203]

[0204] Among them, k represents the number of categories output by the neural network, v is the output vector, vj is the jth output value in v, i represents the category that needs to be calculated currently, the calculation result is between 0 and 1, and the sum of the softmax values ​​of all categories is 1.

[0205] In some embodiments, when the processor 702 implements the step of determining the cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image, the processor 702 specifically implements the following steps:

[0206] The cross entropy loss value is determined according to a preset cross entropy loss calculation formula, the prediction result, and the annotation corresponding to the sandstone sample image, wherein the cross entropy loss calculation formula is:

[0207]

[0208] Among them, x represents the sample and n represents the total number of samples.

[0209] It should be understood that in the embodiment of the present application, the processor 702 may be a central processing unit (CPU), and the processor 702 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0210] It can be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0211] Therefore, the present application also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the following steps:

[0212] Acquire multiple sandstone images;

[0213] Preprocessing the sandstone image to obtain a plurality of processed sandstone images, wherein the preprocessing includes image enhancement processing and filtering and denoising processing;

[0214] Dividing the processed sandstone images into a training set and a test set;

[0215] The sand images and the stone images in the sand and gravel images in the training set are respectively labeled to obtain a labeled training set;

[0216] Training model parameters of a preset semantic segmentation model according to the labeled training set to obtain a trained semantic segmentation model;

[0217] Performing a simulation test on the trained semantic segmentation model according to the test set to obtain a training effect;

[0218] If the segmentation accuracy in the training result is greater than a preset value, the gravel image to be segmented is processed by gravel segmentation according to the trained semantic segmentation model.

[0219] In some embodiments, the preset semantic segmentation model includes a feature extraction module and a semantic mapping module, the feature extraction module includes multiple feature extraction units, each feature extraction unit is composed of a first convolutional layer, a pooling layer and a regularization layer, and the semantic mapping module is composed of an upsampling layer, a second convolutional layer and a SoftMax classification layer.

[0220] In some embodiments, when the processor executes the program instructions to implement the step of training the model parameters of the preset semantic segmentation model according to the labeled training set to obtain the trained semantic segmentation model, the following steps are specifically implemented:

[0221] Inputting the labeled sandstone sample images in the training set into the preset semantic segmentation model;

[0222] The first convolution layer, the pooling layer and the regularization layer sequentially perform linear filtering processing, maximum pooling processing and batch regularization processing on the sandstone sample image to obtain semantic information of the sandstone sample image;

[0223] The semantic information is sequentially subjected to upsampling processing, deconvolution processing, and normalized probability calculation processing through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain a prediction result of the sandstone sample image;

[0224] Determine a cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image;

[0225] The preset semantic segmentation model is back-propagated according to the cross entropy loss value to obtain a trained semantic segmentation model.

[0226] In some embodiments, when the processor executes the program instructions to implement the step of performing linear filtering, maximum pooling and batch regularization processing on the sandstone sample image in sequence through the first convolution layer, the pooling layer and the regularization layer to obtain the semantic information of the sandstone sample image, the following steps are specifically implemented:

[0227] The sandstone sample image is sequentially subjected to linear filtering processing, maximum pooling processing and batch regularization processing by using the linear filtering formula preset in the first convolutional layer, the maximum pooling formula preset in the pooling layer and the batch regularization formula preset in the regularization layer to obtain semantic information of the sandstone sample image, wherein:

[0228] The linear filtering formula is:

[0229]

[0230] Among them, f(x+s,y+t) is the image matrix for filtering, w(s,t) is the convolution kernel, and g(x,y) is the output image matrix;

[0231] The maximum pooling formula is:

[0232]

[0233] Among them, m, n are the dimensions of the maximum pooling matrix, is the feature matrix after maximum pooling;

[0234] In the regularization layer, the mean μ and variance σ of each batch are calculated along the channel. 2 , normalize, and add the scaling and translation amounts γ and β, and batch regularization can be calculated as:

[0235]

[0236]

[0237]

[0238]

[0239] Among them, m is the number of batches, z i Represents the propagation value after batch normalization.

[0240] In some embodiments, when the processor executes the program instructions to implement the step of performing upsampling processing, deconvolution processing, and normalized probability calculation processing on the semantic information in sequence through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain the prediction result of the sandstone sample image, the following steps are specifically implemented:

[0241] Upsampling the feature image corresponding to the semantic information through the upsampling layer to obtain an upsampled feature image;

[0242] Performing deconvolution processing on the upsampled feature image by using the linear filtering formula in the second convolution layer to obtain a deconvolved feature image;

[0243] Performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result;

[0244] The result corresponding to the probability with the largest value in the probability results is determined as the predicted result.

[0245] In some embodiments, when the processor executes the program instructions to implement the step of performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result, the following steps are specifically implemented:

[0246] The probability calculation result is obtained by performing normalized probability calculation on the deconvolved feature image through the probability calculation formula in the SoftMax classification layer, wherein the probability calculation formula is:

[0247]

[0248] Among them, k represents the number of categories output by the neural network, v is the output vector, vj is the jth output value in v, i represents the category that needs to be calculated currently, the calculation result is between 0 and 1, and the sum of the softmax values ​​of all categories is 1.

[0249] In some embodiments, when the processor executes the program instructions to implement the step of determining the cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image, the processor specifically implements the following steps:

[0250] The cross entropy loss value is determined according to a preset cross entropy loss calculation formula, the prediction result, and the annotation corresponding to the sandstone sample image, wherein the cross entropy loss calculation formula is:

[0251]

[0252] Among them, x represents the sample and n represents the total number of samples.

[0253] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, or any other computer-readable storage medium that can store program codes.

[0254] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0255] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0256] The steps in the method of the embodiment of the present application can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present application can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0257] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application.

[0258] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A sandstone image segmentation method based on semantic segmentation, characterized in that: include: Acquire multiple sandstone images; Preprocessing the sandstone image to obtain a plurality of processed sandstone images, wherein the preprocessing includes image enhancement processing and filtering and denoising processing; Dividing the processed sandstone images into a training set and a test set; The sand images and the stone images in the sand and gravel images in the training set are respectively labeled to obtain a labeled training set; Training model parameters of a preset semantic segmentation model according to the labeled training set to obtain a trained semantic segmentation model; Performing a simulation test on the trained semantic segmentation model according to the test set to obtain a training effect; If the segmentation accuracy in the training result is greater than a preset value, performing sand and gravel segmentation processing on the sand and gravel image to be segmented according to the trained semantic segmentation model; The preset semantic segmentation model includes a feature extraction module and a semantic mapping module, the feature extraction module includes a plurality of feature extraction units, each feature extraction unit is composed of a first convolution layer, a pooling layer and a regularization layer, and the semantic mapping module is composed of an upsampling layer, a second convolution layer and a SoftMax classification layer; The training of model parameters of a preset semantic segmentation model according to the labeled training set to obtain a trained semantic segmentation model includes: Inputting the labeled sandstone sample images in the training set into the preset semantic segmentation model; The first convolution layer, the pooling layer and the regularization layer sequentially perform linear filtering processing, maximum pooling processing and batch regularization processing on the sandstone sample image to obtain semantic information of the sandstone sample image; The semantic information is sequentially subjected to upsampling processing, deconvolution processing, and normalized probability calculation processing through the upsampling layer, the second convolution layer, and the SoftMax classification layer to obtain a prediction result of the sandstone sample image; Determine a cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image; Back-propagating the preset semantic segmentation model according to the cross entropy loss value to obtain a trained semantic segmentation model; The method sequentially performs linear filtering processing, maximum pooling processing and batch regularization processing on the sandstone sample image through the first convolution layer, the pooling layer and the regularization layer to obtain semantic information of the sandstone sample image, including: The sandstone sample image is sequentially subjected to linear filtering processing, maximum pooling processing and batch regularization processing by using the linear filtering formula preset in the first convolutional layer, the maximum pooling formula preset in the pooling layer and the batch regularization formula preset in the regularization layer to obtain semantic information of the sandstone sample image, wherein: The linear filtering formula is: Among them, f(x+s,y+t) is the image matrix for filtering, w(s,t) is the convolution kernel, and g(x,y) is the output image matrix; The maximum pooling formula is: Among them, m, n are the dimensions of the maximum pooling matrix, is the feature matrix after maximum pooling; In the regularization layer, the mean μ and variance σ of each batch are calculated along the channel. 2 , normalize, and add scaling and translation γ and β, batch regularization can be calculated: Among them, m is the number of batches, z i Represents the propagation value after batch normalization.

2. The method according to claim 1, characterized in that The upsampling layer, the second convolution layer and the SoftMax classification layer sequentially perform upsampling processing, deconvolution processing and normalized probability calculation processing on the semantic information to obtain the prediction result of the sandstone sample image, including: Upsampling the feature image corresponding to the semantic information through the upsampling layer to obtain an upsampled feature image; Performing deconvolution processing on the upsampled feature image by using a linear filtering formula in the second convolutional layer to obtain a deconvolved feature image; Performing normalized probability calculation on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result; The result corresponding to the probability with the largest value in the probability results is determined as the predicted result.

3. The method according to claim 2, characterized in that The normalized probability calculation is performed on the deconvolved feature image through the SoftMax classification layer to obtain a probability calculation result, including: The probability calculation result is obtained by performing normalized probability calculation on the deconvolved feature image through the probability calculation formula in the SoftMax classification layer, wherein the probability calculation formula is: Among them, k represents the number of categories output by the neural network, v is the output vector, vj is the jth output value in v, i represents the category that needs to be calculated currently, the calculation result is between 0 and 1, and the sum of the softmax values ​​of all categories is 1.

4. The method according to claim 1, characterized in that: The determining the cross entropy loss value according to the prediction result and the annotation corresponding to the sandstone sample image includes: The cross entropy loss value is determined according to a preset cross entropy loss calculation formula, the prediction result, and the annotation corresponding to the sandstone sample image, wherein the cross entropy loss calculation formula is: Among them, x represents the sample and n represents the total number of samples.

5. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 4 when executing the computer program.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the method according to any one of claims 1 to 4 can be implemented.

Citation Information

Patent Citations

  • Semantic image segmentation model training method, readable storage medium and electronic equipment

    CN108596184A

  • Aero-engine borescope image damage intelligent segmentation method based on context coding network

    CN110895814A