Backfat thickness detection system and method

Abstract

The invention relates to the field of backfat thickness detection, in particular to a backfat thickness detection system and method. The method comprises the following steps: setting a channel detection region, collecting animal image data and animal identity information through the channel detection region, analyzing the animal image data and the animal identity information, and constructing a three-dimensional target model; extracting the three-dimensional target model to obtain a three-dimensional target animal area, and identifying the three-dimensional target animal area to obtain a backfat detection area; processing the backfat detection area to obtain a backfat perspective model and a back section point set; in combination with a backfat perspective model, analyzing the back section point set to obtain a back boundary point set; and obtaining a detection thickness value according to the back boundary point set. According to the invention, the backfat thickness detection efficiency can be improved.

Description

Technical Field

[0001] This invention relates to the field of back fat thickness detection, and specifically to a back fat thickness detection system and method. Background Technology

[0002] In modern livestock production, the body condition indicators of individual animals are important bases affecting breeding efficiency, feeding management, health assessment, and slaughter grading. Backfat thickness, as an important parameter for measuring the level of fat deposition in livestock such as pigs and cattle, is directly related to the animal's nutritional status, growth stage, and meat quality grade, and is one of the key indicators in livestock breeding, refined feeding management, and product grading standards.

[0003] With the development of intelligent and information-based aquaculture, non-contact detection technology has gradually become a research and application hotspot. Especially driven by technologies such as image recognition, computer vision, deep learning, and 3D reconstruction, backfat thickness detection systems based on visual perception have advantages such as non-contact, high precision, strong real-time performance, and good traceability, making them an important direction for promoting the upgrading of the aquaculture industry.

[0004] Chinese Patent Publication No. CN110754426A discloses an automatic backfat detection and automatic animal feeding method and system, comprising: acquiring an image sequence of an animal at a specific location; extracting animal body shape regions from each image in the image sequence to obtain several animal body shape region images; marking key points in each of the several animal body shape region images; transforming the several animal body shape region images into standard body shape images based on the marked key points; comparing and analyzing the standard body shape images with a pre-established standard body shape template information database; determining parameter values ​​related to backfat based on the comparison and analysis results; and calculating the current body weight value of the animal based on the parameter values ​​and the current information of the individual animal.

[0005] In existing technologies, farms mainly rely on traditional methods such as manual palpation or ultrasound to detect backfat thickness. While these methods can obtain information on the thickness of animal surface or subcutaneous fat to some extent, they generally have the following problems: operation depends on manual judgment or professional skills, resulting in low detection efficiency; animals need to maintain a fixed posture or cooperate with the measurement, which can easily cause animal stress; it is difficult to achieve batch, continuous, and automated detection, affecting the application effect in large-scale farms; and the detection results are highly subjective, with poor repeatability and consistency, affecting data reliability. These are the problems we need to solve. Summary of the Invention

[0006] The purpose of this invention is to address the problems existing in the background technology by proposing a method for detecting backfat thickness.

[0007] The technical solution of the present invention: a method for detecting backfat thickness, comprising the following steps:

[0008] S1. Set up a channel detection area, collect animal image data and animal identity information through the channel detection area, analyze the animal image data and animal identity information, and construct a three-dimensional target model;

[0009] S2. Extract the three-dimensional target model to obtain the three-dimensional target animal region, identify the three-dimensional target animal region, and obtain the backfat detection region;

[0010] S3. Process the back fat detection area to obtain the back fat perspective model and the set of back cross-section points;

[0011] S4. Combine the back fat perspective model to analyze the set of points on the back section and obtain the set of back boundary points; based on the set of back boundary points, obtain the detected thickness value.

[0012] Preferably, the process of setting up a channel detection area and collecting animal image data and animal identification information through the channel detection area includes:

[0013] A channel detection area is set up, and the channel detection area is equipped with a detection unit. The detection unit includes an RGB camera, a depth camera, a laser rangefinder, an infrared sensor, a pressure sensor, a wireless radio frequency identification reader, and an object motion sensor.

[0014] The channel detection area includes a channel detection start point and a channel detection end point;

[0015] When the target animal moves from the beginning to the end of the detection channel or from the end to the beginning of the detection channel, the detection unit uses infrared beams and pressure sensors to determine whether the animal has completely passed through the beginning and end of the detection channel. If it has, the detection unit's acquisition mechanism is triggered, and the detection unit performs synchronous detection on the target animal, acquiring animal image data and animal identification information.

[0016] The animal image data includes several point cloud data and animal bitmap data; the animal identification information includes RFID and QR code recognition.

[0017] Preferably, the process of analyzing animal image data and animal identification information to construct a three-dimensional target model includes:

[0018] Animal image data is processed by RFID and QR code recognition of animal identity information. Multiple point cloud data and animal bitmap data are grouped to obtain animal bitmap point cloud groups. RFID is converted into codes and recorded as animal codes. Animal codes are then assigned to animal bitmap point cloud groups and recorded as the number of animal bitmap point cloud groups.

[0019] The data within the animal bitmap point cloud group is analyzed. The point cloud data and animal bitmap data corresponding to the same area are grouped together, and the animal bitmap data from all angles are input into the point cloud data to obtain three-dimensional point cloud data. Three-dimensional modeling is performed on the obtained three-dimensional point cloud data to obtain a three-dimensional target model.

[0020] Preferably, the process of extracting the three-dimensional target model to obtain the three-dimensional target animal region is as follows:

[0021] The three-dimensional target model is analyzed, and the target animal is detected in the image within the three-dimensional target model using the deep learning algorithm YOLOv5. The overall outline of the detected target animal is identified, and the overall outline of the animal in the three-dimensional target model is drawn to obtain the three-dimensional target animal region.

[0022] Using machine learning and computer vision technologies, the integrity of the 3D target animal region is assessed. By extracting the animal's posture, the integrity of the 3D target animal region image is determined. If complete, the analysis of the 3D target animal region continues; if incomplete, the corresponding region of the target animal is located by using adjacent images from all angles and directions of the incomplete image, and a new image is acquired. The re-acquired image data is then used to complete the 3D target animal region, and the outline of the 3D target animal region automatically changes as the region changes.

[0023] Preferably, the process of identifying the three-dimensional target animal region and obtaining the backfat detection region includes:

[0024] Using an animal back database, the back region of a 3D target animal is identified, the back region of the target animal is extracted, and key points of interest are marked within the 3D target animal region. Key skeletal points of the target back region are identified and recorded as back bone nodes. The back bone nodes are connected according to the bone direction within the target back region, and lines are drawn within the target back region to construct back bone lines. The back fat detection location is determined by the direction of the back bone lines within the target back region and recorded as the back fat detection area.

[0025] Preferably, the process of processing the backfat detection area to obtain the backfat perspective model and the set of back cross-sectional points includes:

[0026] Acquire multi-angle image data of the backfat detection area and input the acquired multi-angle image data into the 3D model to construct the backfat point cloud model. Generate density point cloud data through the RGB-D method and combine it with the detection unit to extract the depth image data related to the backfat detection area. Convert the 2D bitmap data and depth image data in the backfat detection area into 3D point cloud coordinates.

[0027] Image segmentation is used to process all three-dimensional point cloud coordinates in the backfat detection area. Three-dimensional point cloud coordinates that are not related to backfat are deleted. Noise points in the three-dimensional point cloud coordinates are identified and removed by a radius filter. The three-dimensional point cloud coordinates are downsampled by a voxel mesh method to unify the point density of the three-dimensional point cloud coordinates. The surface normal vector of the three-dimensional point cloud coordinates is calculated by the point cloud neighborhood to obtain the three-dimensional point cloud vector.

[0028] The position and orientation of the three-dimensional point cloud coordinates are calibrated using three-dimensional point cloud vectors and three-dimensional coordinate space. Then, the three-dimensional point cloud coordinates are recalibrated using an iterative nearest point algorithm. The calibrated three-dimensional point cloud coordinates are unified into the three-dimensional coordinate system. The back curve is reconstructed and calibrated using three-dimensional point cloud vectors to restore the back angle of the back fat detection area as much as possible and construct a back fat perspective model.

[0029] Set the cross-sectional offset value, and use the cross-sectional offset value to cut the back section of the back fat perspective model to obtain several sets of back section points.

[0030] Preferably, by combining the backfat perspective model, the set of points on the back cross section is analyzed to obtain the set of back boundary points; the process of obtaining the detection thickness value based on the set of back boundary points includes:

[0031] Using a curvature change algorithm combined with a backfat perspective model, the outer contour of the skin and the boundary of the back muscles of the target animal are identified and detected. The set of back cross-section points of each group is filtered to obtain the set of back boundary points.

[0032] The back boundary point set is sorted according to the point cloud angle and spatial position, and the back boundary point set is connected in sequence to construct the back curve point sequence. The back curve point sequence is highlighted at the corresponding position of the back fat perspective model, and the unhighlighted area of ​​the same cross section is resampled to ensure that the back boundary points are evenly distributed. The back curve point sequence is denoised by a local smoothing algorithm to remove the rough parts of the back surface.

[0033] Calculate the minimum vertical distance between two points within each set of back cross-section points to obtain the local thickness value. Calculate the average local thickness value of all sets of back cross-section points to obtain the backfat thickness value of the target animal.

[0034] Repeat the above steps to calculate the average value of the obtained backfat thickness values ​​and obtain the detected thickness value.

[0035] This invention also discloses a backfat thickness detection system, including a management center, which is communicatively connected to a target data acquisition module, a backfat identification module, a backfat detection and analysis module, and a backfat thickness detection module.

[0036] The target data acquisition module is used to set the channel detection area, collect animal image data and animal identification information through the channel detection area, and analyze the animal image data and animal identification information to construct a three-dimensional target model;

[0037] The backfat recognition module is used to extract the three-dimensional target model, obtain the three-dimensional target animal region, and recognize the three-dimensional target animal region to obtain the backfat detection region;

[0038] The back fat detection and analysis module is used to process the back fat detection area to obtain a back fat perspective model and a set of back cross-sectional points;

[0039] The backfat thickness detection module is used to analyze the set of points on the back section by combining the backfat perspective model to obtain the set of back boundary points; and obtain the detected thickness value based on the set of back boundary points.

[0040] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects:

[0041] A channel detection area is set up, and animal image data and animal identification information are collected through the channel detection area. The animal image data and animal identification information are analyzed to construct a 3D target model. The 3D target model is extracted to obtain the 3D target animal region. The 3D target animal region is identified to obtain the backfat detection area. The backfat detection area is processed to obtain the backfat perspective model and the set of back cross-sectional points. Combined with the backfat perspective model, the set of back cross-sectional points is analyzed to obtain the set of back boundary points. Based on the set of back boundary points, the detected thickness value is obtained. This achieves non-contact measurement, improves measurement accuracy and data reliability, and improves the efficiency of backfat thickness detection. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of one embodiment of the present invention. Detailed Implementation

[0043] Example 1, as Figure 1 As shown, the present invention proposes a method for detecting backfat thickness, comprising the following steps:

[0044] S1. Set up a channel detection area, collect animal image data and animal identity information through the channel detection area, analyze the animal image data and animal identity information, and construct a three-dimensional target model;

[0045] S2. Extract the three-dimensional target model to obtain the three-dimensional target animal region, identify the three-dimensional target animal region, and obtain the backfat detection region;

[0046] S3. Process the back fat detection area to obtain the back fat perspective model and the set of back cross-section points;

[0047] S4. Combine the back fat perspective model to analyze the set of points on the back section and obtain the set of back boundary points; based on the set of back boundary points, obtain the detected thickness value.

[0048] It should be further explained that, in the specific implementation process, the process of setting up a channel detection area, collecting animal image data and animal identification information through the channel detection area, and analyzing the animal image data and animal identification information to construct a 3D target model is as follows:

[0049] A channel detection area is set up, which is located in the fixed activity area of ​​the target animal. A detection unit is set up inside the channel detection area. The detection unit includes an RGB camera, a depth camera, a laser rangefinder, an infrared sensor, a pressure sensor, a wireless radio frequency identification reader, and an object motion sensor.

[0050] The channel detection area includes a channel detection start point and a channel detection end point;

[0051] The target animal for testing refers to the animal for which backfat thickness testing is required, such as pigs and cattle.

[0052] When the target animal moves from the beginning to the end of the detection channel or from the end to the beginning of the detection channel, the detection unit uses infrared beams and pressure sensors to determine whether the animal has completely passed through the beginning and end of the detection channel. If it has, the detection unit's acquisition mechanism is triggered, and the detection unit performs synchronous detection on the target animal, acquiring animal image data and animal identification information.

[0053] The animal image data includes several point cloud data and animal bitmap data; the animal identification information includes RFID and QR code recognition; the animal bitmap data is synchronous image data of all angles of the target animal being detected, while ensuring that the bitmap data of all angles can cover the entire range of the target animal being detected, and that there is some overlap between the bitmap data of adjacent angles.

[0054] Animal image data is processed by RFID and QR code recognition of animal identity information. Multiple point cloud data and animal bitmap data are grouped together. Point cloud data and animal bitmap data belonging to the same animal identity information are grouped together and called animal bitmap point cloud group. RFID is converted into code and called animal code. Animal code is assigned to animal bitmap point cloud group and called animal bitmap point cloud group number.

[0055] The data within the animal bitmap point cloud group is analyzed. The point cloud data and animal bitmap data corresponding to the same area are grouped together, and the animal bitmap data from all angles are input into the point cloud data to obtain three-dimensional point cloud data. Three-dimensional modeling is performed on the obtained three-dimensional point cloud data to obtain a three-dimensional target model.

[0056] It should be further explained that, in the specific implementation process, the process of extracting the three-dimensional target model to obtain the three-dimensional target animal region, and identifying the three-dimensional target animal region to obtain the backfat detection region is as follows:

[0057] The three-dimensional target model is analyzed, and the target animal is detected in the image within the three-dimensional target model using the deep learning algorithm YOLOv5. The overall outline of the detected target animal is identified, and the overall outline of the animal in the three-dimensional target model is drawn to obtain the three-dimensional target animal region.

[0058] Using machine learning and computer vision technology, the integrity of the 3D target animal region is judged. By extracting the animal's posture, the integrity of the 3D target animal region image is determined. If it is complete, the 3D target animal region is analyzed further. If it is incomplete, the corresponding region of the target animal is located by using adjacent images from all angles and directions of the incomplete image, and the image is re-acquired. The re-acquired image data is then used to complete the 3D target animal region, and the outline of the 3D target animal region is automatically changed as the region changes.

[0059] Using an animal back database, the back region of a 3D target animal is identified, the back region of the target animal is extracted, and key points of interest are marked within the 3D target animal region. Key skeletal points of the target back region are identified and recorded as back bone nodes. The back bone nodes are connected according to the bone direction within the target back region, and lines are drawn within the target back region to construct back bone lines. The back fat detection location is determined by the direction of the back bone lines within the target back region and recorded as the back fat detection area.

[0060] It should be further explained that, in the specific implementation process, the process of processing the backfat detection area to obtain the backfat perspective model and the set of back cross-sectional points is as follows:

[0061] Acquire multi-angle image data of the backfat detection area and input the acquired multi-angle image data into the 3D model to construct the backfat point cloud model. Generate density point cloud data through the RGB-D method and combine it with the detection unit to extract the depth image data related to the backfat detection area. Convert the 2D bitmap data and depth image data in the backfat detection area into 3D point cloud coordinates.

[0062] Image segmentation is used to process all three-dimensional point cloud coordinates in the backfat detection area. Three-dimensional point cloud coordinates that are not related to backfat are deleted. Noise points in the three-dimensional point cloud coordinates are identified and removed by a radius filter. The three-dimensional point cloud coordinates are downsampled by a voxel mesh method to unify the point density of the three-dimensional point cloud coordinates. The surface normal vector of the three-dimensional point cloud coordinates is calculated by the point cloud neighborhood to obtain the three-dimensional point cloud vector.

[0063] The position and orientation of the three-dimensional point cloud coordinates are calibrated using three-dimensional point cloud vectors and three-dimensional coordinate space. Then, the three-dimensional point cloud coordinates are recalibrated using an iterative nearest point algorithm. The calibrated three-dimensional point cloud coordinates are unified into the three-dimensional coordinate system. The back curve is reconstructed and calibrated using three-dimensional point cloud vectors to restore the back angle of the back fat detection area as much as possible and construct a back fat perspective model.

[0064] Set the cross-sectional offset value, and use the cross-sectional offset value to cut the back section of the back fat perspective model to obtain several sets of back section points.

[0065] It should be further explained that, in the specific implementation process, the set of points on the back section is analyzed in conjunction with the back fat perspective model to obtain the set of back boundary points; the process of obtaining the detection thickness value based on the back boundary point set is as follows:

[0066] Using a backfat perspective model, the thickness of each set of back cross-sectional points is measured.

[0067] Using a curvature change algorithm combined with a backfat perspective model, the outer contour of the skin and the boundary of the back muscles of the target animal are identified and detected. The set of back cross-section points of each group is filtered to obtain the set of back boundary points.

[0068] The back boundary point set is sorted according to the point cloud angle and spatial position, and the back boundary point set is connected in sequence to construct the back curve point sequence. The back curve point sequence is highlighted at the corresponding position of the back fat perspective model, and the unhighlighted area of ​​the same cross section is resampled to ensure that the back boundary points are evenly distributed. The back curve point sequence is denoised by a local smoothing algorithm to remove the rough parts of the back surface.

[0069] Calculate the minimum vertical distance between two points within each set of back cross-section points to obtain the local thickness value. Calculate the average local thickness value of all sets of back cross-section points to obtain the backfat thickness value of the target animal.

[0070] Repeat the above steps to calculate the average value of the obtained backfat thickness values ​​and obtain the detected thickness value.

[0071] Example 2: The backfat thickness detection system proposed in this invention is applied to the backfat thickness detection method described in Example 1. Specifically, it includes a management center, which is communicatively connected to a target data acquisition module, a backfat identification module, a backfat detection and analysis module, and a backfat thickness detection module.

[0072] The target data acquisition module is used to set the channel detection area, collect animal image data and animal identification information through the channel detection area, and analyze the animal image data and animal identification information to construct a three-dimensional target model;

[0073] The backfat recognition module is used to extract the three-dimensional target model, obtain the three-dimensional target animal region, and recognize the three-dimensional target animal region to obtain the backfat detection region;

[0074] The back fat detection and analysis module is used to process the back fat detection area to obtain a back fat perspective model and a set of back cross-sectional points;

[0075] The backfat thickness detection module is used to analyze the set of points on the back section by combining the backfat perspective model to obtain the set of back boundary points; and obtain the detected thickness value based on the set of back boundary points.

[0076] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for detecting backfat thickness, characterized in that, Includes the following steps: S1. Set up a channel detection area, collect animal image data and animal identity information through the channel detection area, analyze the animal image data and animal identity information, and construct a three-dimensional target model; S2. Extract the three-dimensional target model to obtain the three-dimensional target animal region, identify the three-dimensional target animal region, and obtain the backfat detection region; S3. Process the back fat detection area to obtain the back fat perspective model and the set of back cross-section points; S4. Combine the back fat perspective model to analyze the set of points on the back section and obtain the set of back boundary points; based on the set of back boundary points, obtain the detected thickness value.

2. The method for detecting backfat thickness according to claim 1, characterized in that, The process of setting up a channel detection area and collecting animal image data and animal identification information through the channel detection area includes: A channel detection area is set up, and the channel detection area is equipped with a detection unit. The detection unit includes an RGB camera, a depth camera, a laser rangefinder, an infrared sensor, a pressure sensor, a wireless radio frequency identification reader, and an object motion sensor. The channel detection area includes a channel detection start point and a channel detection end point; When the target animal moves from the beginning to the end of the detection channel or from the end to the beginning of the detection channel, the detection unit uses infrared beams and pressure sensors to determine whether the animal has completely passed through the beginning and end of the detection channel. If it has, the detection unit's acquisition mechanism is triggered, and the detection unit performs synchronous detection on the target animal, acquiring animal image data and animal identification information. The animal image data includes several point cloud data and animal bitmap data; the animal identification information includes RFID and QR code recognition.

3. The method for detecting backfat thickness according to claim 2, characterized in that, The process of analyzing animal image data and animal identification information to construct a 3D target model includes: Animal image data is processed by RFID and QR code recognition of animal identity information. Multiple point cloud data and animal bitmap data are grouped to obtain animal bitmap point cloud groups. RFID is converted into codes and recorded as animal codes. Animal codes are then assigned to animal bitmap point cloud groups and recorded as the number of animal bitmap point cloud groups. The data within the animal bitmap point cloud group is analyzed. The point cloud data and animal bitmap data corresponding to the same area are grouped together, and the animal bitmap data from all angles are input into the point cloud data to obtain three-dimensional point cloud data. Three-dimensional modeling is performed on the obtained three-dimensional point cloud data to obtain a three-dimensional target model.

4. The method for detecting backfat thickness according to claim 3, characterized in that, The process of extracting the three-dimensional target model to obtain the three-dimensional target animal region is as follows: The three-dimensional target model is analyzed, and the target animal is detected in the image within the three-dimensional target model using the deep learning algorithm YOLOv5. The overall outline of the detected target animal is identified, and the overall outline of the animal in the three-dimensional target model is drawn to obtain the three-dimensional target animal region. Using machine learning and computer vision technologies, the integrity of the 3D target animal region is assessed. By extracting the animal's posture, the integrity of the 3D target animal region image is determined. If complete, the analysis of the 3D target animal region continues; if incomplete, the corresponding region of the target animal is located by using adjacent images from all angles and directions of the incomplete image, and a new image is acquired. The re-acquired image data is then used to complete the 3D target animal region, and the outline of the 3D target animal region automatically changes as the region changes.

5. The method for detecting backfat thickness according to claim 4, characterized in that, The process of identifying the three-dimensional target animal region and obtaining the backfat detection region includes: Using an animal back database, the back region of a 3D target animal is identified, the back region of the target animal is extracted, and key points of interest are marked within the 3D target animal region. Key skeletal points of the target back region are identified and recorded as back bone nodes. The back bone nodes are connected according to the bone direction within the target back region, and lines are drawn within the target back region to construct back bone lines. The back fat detection location is determined by the direction of the back bone lines within the target back region and recorded as the back fat detection area.

6. A method for detecting backfat thickness according to claim 1 or 5, characterized in that, The process of processing the back fat detection area to obtain the back fat perspective model and the set of back cross-sectional points includes: Acquire multi-angle image data of the backfat detection area and input the acquired multi-angle image data into the 3D model to construct the backfat point cloud model. Generate density point cloud data through the RGB-D method and combine it with the detection unit to extract the depth image data related to the backfat detection area. Convert the 2D bitmap data and depth image data in the backfat detection area into 3D point cloud coordinates. Image segmentation is used to process all three-dimensional point cloud coordinates in the backfat detection area. Three-dimensional point cloud coordinates that are not related to backfat are deleted. Noise points in the three-dimensional point cloud coordinates are identified and removed by a radius filter. The three-dimensional point cloud coordinates are downsampled by a voxel mesh method to unify the point density of the three-dimensional point cloud coordinates. The surface normal vector of the three-dimensional point cloud coordinates is calculated by the point cloud neighborhood to obtain the three-dimensional point cloud vector. The position and orientation of the three-dimensional point cloud coordinates are calibrated using three-dimensional point cloud vectors and three-dimensional coordinate space. Then, the three-dimensional point cloud coordinates are recalibrated using an iterative nearest point algorithm. The calibrated three-dimensional point cloud coordinates are unified into the three-dimensional coordinate system. The back curve is reconstructed and calibrated using three-dimensional point cloud vectors to restore the back angle of the back fat detection area as much as possible and construct a back fat perspective model. Set the cross-sectional offset value, and use the cross-sectional offset value to cut the back section of the back fat perspective model to obtain several sets of back section points.

7. The method for detecting backfat thickness according to claim 6, characterized in that, By combining the back fat perspective model, the set of points on the back section is analyzed to obtain the set of boundary points on the back. The process of obtaining the detected thickness value based on the set of back boundary points includes: Using a curvature change algorithm combined with a backfat perspective model, the outer contour of the skin and the boundary of the back muscles of the target animal are identified and detected. The set of back cross-section points of each group is filtered to obtain the set of back boundary points. The back boundary point set is sorted according to the point cloud angle and spatial position, and the back boundary point set is connected in sequence to construct the back curve point sequence. The back curve point sequence is highlighted at the corresponding position of the back fat perspective model, and the unhighlighted area of ​​the same cross section is resampled to ensure that the back boundary points are evenly distributed. The back curve point sequence is denoised by a local smoothing algorithm to remove the rough parts of the back surface. Calculate the minimum vertical distance between two points within each set of back cross-section points to obtain the local thickness value. Calculate the average local thickness value of all sets of back cross-section points to obtain the backfat thickness value of the target animal. Repeat the above steps to calculate the average value of the obtained backfat thickness values ​​and obtain the detected thickness value.

8. A backfat thickness detection system, specifically applied to the backfat thickness detection method according to any one of claims 1 to 7, comprising a management center, characterized in that, The management center's communication connections include a target data acquisition module, a backfat identification module, a backfat detection and analysis module, and a backfat thickness detection module. The target data acquisition module is used to set the channel detection area, collect animal image data and animal identification information through the channel detection area, and analyze the animal image data and animal identification information to construct a three-dimensional target model; The backfat recognition module is used to extract the three-dimensional target model, obtain the three-dimensional target animal region, and recognize the three-dimensional target animal region to obtain the backfat detection region; The back fat detection and analysis module is used to process the back fat detection area to obtain a back fat perspective model and a set of back cross-sectional points; The backfat thickness detection module is used to analyze the set of points on the back section by combining the backfat perspective model to obtain the set of back boundary points; and obtain the detected thickness value based on the set of back boundary points.

Images