Broiler pore density estimation method and device

Abstract

The invention discloses a broiler pore density estimation method and device, and belongs to the technical field of poultry breeding. The method comprises the following steps: S1, obtaining broiler skin image data, and carrying out manual labeling and anti-interference processing on the image data to obtain an image data set; s2, constructing a pore prediction model, and training the pore prediction model by adopting the image data set to obtain a density estimation model; and S3, performing pore density estimation on the broiler chicken by adopting the density estimation model to obtain an estimation result. According to the method, the pore density of the live chicken is estimated by combining a visual acquisition technology and a deep learning technology. And a certain reference value can be provided for broiler chicken breeding and broiler chicken markets.

Description

Technical Field

[0001] The invention belongs to the technical field of livestock breeding, and in particular relates to a method and a device for estimating the pore density of broiler chickens. Background Art

[0002] The number of pores in broiler chickens involves many aspects, including the richness of poultry genetic resources, changes in market demand, limitations of traditional breeding methods, and the application of modern molecular biology techniques. Richness of poultry genetic resources China is one of the richest poultry genetic resources in the world. The scale of Guangxi's high-quality broiler industry ranks second in the country and is one of the most scaled and industrialized advantageous industries in Guangxi's animal husbandry. Guangxi has rich local chicken germplasm resources, such as Nandan Yao chicken, which have unique genetic characteristics and excellent meat quality. Among these chicken breeds, the number of pores, as an important appearance trait, has received more and more attention. Changes in market demand As the chilled chicken market gradually replaces the live poultry market, consumers have put forward higher requirements for the appearance quality of broilers and the appearance quality of carcasses. The number of pores, as an important indicator reflecting the appearance quality of carcasses, is directly related to the market competitiveness of broilers. Therefore, accurately estimating the number of pores in broilers is of great significance to meeting market demand and increasing the added value of broiler products. Summary of the invention

[0003] The purpose of the present invention is to propose a method and device for estimating the pore density of broiler chickens. By combining visual acquisition technology and deep learning technology, the pore density of live broiler chickens can be estimated by only partially removing feathers from the live broiler chickens.

[0004] To achieve the above object, the present invention provides the following scheme: a method for estimating pore density of broiler chickens, comprising the following steps:

[0005] S1. Acquire broiler skin image data, perform manual annotation and anti-interference processing on the image data, and obtain an image data set;

[0006] S2, constructing a pore prediction model, and using the image data set to train the pore prediction model to obtain a density estimation model;

[0007] S3. Use the density estimation model to estimate the pore density of broilers to obtain an estimation result.

[0008] Further preferably, S1 comprises the following steps:

[0009] S11, obtaining a complete skin image of a broiler chicken, and intercepting unit areas at different parts as the image data;

[0010] S12, performing anti-interference processing on the image data to obtain a processed image;

[0011] S13. Manually annotate the processed image to obtain an annotated image; the annotated image constitutes the image data set.

[0012] Further preferably, the density estimation model comprises: a pore number calculation module, a surface area calculation module and a density calculation module;

[0013] The pore number calculation module is used to calculate the number of pores per unit area;

[0014] The surface area calculation module is used to calculate the surface area of ​​the broiler;

[0015] The density calculation module is used to calculate the pore density based on the number of pores and the surface area to obtain the estimation result.

[0016] Further preferably, the pore number calculation module includes: a normalization unit, a feature extraction unit, an anchor point acquisition unit, a classification unit, a pooling unit, an adjustment unit and a counting unit;

[0017] The normalization unit is used to perform normalization processing on the input image to obtain a scaled image of the same size;

[0018] The feature extraction unit is used to extract features from the zoomed image to obtain a feature map of the zoomed image;

[0019] The anchor point acquisition unit is used to generate all initial anchor points based on the pixel points on the feature map;

[0020] The classification unit is used to regress and classify the initial anchor points generated by the anchor point acquisition unit to obtain proposal points;

[0021] The pooling unit is used to perform a pooling operation on the proposed points to obtain the proposed points of the same size;

[0022] The adjustment unit is used to perform an adjustment operation on the proposed point output by the pooling unit to obtain a detection frame of the proposed point;

[0023] The counting unit is used to count the detection frames to obtain the number of pores.

[0024] Further preferably, the surface area calculation module comprises: an image acquisition unit, a matching unit, a fusion reconstruction unit and a calculation unit;

[0025] The image acquisition unit is used to collect optical image data of broilers to obtain original images;

[0026] The matching unit is used to perform stereo matching on the original image to obtain a matching image;

[0027] The fusion reconstruction unit is used to fuse and register the matching images to obtain a broiler three-dimensional model;

[0028] The calculation unit is used to calculate the surface area of ​​the broiler based on the broiler three-dimensional model.

[0029] The present invention also provides a broiler pore density estimation device, comprising: a data set construction system, a model construction system and an estimation system;

[0030] The data set construction system is used to obtain broiler skin image data, perform manual annotation and anti-interference processing on the image data, and obtain an image data set;

[0031] The model building system is used to build a pore prediction model, and the image data set is used to train the pore prediction model to obtain a density estimation model;

[0032] The estimation system uses the density estimation model to estimate the pore density of broilers to obtain an estimation result.

[0033] Further preferably, the data set construction system comprises: a data set acquisition module, a processing module and a labeling module;

[0034] The data set acquisition module is used to obtain the complete skin image of the broiler chicken and intercept the unit area at different parts as the image data;

[0035] The processing module is used to perform anti-interference processing on the image data to obtain a processed image;

[0036] The annotation module is used to manually annotate the processed image to obtain an annotated image; the annotated image constitutes the image data set.

[0037] Further preferably, the density estimation model comprises: a pore number calculation module, a surface area calculation module and a density calculation module;

[0038] The pore number calculation module is used to calculate the number of pores per unit area;

[0039] The surface area calculation module is used to calculate the surface area of ​​the broiler;

[0040] The density calculation module is used to calculate the pore density based on the number of pores and the surface area to obtain the estimation result.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention estimates the pore density of live chickens by combining visual acquisition technology and deep learning technology, which can provide certain reference value for broiler breeding and the broiler market. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 The present invention is a schematic flow chart of a method for estimating pore density of broilers according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Embodiment 1:

[0048] like Figure 1 As shown, this embodiment provides a method for estimating pore density of broiler chickens, comprising the following steps:

[0049] S1. Acquire broiler skin image data, perform manual annotation and anti-interference processing on the image data, and obtain an image data set.

[0050] Specifically, S1 includes the following steps:

[0051] S11, obtaining a complete skin image of a broiler chicken, and capturing unit areas at different parts as the image data.

[0052] S12, performing anti-interference processing on the image data to obtain a processed image.

[0053] Anti-interference processing methods include:

[0054] 1. Pigment separation:

[0055] During the imaging process, the incident light first contacts the stratum corneum, a small part of which will be reflected by the upper epidermis, and the remaining incident light will pass through the stratum corneum and enter the epidermis. The epidermis contains a large amount of melanin, part of the light is absorbed by the melanin, and part is scattered and transmitted into the dermis; the dermis contains a large amount of hemoglobin, and the light is absorbed by the hemoglobin and scattered and transmitted. The fiber structure of the dermis will also produce lower epidermal reflection and reflect back to the skin surface. The absorption of light by melanin and hemoglobin largely determines the true color of the skin.

[0056] Therefore, this embodiment first uses the improved Lambert-Beer law to construct the relationship expression between the image pixel color value and the pigment concentration:

[0057]

[0058] in,

[0059] In the formula, C represents the difference between R, G, and B channels; the superscript log represents the logarithm; V represents the linear combination of the relative absorption vectors of the melanin layer and the hemoglobin layer; c represents the skin pigment concentration; Represents a very small quantity related to the incident spectral radiation.

[0060] Afterwards, the pigment layer was removed by independent component analysis algorithm.

[0061] 2. Remove skin interference features:

[0062] The specific steps include:

[0063] First, perform feature detection on the image data to obtain the feature response value. Then use the K-means algorithm to cluster the feature response values ​​corresponding to each image data; calculate the cluster center set W with strong significance and the cluster center set N with weak significance. At the same time, calculate the cluster radius set R of the area with strong significance and the cluster radius set S of the area with weak significance; calculate S=NA, R=N+S; where A represents the minimum feature response value. Select the maximum value of R as the threshold for feature detection, perform clustering based on the threshold, and obtain the cluster distribution.

[0064] Afterwards, the interference item of hemoglobin is removed by the Euclidean distance; and the interference item of melanin is screened out by the clustering threshold determined above.

[0065] Specifically, threshold 3 is set by the threshold calculated above. Interference items are detected on the initial image data and the hemoglobin layer image by using all the position information of the detected points obtained in the melanin layer image, and the position information of the interference items on the initial skin image and the hemoglobin layer image obtained by setting thresholds 1 and 2 respectively; and all the detection information of the feature points on the initial image data is obtained according to the optimal scale of the interference items detected on the initial image data.

[0066] Using the position information L detected on the image data s and the position information L detected on the hemoglobin layer image h , remove the interference items including hemoglobin and keep the interference items including melanin. Use its Euclidean distance to remove:

[0067]

[0068] Where, L s =[x s ,y s ], L h =[x h ,y h ].

[0069] When the Euclidean distance is less than the maximum value of the similarity of the position information between the hemoglobin layer and the image data detection point, the corresponding feature point is deleted and all the detection information is updated.

[0070] Using the updated position information L′ s and the position information L detected on the hemoglobin layer image m , remove interference items on the melanin layer and accurately locate pores; use its Euclidean distance to remove:

[0071]

[0072] When the Euclidean distance is less than the updated position information L′ s When the optimal scale of the corresponding interference item containing melanin is obtained, the corresponding position information is deleted to obtain the final detection information, that is, the pore position.

[0073] S13. Manually annotate the processed image to obtain an annotated image; the annotated image constitutes the image data set.

[0074] S2. Construct a pore prediction model, and use the image data set to train the pore prediction model to obtain a density estimation model.

[0075] In this embodiment, the density estimation model includes: a pore number calculation module, a surface area calculation module and a density calculation module; the pore number calculation module is used to calculate the number of pores per unit area; the surface area calculation module is used to calculate the surface area of ​​broilers; the density calculation module is used to calculate the pore density based on the pore number and the surface area to obtain the estimation result.

[0076] Specifically, the pore number calculation module includes: a normalization unit, a feature extraction unit, an anchor point acquisition unit, a classification unit, a pooling unit, an adjustment unit and a counting unit; the normalization unit is used to normalize the input image to obtain a scaled image of the same size; the feature extraction unit is used to extract features from the scaled image to obtain a feature map of the scaled image; in this embodiment, the feature extraction unit uses shared convolution layers such as VGG and Res50 to achieve feature extraction. The anchor point acquisition unit is used to generate all initial anchor points based on the pixel points on the feature map; on each pixel point of the feature map obtained by shared convolution, 9 anchors of different sizes and aspect ratios are generated through 3 different sizes and different aspect ratios. Using a sliding window, the initial anchors are generated with each pixel point of the feature map as the center. The classification unit is used to regress and classify the initial anchor points generated by the anchor point acquisition unit to obtain proposal points. Specifically, the RPN structure is used to regress and classify each initial anchor, and the regression operation obtains the offset of the initial anchors in each position parameter; the classification operation obtains the scores of the initial anchors in the two categories of background and non-background. The classification score of each anchor in the target object class, that is, non-background, is taken as the classification result of the anchor. Combined with the above regression and classification results, the obtained offset is used to adjust the position of the anchor's border to obtain the proposal point. By presetting parameters, several anchors with the highest scores are obtained. NMS is used to remove redundancy and obtain the final proposal point. The classification stage will continue to obtain the offset of each proposal point through a fully connected layer, and the position and size of the proposal point will be fine-tuned based on the offset to obtain the final bounding box.

[0077] The calculation method of the offset includes:

[0078]

[0079] Where x, y, w, and h represent the horizontal and vertical coordinates of the center position, width, and height of the bounding box corresponding to the anchor point, respectively; the subscript a represents the quantity describing the anchor point; the superscript * represents the quantity describing the real bounding box of the anchor point; and t represents the offset of each quantity.

[0080] The pooling unit is used to perform a pooling operation on the proposed points to obtain the proposed points of the same size; the adjustment unit uses two fully connected layer branches to perform specific attribute category classification and position adjustment operations on the proposed points output by the pooling unit to obtain the detection frame of the proposed point; the counting unit is used to count the detection frame to obtain the number of pores.

[0081] Specifically, the surface area calculation module includes: an image acquisition unit, a matching unit, a fusion reconstruction unit and a calculation unit; the image acquisition unit is used to acquire optical image data of broilers to obtain original images; wherein the original images include binocular visible light images and thermal infrared images. The matching unit is used to perform stereo matching on the original images to obtain matching images; the fusion reconstruction unit is used to perform fusion registration on the matching images to obtain a broiler three-dimensional model.

[0082] Among them, the fusion registration methods include:

[0083] After stereo matching of the binocular visible light image, a disparity map is obtained, and the disparity map is converted into a depth map. The depth map is semantically segmented to obtain a segmentation mask map. The matched thermal infrared image and the binocular visible light image are heterogeneously registered to obtain a registered thermal infrared image. The binocular visible light image, the depth map, the segmentation mask map and the registered thermal infrared image are three-dimensionally reconstructed to obtain a broiler three-dimensional model. The calculation unit is used to calculate the surface area of ​​the broiler based on the broiler three-dimensional model. Since a hairless broiler three-dimensional model is required for calculating the pores of the broiler, when performing non-contact and non-destructive testing on the broiler, a linear expression is calculated through the linear relationship between its hairy surface area and its hairless surface area. The hairless surface area of ​​the broiler is obtained by collecting images of the hairy broiler.

[0084] S3. Use the density estimation model to estimate the pore density of broilers to obtain an estimation result.

[0085] Embodiment 2:

[0086] The present embodiment provides a broiler pore density estimation device, comprising: a data set construction system, a model construction system and an estimation system; the data set construction system is used to obtain broiler skin image data, manually annotate and perform anti-interference processing on the image data to obtain an image data set; the model construction system is used to construct a pore prediction model, and the pore prediction model is trained using the image data set to obtain a density estimation model; the estimation system uses the density estimation model to estimate the pore density of broilers to obtain an estimation result.

[0087] In this embodiment, the dataset construction system includes: a dataset acquisition module, a processing module and a labeling module; the dataset acquisition module is used to obtain the complete skin image of the broiler and intercept the unit area at different parts as the image data; the processing module is used to perform anti-interference processing on the image data to obtain a processed image; the labeling module is used to manually label the processed image to obtain a labeled image; the labeled image constitutes the image dataset.

[0088] The method for the processing module to perform anti-interference processing on the image data includes:

[0089] 1. Pigment separation:

[0090] During the imaging process, the incident light first contacts the stratum corneum, a small part of which will be reflected by the upper epidermis, and the remaining incident light will pass through the stratum corneum and enter the epidermis. The epidermis contains a large amount of melanin, part of the light is absorbed by the melanin, and part is scattered and transmitted into the dermis; the dermis contains a large amount of hemoglobin, and the light is absorbed by the hemoglobin and scattered and transmitted. The fiber structure of the dermis will also produce lower epidermal reflection and reflect back to the skin surface. The absorption of light by melanin and hemoglobin largely determines the true color of the skin.

[0091] Therefore, this embodiment first uses the improved Lambert-Beer law to construct the relationship expression between the image pixel color value and the pigment concentration:

[0092]

[0093] in,

[0094] In the formula, C represents the difference between R, G, and B channels; the superscript log represents the logarithm; V represents the linear combination of the relative absorption vectors of the melanin layer and the hemoglobin layer; c represents the skin pigment concentration; Represents a very small quantity related to the incident spectral radiation.

[0095] Afterwards, the pigment layer was removed by independent component analysis algorithm.

[0096] 2. Remove skin interference features:

[0097] The specific steps include:

[0098] First, perform feature detection on the image data to obtain the feature response value. Then use the K-means algorithm to cluster the feature response values ​​corresponding to each image data; calculate the cluster center set W with strong significance and the cluster center set N with weak significance. At the same time, calculate the cluster radius set R of the area with strong significance and the cluster radius set S of the area with weak significance; calculate S=NA, R=N+S; where A represents the minimum feature response value. Select the maximum value of R as the threshold for feature detection, perform clustering based on the threshold, and obtain the cluster distribution.

[0099] Afterwards, the interference item of hemoglobin is removed by the Euclidean distance; and the interference item of melanin is screened out by the clustering threshold determined above.

[0100] Specifically, threshold 3 is set by the threshold calculated above. Interference items are detected on the initial image data and the hemoglobin layer image by using all the position information of the detected points obtained in the melanin layer image, and the position information of the interference items on the initial skin image and the hemoglobin layer image obtained by setting thresholds 1 and 2 respectively; and all the detection information of the feature points on the initial image data is obtained according to the optimal scale of the interference items detected on the initial image data.

[0101] Using the position information L detected on the image data s and the position information L detected on the hemoglobin layer image h , remove the interference items including hemoglobin and keep the interference items including melanin. Use its Euclidean distance to remove:

[0102]

[0103] Where, L s =[x s ,y s ], L h =[x h ,y h ].

[0104] When the Euclidean distance is less than the maximum value of the similarity of the position information between the hemoglobin layer and the image data detection point, the corresponding feature point is deleted and all the detection information is updated.

[0105] Using the updated position information L′ s and the position information L detected on the hemoglobin layer image m , remove interference items on the melanin layer and accurately locate pores; use its Euclidean distance to remove:

[0106]

[0107] When the Euclidean distance is less than the updated position information L′ sWhen the optimal scale of the corresponding interference item containing melanin is obtained, the corresponding position information is deleted to obtain the final detection information, that is, the pore position.

[0108] Further preferably, the density estimation model includes: a pore number calculation module, a surface area calculation module and a density calculation module; the pore number calculation module is used to calculate the number of pores per unit area; the surface area calculation module is used to calculate the surface area of ​​broilers; the density calculation module is used to calculate the pore density based on the pore number and the surface area to obtain the estimation result.

[0109] Specifically, the pore number calculation module includes: a normalization unit, a feature extraction unit, an anchor point acquisition unit, a classification unit, a pooling unit, an adjustment unit and a counting unit; the normalization unit is used to normalize the input image to obtain a scaled image of the same size; the feature extraction unit is used to extract features from the scaled image to obtain a feature map of the scaled image; in this embodiment, the feature extraction unit uses shared convolution layers such as VGG and Res50 to achieve feature extraction. The anchor point acquisition unit is used to generate all initial anchor points based on the pixel points on the feature map; on each pixel point of the feature map obtained by shared convolution, 9 anchors of different sizes and aspect ratios are generated through 3 different sizes and different aspect ratios. Using a sliding window, the initial anchors are generated with each pixel point of the feature map as the center. The classification unit is used to regress and classify the initial anchor points generated by the anchor point acquisition unit to obtain proposal points. Specifically, the RPN structure is used to regress and classify each initial anchor, and the regression operation obtains the offset of the initial anchors in each position parameter; the classification operation obtains the scores of the initial anchors in the two categories of background and non-background. The classification score of each anchor in the target object class, that is, non-background, is taken as the classification result of the anchor. Combined with the above regression and classification results, the obtained offset is used to adjust the position of the anchor's border to obtain the proposal point. By presetting parameters, several anchors with the highest scores are obtained. NMS is used to remove redundancy and obtain the final proposal point. The classification stage will continue to obtain the offset of each proposal point through a fully connected layer, and the position and size of the proposal point will be fine-tuned based on the offset to obtain the final bounding box.

[0110] The calculation method of the offset includes:

[0111]

[0112] Where x, y, w, and h represent the horizontal and vertical coordinates of the center position, width, and height of the bounding box corresponding to the anchor point, respectively; the subscript a represents the quantity describing the anchor point; the superscript * represents the quantity describing the real bounding box of the anchor point; and t represents the offset of each quantity.

[0113] The pooling unit is used to perform a pooling operation on the proposed points to obtain the proposed points of the same size; the adjustment unit uses two fully connected layer branches to perform specific attribute category classification and position adjustment operations on the proposed points output by the pooling unit to obtain the detection frame of the proposed point; the counting unit is used to count the detection frame to obtain the number of pores.

[0114] Specifically, the surface area calculation module includes: an image acquisition unit, a matching unit, a fusion reconstruction unit and a calculation unit; the image acquisition unit is used to acquire optical image data of broilers to obtain original images; wherein the original images include binocular visible light images and thermal infrared images. The matching unit is used to perform stereo matching on the original images to obtain matching images; the fusion reconstruction unit is used to perform fusion registration on the matching images to obtain a broiler three-dimensional model.

[0115] Among them, the fusion registration methods include:

[0116] After stereo matching of the binocular visible light image, a disparity map is obtained, and the disparity map is converted into a depth map. The depth map is semantically segmented to obtain a segmentation mask map. The matched thermal infrared image and the binocular visible light image are heterogeneously registered to obtain a registered thermal infrared image. The binocular visible light image, the depth map, the segmentation mask map and the registered thermal infrared image are three-dimensionally reconstructed to obtain a broiler three-dimensional model. The calculation unit is used to calculate the surface area of ​​the broiler based on the broiler three-dimensional model. Since a hairless broiler three-dimensional model is required for calculating the pores of the broiler, when performing non-contact and non-destructive testing on the broiler, a linear expression is calculated through the linear relationship between its hairy surface area and its hairless surface area. The hairless surface area of ​​the broiler is obtained by collecting images of the hairy broiler.

[0117] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for estimating pore density of broiler chickens, characterized in that: The following steps are involved: S1. Acquire broiler skin image data, perform manual annotation and anti-interference processing on the image data, and obtain an image data set; S2, constructing a pore prediction model, and using the image data set to train the pore prediction model to obtain a density estimation model; S3. Use the density estimation model to estimate the pore density of broilers to obtain an estimation result.

2. A method for estimating pore density of broiler chickens according to claim 1, characterized in that: S1 includes the following steps: S11, obtaining a complete skin image of a broiler chicken, and intercepting unit areas at different parts as the image data; S12, performing anti-interference processing on the image data to obtain a processed image; S13, manually annotating the processed image to obtain an annotated image; The annotated images constitute the image dataset.

3. The method for estimating pore density of broiler chickens according to claim 1, characterized in that: The density estimation model includes: a pore number calculation module, a surface area calculation module and a density calculation module; The pore number calculation module is used to calculate the number of pores per unit area; The surface area calculation module is used to calculate the surface area of ​​the broiler; The density calculation module is used to calculate the pore density based on the number of pores and the surface area to obtain the estimation result.

4. A method for estimating pore density of broiler chickens according to claim 3, characterized in that: The pore number calculation module includes: a normalization unit, a feature extraction unit, an anchor point acquisition unit, a classification unit, a pooling unit, an adjustment unit and a counting unit; The normalization unit is used to perform normalization processing on the input image to obtain a scaled image with the same size; The feature extraction unit is used to extract features from the zoomed image to obtain a feature map of the zoomed image; The anchor point acquisition unit is used to generate all initial anchor points based on the pixel points on the feature map; The classification unit is used to regress and classify the initial anchor points generated by the anchor point acquisition unit to obtain proposal points; The pooling unit is used to perform a pooling operation on the proposed points to obtain the proposed points of the same size; The adjustment unit is used to perform an adjustment operation on the proposed point output by the pooling unit to obtain a detection frame of the proposed point; The counting unit is used to count the detection frames to obtain the number of pores.

5. A method for estimating pore density of broiler chickens according to claim 3, characterized in that: The surface area calculation module includes: an image acquisition unit, a matching unit, a fusion reconstruction unit and a calculation unit; The image acquisition unit is used to collect optical image data of broilers to obtain original images; The matching unit is used to perform stereo matching on the original image to obtain a matching image; The fusion reconstruction unit is used to fuse and register the matching images to obtain a broiler three-dimensional model; The calculation unit is used to calculate the surface area of ​​the broiler based on the broiler three-dimensional model.

6. A device for estimating pore density of broiler chickens, the device being used to implement the method according to any one of claims 1 to 5, characterized in that: include: Dataset construction system, model construction system, and estimation system; The data set construction system is used to obtain broiler skin image data, perform manual annotation and anti-interference processing on the image data, and obtain an image data set; The model building system is used to build a pore prediction model, and the image data set is used to train the pore prediction model to obtain a density estimation model; The estimation system uses the density estimation model to estimate the pore density of broilers to obtain an estimation result.

7. A broiler pore density estimation device according to claim 6, characterized in that: The data set construction system includes: a data set acquisition module, a processing module and a labeling module; The data set acquisition module is used to obtain the complete skin image of the broiler chicken and intercept the unit area at different parts as the image data; The processing module is used to perform anti-interference processing on the image data to obtain a processed image; The annotation module is used to manually annotate the processed image to obtain an annotated image; the annotated image constitutes the image data set.

8. A broiler pore density estimation device according to claim 6, characterized in that: The density estimation model includes: a pore number calculation module, a surface area calculation module and a density calculation module; The pore number calculation module is used to calculate the number of pores per unit area; The surface area calculation module is used to calculate the surface area of ​​the broiler; The density calculation module is used to calculate the pore density based on the number of pores and the surface area to obtain the estimation result.

Images