Non-destructive determination of backfat thickness, loin area and intramuscular fat content in pigs

Abstract

The present invention discloses a non-destructive measurement method for the backfat thickness, eye muscle area and intramuscular fat content of a pig. First, a B-ultrasound image of the pig is obtained; the B-ultrasound image of the pig is optimized and data enhanced, and input into an image segmentation model, and is segmented into a backfat region mask, an eye muscle region mask and a backfat and eye muscle region; the backfat region mask area and perimeter, and the eye muscle area are calculated using Green's formula, and the backfat thickness of the pig is determined according to the backfat region mask area and perimeter; the backfat region image in the backfat and eye muscle region image is removed to obtain an eye muscle region image; the eye muscle region image is subjected to color histogram equalization to extract grayscale-gradient co-occurrence matrix parameters; the backfat thickness, eye muscle area and grayscale-gradient co-occurrence matrix parameters of the pig are input into a recurrent neural network prediction model for prediction to obtain intramuscular fat content. Based on the non-destructive measurement method of the present invention, the method of slaughtering pigs in the past is no longer necessary.

Description

Technical Field

[0001] The invention relates to the technical field of pig breeding and farming, in particular to a non-destructive determination method of pig backfat thickness, eye muscle area and intramuscular fat content. Background Art

[0002] Backfat thickness, loin eye area, and intramuscular fat content are important factors affecting pig carcass quality. Backfat thickness can increase the number of piglets born alive and litter weight. Loin eye area, as a growth indicator, can be used to select high-quality breeds with fast growth and high meat yield, and is an important indicator for measuring pig carcass quality and fattening results. Intramuscular fat content determines the taste of the meat. Lower intramuscular fat content results in less tenderness, juiciness, and flavor. However, higher intramuscular fat content can have adverse effects on human health.

[0003] Traditional testing methods are unable to quickly, accurately, and non-destructively measure backfat thickness, muscle eye area, and intramuscular fat content. With the development of ultrasonic technology, non-destructive testing has become a mainstream method. However, traditional ultrasonic equipment requires manual marking of the backfat and muscle eye areas. The measurement speed depends on the employee's experience, which is highly subjective and prone to large measurement errors. It is also impossible to measure intramuscular fat content.

[0004] Method content

[0005] The purpose of this application is to provide a non-destructive determination method for pig backfat thickness, eye muscle area and intramuscular fat content in order to address the technical defects in the prior art.

[0006] The technical solutions adopted to achieve the purpose of this application are:

[0007] A non-destructive method for measuring pig backfat thickness, eye muscle area and intramuscular fat content comprises the following steps:

[0008] Step 1, obtaining a B-ultrasound image of a pig;

[0009] Step 2: Optimize and data enhance the pig's B-ultrasound image obtained in step 1, input the data-enhanced pig's B-ultrasound image into a pre-trained image segmentation model, and perform region of interest segmentation based on the pre-trained image segmentation model, and segment the optimized pig's B-ultrasound image into a backfat region mask, an eye muscle region mask, and backfat and eye muscle regions;

[0010] Step 3, using Green's formula to calculate the mask area and perimeter of the backfat region and the eye muscle area in the eye muscle region mask image, and determining the backfat thickness of the pig based on the mask area and perimeter of the backfat region;

[0011] Step 4, removing the backfat area image from the backfat and eye muscle area images to obtain the eye muscle area image;

[0012] Step 5: Perform color histogram equalization on the eye muscle area image to extract the grayscale-gradient co-occurrence matrix parameters;

[0013] In step 6, the backfat thickness, eye muscle area and grayscale-gradient co-occurrence matrix parameters of the pig are input into the pre-trained recurrent neural network prediction model for prediction to obtain the intramuscular fat content.

[0014] In the above technical solution, the B-ultrasound image of the pig in step 1 is a transverse B-ultrasound image between the 10th and 11th ribs on the left side of the pig; the optimization of the B-ultrasound image of the pig obtained in step 1 includes marking the back fat area mask with green and the background with black; marking the eye muscle area mask with red and the background with black; and the back fat and eye muscle area cutout background is black.

[0015] In the above technical solution, step 2 includes the following steps:

[0016] S2.1: Cropping the pig ultrasound image obtained in step 1 into a 220x160 image, and optimizing the cropped image using wavelet denoising;

[0017] S2.2: Data enhancement is performed on the optimized pig B-ultrasound images using rotation, translation, horizontal flipping, brightening, darkening, salt and pepper noise, and logarithmic transformation methods;

[0018] S2.3: Build a deep learning image segmentation model, input the data-augmented pig B-ultrasound images into the deep learning model for training and testing, and obtain a trained deep learning image segmentation model;

[0019] S2.4: Use the trained image segmentation model to segment the data-enhanced pig B-ultrasound image into a backfat region mask image, an eye muscle region mask image, and backfat and eye muscle region images.

[0020] In the above technical solution, the deep learning image segmentation model is preferably a U-Net convolutional neural network model.

[0021] In the above technical solution, the training process of the U-Net convolutional neural network model includes:

[0022] S2.3.1: Input the data-augmented pig ultrasound images into the U-Net convolutional neural network model and perform four pooling operations to obtain feature images of different sizes: 256x256, 128x128, 64x64, and 32x32.

[0023] S2.3.2: Upsample the 32x32 feature image to obtain a 64x64 feature image;

[0024] S2.3.3: Concatenate the upsampled 64x64 feature image with the pooled 64x64 feature image on the channels, then convolve and upsample the concatenated feature image to obtain a 128x128 feature image.

[0025] S2.3.4: Concatenate the 128x128 feature image obtained by convolution and upsampling with the 128x128 feature image obtained after pooling. Then, convolve and upsample the concatenated feature image to obtain a 256x256 feature image.

[0026] S2.3.5: Concatenate the 256x256 feature image obtained by convolution and upsampling with the 256x256 feature image obtained after pooling. Then convolve and upsample the concatenated feature image to obtain a 512x512 feature image.

[0027] In the above technical solution, the U-Net convolutional neural network model modifies the data-augmented pig B-ultrasound image of size 220x160 to a data-augmented pig B-ultrasound image of size 512x512 before training, so the image size of the U-Net convolutional neural network model training input is 512x512.

[0028] In the above technical solution, step 3 includes the following steps:

[0029] S3.1: converting the backfat area mask image and the eye muscle area image into binary images;

[0030] S3.2: Perform pixel traversal on the binary images of the backfat region mask and the eye muscle region respectively, and store all contour points of the binary images of the backfat region mask and the eye muscle region when the position difference between two adjacent pixels does not exceed 1;

[0031] S3.3: Using Green's formula, synthesize all stored contour points and integrate the synthesized contour to obtain the mask area of ​​the backfat region and the eye muscle area in the mask image of the eye muscle region;

[0032] S3.4: Calculate the pixels of all stored contour points to obtain the perimeter of the backfat area mask;

[0033] S3.5: Based on the masked area and perimeter of the backfat area, calculate the width of the backfat area, which is the backfat thickness of the pig.

[0034] In the above technical solution, step 4 specifically includes the following steps:

[0035] S4.1: converting the backfat region mask image and the backfat and eye muscle region images into binary images;

[0036] S4.2: Perform pixel traversal on the backfat region mask image and the binary images of the backfat and eye muscle regions, respectively, and store all contour points when the distance between two adjacent pixel points does not exceed 1;

[0037] S4.3: Using Green's formula, synthesize all the stored contour points into a contour, and integrate the synthesized contour to obtain the mask area of ​​the backfat region, the backfat area and the muscleeye area in the backfat and muscleeye region images, and the area of ​​the uninterested region in the backfat and muscleeye region images;

[0038] S4.4: Compare the sizes of the backfat mask area, the backfat area and the muscle eye area in the backfat and muscle eye area images, and the area of ​​the uninterested region in the backfat and muscle eye area images;

[0039] S4.5: When the difference between the backfat area mask area and the backfat area and the muscle eye area in the backfat and muscle eye area images is less than 10, set the pixel points of the backfat area image in the backfat and muscle eye area images to black to obtain the muscle eye area image.

[0040] In the above technical solution, the extraction of grayscale-gradient co-occurrence matrix parameters includes extracting 15 parameters: small gradient advantage, large gradient advantage, grayscale entropy, gradient entropy, grayscale average, gradient average, grayscale distribution unevenness, gradient distribution unevenness, grayscale variance, gradient variance, energy, inverse moment, mixed entropy, correlation and inertia.

[0041] In the above technical solution, the training of the recurrent neural network prediction model includes the following steps:

[0042] S6.1: Build a recurrent neural network prediction model;

[0043] S6.2: Input the grayscale-gradient co-occurrence matrix parameters extracted in step 5, the eye muscle area in the eye muscle area mask image in step 3, and the pig's backfat thickness into the recurrent neural network prediction model for training to obtain a trained recurrent neural network prediction model.

[0044] The beneficial effects of the present invention are as follows:

[0045] 1. This application establishes an image segmentation model based on deep learning (U-Net convolutional neural network model), which can accurately segment the backfat area mask, eye muscle area mask and backfat and eye muscle area images based on pig B-ultrasound images, with a segmentation accuracy of 98%. At the same time, a calculation algorithm (Green's formula) for pig backfat thickness and eye muscle area was established, with an average standard error of 0.19 and high measurement accuracy. This allows the calculation of pig backfat thickness and eye muscle area to break away from the level of manual subjective experience, greatly improving the universality of detection.

[0046] 2. This application also establishes a recurrent neural network prediction model, which can measure the intramuscular fat content based on data extracted from pig B-ultrasound images. The measurement accuracy reaches 73%, and the measurement precision is high. The determination of intramuscular fat content can enable pigs to be measured quickly and accurately using B-ultrasound images, getting rid of the previous destructive methods such as slaughtering pigs. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only 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.

[0048] Figure 1 Shown is a flow chart of the non-destructive determination method of pig backfat thickness, eye muscle area and intramuscular fat content of the present invention.

[0049] Figure 2 Shown is a schematic diagram of a pig's B-ultrasound image according to the present invention.

[0050] Figure 3 Shown is a schematic diagram of the U-Net neural network model structure of the present invention.

[0051] Figure 4 Shown is a schematic diagram of the backfat area mask of the present invention.

[0052] Figure 5 Shown is a schematic diagram of the eye muscle area mask of the present invention.

[0053] Figure 6 Shown is a schematic diagram of the back fat and eye muscle area of ​​the present invention.

[0054] Figure 7 Shown is a schematic diagram of the eye muscle area of ​​the present invention.

[0055] Figure 8 Shown is a schematic diagram of a binary image of the backfat area mask of the present invention.

[0056] Figure 9 Shown is a schematic diagram of a binary image of the eye muscle area of ​​the present invention.

[0057] Figure 10 Shown is a schematic diagram of the structure of the recurrent neural network prediction model of the present invention. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the present method, the technical solution of the present method is further described below with reference to specific embodiments.

[0059] See also Figure 1 A non-destructive method for measuring pig backfat thickness, eye muscle area and intramuscular fat content comprises the following steps:

[0060] Step 1: Obtain a pig's B-ultrasound image; wherein the pig's B-ultrasound image can be found in Figure 2 .

[0061] Furthermore, the B-ultrasound image of the pig in step 1 is a transverse B-ultrasound image between the 10th and 11th ribs on the left side of the pig. When acquiring the B-ultrasound image of the pig, the hair between the 10th and 11th ribs on the left side of the pig is removed, and a coupling agent or cooking oil is evenly applied. The probe is held perpendicular to the pig's body with uniform force. When a clear and bright field of view is obtained, the B-ultrasound image is captured transversely.

[0062] Step 2: Optimize and enhance the pig's B-ultrasound image obtained in step 1, input the data-enhanced pig's B-ultrasound image into a pre-trained image segmentation model, and perform region of interest segmentation based on the pre-trained image segmentation model, and segment the optimized pig's B-ultrasound image into a back fat region mask, an eye muscle region mask, and back fat and eye muscle regions.

[0063] Furthermore, the optimization of the B-ultrasound image of the pig obtained in step 1 includes marking the backfat area mask with green and a black background; marking the eye muscle area mask with red and a black background; and cutting out the backfat and eye muscle area with a black background.

[0064] Furthermore, the step 2 includes the following steps:

[0065] S2.1: Cropping the pig ultrasound image obtained in step 1 into a 220x160 image, and optimizing the cropped image using wavelet denoising;

[0066] S2.2: Data enhancement is performed on the optimized pig B-ultrasound images using rotation, translation, horizontal flipping, brightening, darkening, salt and pepper noise, and logarithmic transformation methods;

[0067] S2.3: Build a deep learning image segmentation model, input the data-augmented pig B-ultrasound images into the deep learning model for training and testing, and obtain a trained deep learning image segmentation model;

[0068] Furthermore, the deep learning image segmentation model is preferably a U-Net convolutional neural network model.

[0069] For details, see Figure 3 , the training process of the U-Net convolutional neural network model includes:

[0070] S2.3.1: Input the data-augmented pig ultrasound images into the U-Net convolutional neural network model and perform four pooling operations to obtain feature images of different sizes: 256x256, 128x128, 64x64, and 32x32.

[0071] S2.3.2: Upsample the 32x32 feature image to obtain a 64x64 feature image;

[0072] S2.3.3: Concatenate the upsampled 64x64 feature image with the pooled 64x64 feature image on the channels, then convolve and upsample the concatenated feature image to obtain a 128x128 feature image.

[0073] S2.3.4: Concatenate the 128x128 feature image obtained by convolution and upsampling with the 128x128 feature image obtained after pooling. Then, convolve and upsample the concatenated feature image to obtain a 256x256 feature image.

[0074] S2.3.5: Concatenate the 256x256 feature image obtained by convolution and upsampling with the 256x256 feature image obtained after pooling. Then convolve and upsample the concatenated feature image to obtain a 512x512 feature image.

[0075] Furthermore, before training, the U-Net convolutional neural network model modifies the data-augmented pig B-ultrasound image of size 220x160 to a data-augmented pig B-ultrasound image of size 512x512, so the image size of the U-Net convolutional neural network model training input is 512x512.

[0076] S2.4: Using the trained image segmentation model, the pig B-ultrasound image after data augmentation is segmented into a backfat region mask image, an eye muscle region mask image, and a backfat and eye muscle region image. The backfat region mask image can be found in Figure 4 The eye muscle area mask image can be found in Figure 5, the back fat and eye muscle area images can be found in Figure 6 , the eye muscle area image can be found in Figure 7 .

[0077] Step 3: Calculate the mask area and perimeter of the backfat region and the muscle eye area in the muscle eye region mask image using Green's formula, and determine the pig's backfat thickness based on the mask area and perimeter of the backfat region. This specifically includes the following steps:

[0078] S3.1: converting the backfat area mask image and the eye muscle area image into binary images;

[0079] S3.2: Perform pixel traversal on the binary images of the backfat region mask and the eye muscle region respectively, and store all contour points of the binary images of the backfat region mask and the eye muscle region when the position difference between two adjacent pixels does not exceed 1;

[0080] S3.3: Using Green's formula, synthesize all stored contour points and integrate the synthesized contour to obtain the mask area of ​​the backfat region and the eye muscle area in the mask image of the eye muscle region;

[0081] Specifically, the specific concept of the Green's formula is to assume that D is a plane region. If the partial region enclosed by any closed curve in D belongs to D, then D is called a plane simply connected region. Intuitively speaking, a simply connected region is a region without space, otherwise it is called a complex connected region. The Green's formula theorem is to assume that the closed region D is enclosed by a piecewise smooth curve L, and the functions P(x, y) and Q(x, y) have first-order continuous partial derivatives on D. The specific expression of the Green's formula is as follows:

[0082]

[0083] Where D represents the closed region, Q represents , P represents , X represents , Y represents , x represents , y represents , and L represents the piecewise smooth curve that encloses the closed region. L is the positive boundary curve of D.

[0084] S3.4: Calculate the pixels of all stored contour points to obtain the perimeter of the backfat area mask;

[0085] S3.5: Based on the masked area and perimeter of the backfat area, calculate the width of the backfat area, which is the backfat thickness of the pig.

[0086] Step 4: remove the backfat area image from the backfat and eye muscle area images to obtain the eye muscle area image.

[0087] The step 4 specifically includes the following steps:

[0088] S4.1: Convert the backfat region mask image and the backfat and eye muscle region images into binary images, wherein the backfat region mask and the backfat and eye muscle region binary images are as follows: Figure 8 、 Figure 9 As shown;

[0089] S4.2: Perform pixel traversal on the backfat region mask image and the binary images of the backfat and eye muscle regions, respectively, and store all contour points when the distance between two adjacent pixel points does not exceed 1;

[0090] S4.3: Using Green's formula, synthesize all the stored contour points into a contour, and integrate the synthesized contour to obtain the mask area of ​​the backfat region, the backfat area and the muscleeye area in the backfat and muscleeye region images, and the area of ​​the uninterested region in the backfat and muscleeye region images;

[0091] S4.4: Compare the sizes of the backfat mask area, the backfat area and the muscle eye area in the backfat and muscle eye area images, and the area of ​​the uninterested region in the backfat and muscle eye area images;

[0092] S4.5: When the difference between the backfat area mask area and the backfat area and the muscle eye area in the backfat and muscle eye area images is less than 10, set the pixels of the backfat area image in the backfat and muscle eye area images to black to obtain the muscle eye area image, wherein, see Figure 7 , the non-black area is the eye muscle area image.

[0093] Step 5: Perform color histogram equalization on the eye muscle area image and extract the grayscale-gradient co-occurrence matrix parameters.

[0094] The grayscale-gradient co-occurrence matrix parameters are extracted, including extracting 15 parameters: small gradient advantage, large gradient advantage, grayscale entropy, gradient entropy, grayscale average, gradient average, grayscale distribution unevenness, gradient distribution unevenness, grayscale variance, gradient variance, energy, inverse moment, mixed entropy, correlation and inertia.

[0095] In step 6, the backfat thickness, eye muscle area and grayscale-gradient co-occurrence matrix parameters of the pig are input into the pre-trained recurrent neural network prediction model for prediction to obtain the intramuscular fat content.

[0096] The training of the recurrent neural network prediction model includes the following steps:

[0097] S6.1: Build a recurrent neural network prediction model;

[0098] S6.2: Input the grayscale-gradient co-occurrence matrix parameters extracted in step 5, the eye muscle area in the eye muscle area mask image in step 3, and the pig's backfat thickness into the recurrent neural network prediction model for training to obtain a trained recurrent neural network prediction model.

[0099] See also Figure 10 , using the recurrent neural network prediction model for training and testing specifically includes: first defining an RNNModel class as a model (the model consists of an RNN layer, two fully connected layers and a Dropout layer), in the forward method, passing the pig's backfat thickness, eye muscle area and grayscale-gradient co-occurrence matrix parameter inputs to the RNN layer, and taking out the output of the last time step; then passing the output to two fully connected layers, and finally using the Dropout layer to reduce overfitting.

[0100] Specifically, the predicted values, true values, and standard deviations of the backfat thickness and eye muscle area of ​​pigs determined based on Green's formula are shown in Table 1. The predicted values, true values, and standard deviations of the intramuscular fat content determined based on the trained recurrent neural network prediction model are shown in Table 2.

[0101] Table 1 Predicted and true values ​​of pig backfat thickness and eye muscle area, as well as standard deviations between the predicted and true values

[0102]

[0103]

[0104] Table 2 Predicted values, true values, and standard deviations between predicted and true values ​​of intramuscular fat content

[0105]

[0106]

[0107] The above description is a preferred embodiment of the present invention and should not be regarded as any form of limitation to the present invention. The preferred embodiments and related technologies in the present invention have been described above, but they do not limit the present invention. For relevant technical personnel familiar with this technical field, without departing from the scope of the technical principles of the present invention, the technical principles described above can be used to make some improvements or replacements, including modifications and equivalent changes based on the core content of the present invention, which are all within the scope of protection of the present invention.

Claims

1. A non-destructive method for measuring backfat thickness, eye muscle area and intramuscular fat content of pigs, characterized in that: The following steps are involved: Step 1, obtaining a B-ultrasound image of a pig; Step 2: Optimize and data enhance the pig's B-ultrasound image obtained in step 1, input the data-enhanced pig's B-ultrasound image into a pre-trained image segmentation model, and perform region of interest segmentation based on the pre-trained image segmentation model, and segment the optimized pig's B-ultrasound image into a backfat region mask, an eye muscle region mask, and backfat and eye muscle regions; Step 3, using Green's formula to calculate the mask area and perimeter of the backfat region and the eye muscle area in the eye muscle region mask image, and determining the backfat thickness of the pig based on the mask area and perimeter of the backfat region; The step 3 comprises the following steps: S3.1: converting the backfat area mask image and the eye muscle area image into binary images; S3.2: Perform pixel traversal on the binary images of the backfat region mask and the eye muscle region respectively, and store all contour points of the binary images of the backfat region mask and the eye muscle region when the position difference between two adjacent pixels does not exceed 1; S3.3: Using Green's formula to synthesize all stored contour points, and integrating the synthesized contour to obtain the mask area of ​​the backfat region and the eye muscle area in the mask image of the eye muscle region; S3.4: Calculate the pixels of all stored contour points to obtain the perimeter of the backfat area mask; S3.5: Calculate the width of the backfat area based on the masked area and perimeter of the backfat area, which is the backfat thickness of the pig; Step 4, removing the backfat area image from the backfat and eye muscle area images to obtain the eye muscle area image; The step 4 specifically includes the following steps: S4.1: converting the backfat region mask image and the backfat and eye muscle region images into binary images; S4.2: Perform pixel traversal on the backfat region mask image and the binary images of the backfat and eye muscle regions, respectively, and store all contour points when the distance between two adjacent pixel points does not exceed 1; S4.3: Using Green's formula, synthesize all the stored contour points into a contour, and integrate the synthesized contour to obtain the mask area of ​​the backfat region, the backfat area and the muscleeye area in the backfat and muscleeye region images, and the area of ​​the uninterested region in the backfat and muscleeye region images; S4.4: Compare the sizes of the backfat mask area, the backfat area and the muscle eye area in the backfat and muscle eye area images, and the area of ​​the uninterested region in the backfat and muscle eye area images; S4.5: When the difference between the backfat area mask area and the backfat area in the backfat and eye muscle area images is less than 10, set the pixels of the backfat area image in the backfat and eye muscle area images to black to obtain the eye muscle area image; Step 5: Perform color histogram equalization on the eye muscle area image to extract the grayscale-gradient co-occurrence matrix parameters; In step 6, the backfat thickness, eye muscle area and grayscale-gradient co-occurrence matrix parameters of the pig are input into the pre-trained recurrent neural network prediction model for prediction to obtain the intramuscular fat content.

2. The non-destructive measurement method according to claim 1, characterized in that The B-ultrasound image of the pig in step 1 is a transverse B-ultrasound image between the 10th and 11th ribs on the left side of the pig; the optimization of the B-ultrasound image of the pig obtained in step 1 includes marking the backfat area mask with green and the background with black; marking the eye muscle area mask with red and the background with black; and the backfat and eye muscle area cutout background is black.

3. The non-destructive measurement method according to claim 1, characterized in that: The step 2 comprises the following steps: S2.1: Cropping the pig ultrasound image obtained in step 1 into a 220x160 image, and optimizing the cropped image using wavelet denoising; S2.2: Data enhancement is performed on the optimized pig B-ultrasound images using rotation, translation, horizontal flipping, brightening, darkening, salt and pepper noise, and logarithmic transformation methods; S2.3: Build a deep learning image segmentation model, input the data-augmented pig B-ultrasound images into the deep learning model for training and testing, and obtain a trained deep learning image segmentation model; S2.4: Use the trained image segmentation model to segment the data-augmented pig B-ultrasound image into a backfat region mask image, an eye muscle region mask image, and backfat and eye muscle region images.

4. The non-destructive measurement method according to claim 1, characterized in that: The image segmentation model of deep learning is the U-Net convolutional neural network model.

5. The non-destructive measurement method according to claim 4, characterized in that: The training process of the U-Net convolutional neural network model includes: S2.3.1: Input the data-augmented pig ultrasound images into the U-Net convolutional neural network model and perform four pooling operations to obtain feature images of different sizes: 256x256, 128x128, 64x64, and 32x32. S2.3.2: Upsample the 32x32 feature image to obtain a 64x64 feature image; S2.3.3: Concatenate the upsampled 64x64 feature image with the pooled 64x64 feature image on the channels, then convolve and upsample the concatenated feature image to obtain a 128x128 feature image. S2.3.4: Concatenate the 128x128 feature image obtained by convolution and upsampling with the 128x128 feature image obtained after pooling. Then, convolve and upsample the concatenated feature image to obtain a 256x256 feature image. S2.3.5: Concatenate the 256x256 feature image obtained by convolution and upsampling with the 256x256 feature image obtained after pooling. Then convolve and upsample the concatenated feature image to obtain a 512x512 feature image.

6. The non-destructive measurement method according to claim 5, characterized in that: Before training, the U-Net convolutional neural network model modifies the data-augmented pig B-ultrasound image of size 220x160 to a data-augmented pig B-ultrasound image of size 512x512, so the image size of the U-Net convolutional neural network model training input is 512x512.

7. The non-destructive measurement method according to claim 1, characterized in that: The grayscale-gradient co-occurrence matrix parameters are extracted, including extracting 15 parameters: small gradient advantage, large gradient advantage, grayscale entropy, gradient entropy, grayscale average, gradient average, grayscale distribution unevenness, gradient distribution unevenness, grayscale variance, gradient variance, energy, inverse moment, mixed entropy, correlation and inertia.

8. The non-destructive measurement method according to claim 1, wherein: The training of the recurrent neural network prediction model includes the following steps: S6.1: Build a recurrent neural network prediction model; S6.2: Input the grayscale-gradient co-occurrence matrix parameters extracted in step 5, the eye muscle area in the eye muscle area mask image in step 3, and the pig's backfat thickness into the recurrent neural network prediction model for training to obtain a trained recurrent neural network prediction model.

Images