Beef cattle carcass computer vision auxiliary segmentation system

Abstract

The invention relates to the technical field of beef carcass segmentation, in particular to a beef carcass computer vision-assisted segmentation system, which comprises a vision acquisition and preprocessing module, a feature extraction module, a segmentation decision module, a dynamic target tracking module, an execution control module and a temperature monitoring and feedback module, the visual acquisition and preprocessing module is used for acquiring image data of beef carcasses; the feature extraction module extracts feature points and edge information; the segmentation decision module generates a segmentation path; the dynamic target tracking module corrects the generated segmentation path; the execution control module is used for cutting beef carcasses; and the temperature monitoring and feedback module monitors the temperature of the cutting equipment and the carcass cutting part in real time. According to the invention, the error and the labor intensity of manual operation are reduced, and the automation level and the overall benefit of the beef carcass segmentation process are improved.

Description

Technical Field

[0001] The invention relates to the technical field of beef cattle carcass segmentation, and in particular to a beef cattle carcass computer vision-assisted segmentation system. Background Art

[0002] With the development of modern agriculture, the beef cattle breeding and processing industries have put forward higher requirements on the efficiency and accuracy of beef cattle carcass segmentation. Traditional beef cattle carcass segmentation relies on manual operation, which is not only labor-intensive and inefficient, but also difficult to ensure segmentation accuracy, resulting in waste of beef cattle resources and increased processing costs. Therefore, the development of an efficient, precise and intelligent beef cattle carcass segmentation system has become an urgent need of the industry.

[0003] Existing beef cattle carcass cutting technology mainly relies on mechanical equipment and simple automated control systems, lacks advanced image processing and intelligent control algorithms, and the accuracy and stability of mechanical equipment are limited. It is unable to track and adjust the cutting path in real time, resulting in low cutting accuracy. In addition, the existing technology lacks temperature monitoring and feedback control during the cutting process, and cannot effectively avoid the problems of reduced cutting quality and equipment damage caused by equipment overheating. These technical bottlenecks limit the improvement of cutting efficiency and quality.

[0004] The present invention aims to provide a computer vision-assisted beef cattle carcass segmentation system, which can dynamically adjust the segmentation path in real time to ensure cutting accuracy and quality, and avoid the problems of cutting quality degradation and equipment damage caused by overheating through real-time temperature monitoring and feedback control, thereby significantly improving the segmentation efficiency and overall benefits. Summary of the invention

[0005] Based on the above objectives, the present invention provides a beef cattle carcass computer vision-assisted segmentation system.

[0006] The computer vision-assisted segmentation system for beef carcasses includes a visual acquisition and preprocessing module, a feature extraction module, a segmentation decision module, a dynamic target tracking module, an execution control module, and a temperature monitoring and feedback module, wherein;

[0007] The visual acquisition and preprocessing module acquires image data of the beef cattle carcass and performs preprocessing on the acquired image data, including noise filtering, image enhancement and preliminary identification of segmented areas;

[0008] The feature extraction module extracts feature points and edge information from the preprocessed image data;

[0009] The segmentation decision module analyzes the extracted feature points and edge information to generate a segmentation path;

[0010] The dynamic target tracking module tracks the position and posture changes of the beef carcass in the segmentation process in real time and corrects the generated segmentation path, specifically including:

[0011] Target recognition: Use image recognition algorithms to detect key parts of beef cattle carcasses in real time;

[0012] Position prediction: Predict the movement trajectory of beef carcasses based on the Kalman filter algorithm;

[0013] Posture analysis: adjust the segmentation path by analyzing the posture changes of the carcass;

[0014] Path correction: According to the results of position prediction and posture analysis, the segmentation path is dynamically corrected in real time;

[0015] The execution control module receives the corrected segmentation path and controls the segmentation equipment to perform a cutting operation on the beef carcass according to the corrected segmentation path;

[0016] The temperature monitoring and feedback module monitors the temperature of the cutting equipment and the carcass cutting parts in real time.

[0017] Optionally, the visual acquisition and preprocessing module includes:

[0018] Multi-angle image acquisition: Use multiple cameras, arranged at different angles, to collect image data of beef carcasses in real time;

[0019] Noise filtering: Adaptive median filtering algorithm is used to filter noise from the collected image data;

[0020] Image enhancement: The histogram equalization algorithm is used to enhance the image data after noise filtering;

[0021] Preliminary identification of segmented areas: The Otsu threshold segmentation algorithm is used to perform preliminary identification of segmented areas on the enhanced image data.

[0022] Optionally, the feature extraction module includes:

[0023] Feature point extraction: The scale-invariant feature transform (SIFT) algorithm is used to extract feature points from the preprocessed image data, expressed as:

[0024] D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))*I(x,y);

[0025] Where D(x, y, σ) is the difference function in the scale space, G(x, y, σ) is the Gaussian function in the scale space, I(x, y) is the input image, and k is a constant;

[0026] Edge information extraction: The Canny edge detection algorithm is used to perform edge detection on the preprocessed image data.

[0027] Optionally, the segmentation decision module includes:

[0028] Feature map generation: Convolutional neural network (CNN) is used to generate feature maps of input feature points and edge information, which can be expressed as:

[0029] y=f(W*x+b);

[0030] Among them, y is the output feature map, W is the convolution kernel, x is the input feature point and edge information, b is the bias, and f is the activation function;

[0031] Data Analysis: Time series data is analyzed using a recurrent neural network (RNN) to generate a dynamic segmentation path, represented as:

[0032] h t =tanh(W h h t-1 +W x x t + b);

[0033] Among them, h t is the hidden state at the current moment, h t-1 is the hidden state at the previous moment, W h and W x are weight matrices, x t is the input at the current moment, b is the bias;

[0034] y t =W y h t +b y ;

[0035] Among them, y t is the output at the current moment, W y is the output weight matrix, b y is the bias of the output layer;

[0036]

[0037] Where n is the number of samples, y i is the actual value, is the predicted value;

[0038] Segmentation path generation: Combining the results of feature map generation and data analysis, the optimal segmentation path is generated through an optimization algorithm, which is expressed as:

[0039]

[0040] Among them, L is the loss function, y i is the actual output, is the predicted output, λ is the regularization parameter, W j is the weight matrix.

[0041] Optionally, the image recognition algorithm adopts a Faster R-CNN algorithm, and the Faster R-CNN algorithm includes:

[0042] Region Proposal Network: A region proposal network (RPN) is used to generate candidate regions, combining multi-scale features to adapt to different sizes of beef carcass parts, expressed as:

[0043]

[0044] Among them, p i is the category prediction probability of the i-th anchor point, is the actual category of the i-th anchor point, t i is the regression prediction of the i-th anchor point, is the actual regression value of the i-th anchor point, L cls is the classification loss function, L reg is the regression loss function, N cls and N reg are the normalization factors for classification and regression, respectively, and λ is the weight balancing coefficient;

[0045] Multi-scale feature fusion strategy: A multi-scale feature fusion strategy is introduced, and a feature pyramid network (FPN) is used to process different scales of beef carcass parts, which can be expressed as:

[0046] P i =Conv(C i )+Up-sample(P i+1 );

[0047] Among them, P i is the i-th layer pyramid feature map, C i is the i-th layer convolution feature map, Cony is the convolution operation, Up-sample is the up-sampling operation, P i+1 is the i+1th layer pyramid feature map;

[0048] ROI pooling: Map the candidate region to the feature map and unify it to a fixed size through the ROI pooling layer, expressed as:

[0049] y roi (i, j) = max (m，n)∈R(i，j) y(m,n);

[0050] Among them, y roiis the output after ROI pooling, R(i, j) is the area mapped to the feature map, and y is the output feature map;

[0051] Classification and regression: Combined with multi-task learning optimization, the features after ROI pooling are classified and bounding box regressed, expressed as:

[0052] L cls (p,u)=-logp u ;

[0053] Among them, p is the predicted category probability vector, and u is the index of the true category;

[0054]

[0055] Among them, t u is the predicted bounding box parameter, v is the actual bounding box parameter, For the smooth L1 loss function, x is the horizontal coordinate of the center of the bounding box, y is the vertical coordinate of the center of the bounding box, w is the width of the bounding box, and h is the height of the bounding box.

[0056] Optionally, the position prediction includes:

[0057] State initialization: Initialize the state vector x0 and covariance matrix P0;

[0058] State prediction: based on the state vector x at the previous moment k-1 And the state transfer matrix F is used for state prediction;

[0059] Covariance prediction: based on the covariance matrix P of the previous moment k-1 And the process noise covariance matrix Q is used for covariance prediction, which is expressed as:

[0060]

[0061] Where Q is the process noise covariance matrix;

[0062] Kalman gain calculation: Calculate the Kalman gain K k , expressed as:

[0063]

[0064] Where H is the observation matrix, R is the observation noise covariance matrix;

[0065] State update: based on observation z k Update the state vector x k , expressed as:

[0066]

[0067] Among them, z k is the observed value at the current moment;

[0068] Covariance update: Update the covariance matrix P k , expressed as:

[0069]

[0070] Where I is the identity matrix.

[0071] Optionally, the posture analysis includes:

[0072] Posture initialization: Initialize the carcass posture vector θ0, including the angles in three directions (pitch angle, roll angle and yaw angle);

[0073] Posture estimation: Use the inertial measurement unit (IMU) to obtain the posture change data of the carcass in real time, expressed as:

[0074] θ k =θ k-1 +Δθ k ;

[0075] Among them, θ k is the posture vector at the current moment, Δθ k is the increment of posture change;

[0076] Posture change matrix calculation: Calculate the posture change matrix R(θ k ), describing the rotation transformation of the carcass posture, is expressed as:

[0077] R(θ k )=R yaw (θ yaw，k )R pitch (θ pitch，k )R rol l(θ roll，k );

[0078] Among them, R yaw , R pitch and R roll are the rotation matrices of yaw, pitch and roll angles respectively;

[0079] Segmentation path adjustment: According to the posture change matrix R(θ k ) adjusts the current segmentation path, expressed as:

[0080] P adjusted =R(θ k ) current ;

[0081] Among them, P adjusted is the adjusted segmentation path, P currentThe current segmentation path.

[0082] Optionally, the path correction includes:

[0083] Get the position prediction result: Get the current position prediction result p of the beef carcass based on the Kalman filter algorithm k ;

[0084] Get posture analysis results: Get the current posture change result R(θ k );

[0085] Initial segmentation path acquisition: Get the initial segmentation path P initial , make adjustments before making corrections;

[0086] Path correction calculation: Calculate the corrected segmentation path P based on the position prediction results and posture analysis results adjusted , expressed as:

[0087] P adjusted =R(θ k )·(P initial +p k );

[0088] Among them, P adjusted is the corrected segmentation path, R(θ k ) is the posture change matrix, p k is the position prediction result;

[0089] Dynamic path adjustment: real-time update and adjustment of the segmentation path, expressed as:

[0090] P final =αP adjusted +(1-α)P current ;

[0091] Among them, P final is the final adjusted segmentation path, P current is the current segmentation path, and α is the adjustment coefficient (0<α≤1).

[0092] Optionally, the execution control module includes:

[0093] Receive segmentation path: Receive the corrected segmentation path P final ;

[0094] Generate control instructions: Based on the received modified segmentation path P final , generate the corresponding motion control instruction C i , including position command, speed command, acceleration command, path command, stop command, start command and emergency stop command;

[0095] Send control command: Send the generated control command C i sent to the controller of the segmentation device;

[0096] Real-time monitoring and adjustment: Real-time monitoring of the execution of the cutting equipment, obtaining feedback information during the cutting process, including current position, speed, acceleration, cutting path deviation, cutting quality, equipment status (such as temperature, pressure, vibration), error signal, execution log and alarm information;

[0097] Adjust control parameters: Dynamically adjust control instructions C according to feedback information i .

[0098] Optionally, the temperature monitoring and feedback module includes:

[0099] Temperature sensor placement: Place temperature sensors on the splitting equipment;

[0100] Temperature data collection: real-time collection of temperature data of cutting equipment and carcass cutting parts;

[0101] Temperature data processing: process and analyze the received temperature data, calculate the average temperature and temperature change rate;

[0102] Temperature anomaly detection: detect temperature anomalies according to preset temperature thresholds and generate alarm information;

[0103] Feedback control: According to temperature data and alarm information, the control instructions of the cutting equipment are adjusted in real time.

[0104] Beneficial effects of the present invention:

[0105] The present invention, by integrating a visual acquisition and preprocessing module, a feature extraction module, a segmentation decision module, a dynamic target tracking module, an execution control module, and a temperature monitoring and feedback module, can adjust the segmentation path in real time and monitor the cutting temperature to ensure the segmentation accuracy and cutting quality, and significantly improve the segmentation efficiency. The automated segmentation operation reduces the error and labor intensity of manual operation, and improves the automation level and overall efficiency of the beef cattle carcass segmentation process.

[0106] The present invention can track the position and posture changes of a beef cattle carcass during the segmentation process in real time through a dynamic target tracking module and an execution control module, and dynamically correct the segmentation path in real time by using the results of position prediction and posture analysis to ensure cutting accuracy. The execution control module receives the corrected segmentation path and controls the segmentation equipment to perform the cutting operation of the beef cattle carcass according to the corrected segmentation path. At the same time, the temperature monitoring and feedback module monitors the temperature of the segmentation equipment and the cutting part of the carcass in real time to prevent the reduction of cutting quality and equipment damage due to overheating, thereby ensuring the safety and quality of the cutting operation.

[0107] The present invention uses a fast regional convolutional neural network image recognition algorithm to accurately extract image features, analyze dynamic changes, and generate an optimal segmentation path. The multi-scale feature fusion strategy and ROI pooling improve the detection accuracy of beef carcasses of different sizes and shapes. The real-time monitoring and dynamic adjustment mechanism ensures the high accuracy and efficiency of the system, and effectively improves the reliability and effect of automated segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0109] Figure 1 A schematic diagram of system function modules according to an embodiment of the present invention;

[0110] Figure 2 Schematic diagram of a dynamic target tracking module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0111] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0112] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0113] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0114] like Figure 1-Figure 2As shown, the beef carcass computer vision-assisted segmentation system includes a visual acquisition and preprocessing module, a feature extraction module, a segmentation decision module, a dynamic target tracking module, an execution control module, and a temperature monitoring and feedback module, wherein;

[0115] The visual acquisition and preprocessing module acquires the image data of the beef cattle carcass and performs preprocessing on the acquired image data, including noise filtering, image enhancement and preliminary identification of segmented areas;

[0116] The feature extraction module extracts feature points and edge information from the preprocessed image data;

[0117] The segmentation decision module analyzes the extracted feature points and edge information to generate a segmentation path;

[0118] The dynamic target tracking module tracks the position and posture changes of the beef carcass in the segmentation process in real time and corrects the generated segmentation path, including:

[0119] Target recognition: Use image recognition algorithms to detect key parts of beef cattle carcasses in real time;

[0120] Position prediction: Predict the movement trajectory of beef carcasses based on the Kalman filter algorithm;

[0121] Posture analysis: adjust the segmentation path by analyzing the posture changes of the carcass;

[0122] Path correction: According to the results of position prediction and posture analysis, the segmentation path is dynamically corrected in real time to ensure cutting accuracy;

[0123] The execution control module receives the corrected segmentation path and controls the segmentation equipment to perform a cutting operation on the beef carcass according to the corrected segmentation path;

[0124] The temperature monitoring and feedback module monitors the temperature of the cutting equipment and carcass cutting parts in real time to avoid the reduction of cutting quality and equipment damage caused by overheating;

[0125] Through the above content, efficient, accurate and intelligent beef cattle carcass cutting operation is realized. The system can adjust the cutting path and monitor the cutting temperature in real time to ensure the cutting accuracy and cutting quality, which significantly improves the cutting efficiency, reduces the errors and labor intensity of manual operation, and improves the automation level and overall benefits of the beef cattle carcass cutting process.

[0126] The visual acquisition and preprocessing modules include:

[0127] Multi-angle image acquisition: Use multiple cameras, arranged at different angles, to collect image data of beef cattle carcasses in real time to ensure that complete image data of beef cattle carcasses is collected;

[0128] Noise filtering: Adaptive median filtering algorithm is used to filter the noise of the collected image data, which is expressed as:

[0129]

[0130] Where f(i, j) is the grayscale value of pixel (i, j), (S(i, j)) is the pixel set in the neighborhood of pixel (i, j), median(S(i, j)) is the median of the neighborhood pixels, and T is the preset threshold;

[0131] Image enhancement: The histogram equalization algorithm is used to enhance the image data after noise filtering, which is expressed as:

[0132]

[0133] Where f′(i, j) is the pixel value of the enhanced image, L is the grayscale level, MN is the total number of pixels in the image, and h(k) is the number of pixels with grayscale value k;

[0134] Preliminary identification of segmented regions: The 0tsu threshold segmentation algorithm is used to perform preliminary identification of segmented regions on the enhanced image data, expressed as:

[0135]

[0136] in, is the inter-class variance, ω1(t) and ω2(t) are the probabilities of foreground and background, μ1(t) and μ2(t) are the average grayscale values ​​of foreground and background, and t is the threshold;

[0137] Through the above steps, noise interference is effectively reduced, image quality is enhanced, segmented areas are accurately identified, and high-quality image data is provided for subsequent processing, thereby improving the segmentation accuracy and efficiency of the system.

[0138] The feature extraction module includes:

[0139] Feature point extraction: The scale-invariant feature transform (SIFT) algorithm is used to extract feature points from the preprocessed image data, expressed as:

[0140] D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))*I(x,y);

[0141] Where D(x, y, σ) is the difference function in the scale space, G(x, y, σ) is the Gaussian function in the scale space, I(x, y) is the input image, and k is a constant;

[0142] Edge information extraction: The Canny edge detection algorithm is used to perform edge detection on the preprocessed image data, including:

[0143] Smooth the image: Apply a Gaussian filter to smooth the image, expressed as:

[0144]

[0145] Where G(x, y) is a Gaussian filter and σ is the standard deviation;

[0146] Calculate image gradient: Use the Sobel operator to calculate the image gradient, expressed as:

[0147]

[0148] Among them, G x and G y are the gradients in the horizontal and vertical directions respectively;

[0149] Compute the gradient magnitude and direction:

[0150] Apply non-maximum suppression to remove spurious responses: Non-maximum suppression compares the pixel with its neighboring pixels in the gradient direction and retains the local maximum value, which is expressed as:

[0151]

[0152] Apply double threshold detection and edge connection to generate the final edge map: Use two thresholds T low and T high Perform edge detection, expressed as:

[0153]

[0154] Through the above content, the accuracy and reliability of image feature extraction are ensured, providing high-quality data support for subsequent segmentation decisions, greatly improving the accuracy of the segmentation path and the overall efficiency of the system.

[0155] The segmentation decision module includes:

[0156] Feature map generation: Convolutional neural network (CNN) is used to generate feature maps of input feature points and edge information, which can be expressed as:

[0157] y=f(W*x+b);

[0158] Among them, y is the output feature map, W is the convolution kernel, x is the input feature point and edge information, b is the bias, and f is the activation function;

[0159] Data Analysis: Time series data is analyzed using a recurrent neural network (RNN) to generate a dynamic segmentation path, represented as:

[0160] h t=tanh(W h h t-1 +W x x t + b);

[0161] Among them, h t is the hidden state at the current moment, h t-1 is the hidden state at the previous moment, W h and W x are weight matrices, x t is the input at the current moment, b is the bias;

[0162] y t =W y h t +b y ;

[0163] Among them, y t is the output at the current moment, W y is the output weight matrix, b y is the bias of the output layer;

[0164]

[0165] Where n is the number of samples, y i is the actual value, is the predicted value;

[0166] Segmentation path generation: Combining the results of feature map generation and data analysis, the optimal segmentation path is generated through an optimization algorithm, which is expressed as:

[0167]

[0168] Among them, L is the loss function, y i is the actual output, is the predicted output, λ is the regularization parameter, W j is the weight matrix;

[0169] Through the above content, the spatial features of the image are effectively extracted, and the recurrent neural network captures the dynamic changes in the time series. The system can dynamically adjust the segmentation path to ensure high-precision and high-efficiency beef carcass segmentation, greatly improving the reliability and effect of automated segmentation.

[0170] Convolutional Neural Network (CNN) specifically includes:

[0171] Convolution operation:

[0172] Where W is the convolution kernel, size is M×N, x is the input image, (i, j) is the output position, and m and n are the indices of the convolution kernel;

[0173] Add bias: y(i, j) = (W*x)(i, j) + b;

[0174] Where, b is the bias;

[0175] Apply activation function: y′=f(y);

[0176] Where, f(z)=max(0,z);

[0177] Pooling layer: The maximum pooling operation is performed after the convolution layer, expressed as:

[0178] y pool (i, j) = max m，n y′(i+m,j+n);

[0179] Among them, y pool is the output after pooling, m, n are the indexes of the pooling windows.

[0180] The image recognition algorithm uses the Faster R-CNN algorithm, which includes:

[0181] Region Proposal Network: A region proposal network (RPN) is used to generate candidate regions, combining multi-scale features to adapt to different sizes of beef carcass parts, expressed as:

[0182]

[0183] Among them, p i is the category prediction probability of the i-th anchor point, is the actual category of the i-th anchor point, t i is the regression prediction of the i-th anchor point, is the actual regression value of the i-th anchor point, L cls is the classification loss function, L reg is the regression loss function, N cls and N reg are the normalization factors for classification and regression, respectively, and λ is the weight balancing coefficient;

[0184] Multi-scale feature fusion strategy: A multi-scale feature fusion strategy is introduced, and a feature pyramid network (FPN) is used to process different scales of beef carcass parts, which can be expressed as:

[0185] P i =Conv(C i )+Up-sample(P i+1 );

[0186] Among them, P iis the i-th layer pyramid feature map, C i is the i-th layer convolution feature map, Conv is the convolution operation, Up-sample is the up-sampling operation, P i+1 is the i+1th layer pyramid feature map;

[0187] ROI pooling: Map the candidate region to the feature map and unify it to a fixed size through the ROI pooling layer, expressed as:

[0188] y roi (i, j) = max (m，n)∈R(i，j) y(m,n);

[0189] Among them, y roi is the output after ROI pooling, R(i, j) is the area mapped to the feature map, and y is the output feature map;

[0190] Classification and regression: Combined with multi-task learning optimization, the features after ROI pooling are classified and bounding box regressed, expressed as:

[0191] L cls (p,u)=-logp u ;

[0192] Among them, p is the predicted category probability vector, and u is the index of the true category;

[0193]

[0194] Among them, t u is the predicted bounding box parameter, v is the actual bounding box parameter, is the smooth L1 loss function, x is the horizontal coordinate of the center of the bounding box, y is the vertical coordinate of the center of the bounding box, w is the width of the bounding box, and h is the height of the bounding box;

[0195] The smooth L1 loss function is expressed as:

[0196] Through the above content, the detection accuracy of different sizes and shapes of beef cattle carcasses is effectively improved. In addition, ROI pooling and classification and regression ensure the accuracy and reliability of the segmentation path, significantly improve the segmentation efficiency and accuracy of the system, reduce the errors and labor intensity of manual operations, and realize efficient, accurate and intelligent beef cattle carcass segmentation.

[0197] Location predictions include:

[0198] State initialization: Initialize the state vector x0 and covariance matrix P0, expressed as:

[0199]

[0200] Among them, x0 and y0 are the initial positions, v x and v y is the initial velocity, σ x , σ y , and are the initial uncertainties of position and velocity, respectively;

[0201] State prediction: based on the state vector x at the previous moment k-1 And the state transfer matrix F is used for state prediction, which is expressed as:

[0202]

[0203]

[0204] in, is the predicted state vector, Δt is the time interval;

[0205] Covariance prediction: based on the covariance matrix P of the previous moment k-1 And the process noise covariance matrix Q is used for covariance prediction, which is expressed as:

[0206]

[0207] Where Q is the process noise covariance matrix;

[0208] Kalman gain calculation: Calculate the Kalman gain K k , expressed as:

[0209]

[0210] Where H is the observation matrix, R is the observation noise covariance matrix;

[0211] State update: based on observation z k Update the state vector x k , expressed as:

[0212]

[0213] Among them, z k is the observed value at the current moment;

[0214] Covariance update: Update the covariance matrix P k , expressed as:

[0215]

[0216] Where I is the identity matrix;

[0217] Through the above content, the system can accurately predict the movement trajectory of the beef cattle carcass, thereby providing accurate dynamic adjustment basis for subsequent segmentation decisions, ensuring dynamic adjustment and high-precision execution of the segmentation path, and improving the overall performance and reliability of the system.

[0218] Posture analysis includes:

[0219] Posture initialization: Initialize the carcass posture vector θ0, including the angles in three directions (pitch angle, roll angle and yaw angle), expressed as:

[0220]

[0221] Posture estimation: Use the inertial measurement unit (IMU) to obtain the posture change data of the carcass in real time, expressed as:

[0222] θ k =θ k-1 +Δθ k ;

[0223] Among them, θ k is the posture vector at the current moment, Δθ k is the increment of posture change;

[0224] Posture change matrix calculation: Calculate the posture change matrix R(θ k ), describing the rotation transformation of the carcass posture, is expressed as:

[0225] R(θ k )=R yaw (θ yaw，k )R pitch (θ pitch，k )R roll (θ roll，k );

[0226] Among them, R yaw , R pitch and R roll are the rotation matrices of yaw, pitch and roll angles respectively;

[0227] Segmentation path adjustment: According to the posture change matrix R(θ k ) adjusts the current segmentation path, expressed as:

[0228] P adjusted =R(θ k ) current ;

[0229] Among them, P adjusted is the adjusted segmentation path, P current is the current segmentation path;

[0230] Through the above steps, it is ensured that the segmentation path can dynamically adapt to the posture changes of the carcass, improve the accuracy and efficiency of segmentation, effectively deal with the posture changes that may occur in the carcass during the segmentation process, and ensure the accuracy and continuity of the segmentation operation.

[0231] Path corrections include:

[0232] Get the position prediction result: Get the current position prediction result p of the beef carcass based on the Kalman filter algorithm k , expressed as:

[0233]

[0234] Among them, x k and k is the predicted position at the current moment, v x，k and v y，k is the predicted speed at the current moment;

[0235] Get posture analysis results: Get the current posture change result R(θ k );

[0236] Initial segmentation path acquisition: Get the initial segmentation path P initial , adjusted before correction, expressed as:

[0237]

[0238] where x1, x2, ..., x n and y1,y2,...,y n are the coordinates of each point on the initial segmentation path;

[0239] Path correction calculation: Calculate the corrected segmentation path P based on the position prediction results and posture analysis results adjusted , expressed as:

[0240] P adjusted =R(θ k )·(P initial +p k );

[0241] Among them, P adjusted is the corrected segmentation path, R(θ k ) is the posture change matrix, p k is the position prediction result;

[0242] Dynamic path adjustment: The segmentation path is updated and adjusted in real time to ensure the accuracy and continuity of the segmentation operation, which is expressed as:

[0243] P final =αPadjusted +(1-α)P current ;

[0244] Among them, P final is the final adjusted segmentation path, P current is the current segmentation path, α is the adjustment coefficient (0<α≤1);

[0245] Through the above steps, the displacement and posture changes of the carcass during the cutting process can be effectively dealt with, the accuracy and continuity of the cutting path can be ensured, and the precision and efficiency of the cutting operation can be improved.

[0246] The execution control module includes:

[0247] Receive segmentation path: Receive the corrected segmentation path P final ;

[0248] Generate control instructions: Based on the received modified segmentation path P final , generate the corresponding motion control instruction C i , including position command, speed command, acceleration command, path command, stop command, start command and emergency stop command;

[0249] Send control command: Send the generated control command C i sent to the controller of the segmentation device;

[0250] Real-time monitoring and adjustment: Real-time monitoring of the execution of the cutting equipment, obtaining feedback information during the cutting process, including current position, speed, acceleration, cutting path deviation, cutting quality, equipment status (such as temperature, pressure, vibration), error signal, execution log and alarm information;

[0251] Adjust control parameters: Dynamically adjust control instructions C according to feedback information i ;

[0252] Through the above content, feedback information is obtained and the control instructions are dynamically adjusted according to this information to ensure the accuracy and continuity of the cutting path and cutting quality, so as to efficiently and accurately control the cutting equipment and complete the cutting operation of the beef carcass.

[0253] The temperature monitoring and feedback module includes:

[0254] Temperature sensor placement: Place temperature sensors on the splitting equipment;

[0255] Temperature data collection: real-time collection of temperature data of cutting equipment and carcass cutting parts;

[0256] Temperature data processing: Process and analyze the received temperature data, calculate the average temperature and temperature change rate, expressed as:

[0257]

[0258] ΔT device (t) = T device，avg (t)-T device，avg (t-Δt);

[0259] ΔT cutting (t) = T cutting，avg (t)-T cutting，avg (t-Δt);

[0260] Among them, T device，avg (t) and T cutting，avg (t) are the average temperatures of the equipment and cutting parts, ΔT device (t) and ΔT culting (t) are the temperature change rates;

[0261] Temperature anomaly detection: Detect temperature anomalies according to the preset temperature threshold and generate alarm information, which is expressed as:

[0262]

[0263] Among them, T device，threshold and T cutting，threshold They are the temperature thresholds of the equipment and the cutting part, respectively. device (t) and Alarm cutting (t) is the abnormal temperature alarm signal;

[0264] Feedback control: According to temperature data and alarm information, the control instructions of the cutting equipment are adjusted in real time to ensure the safety and quality of the cutting operation;

[0265] Through the above content, the control instructions of the cutting equipment are adjusted in real time to ensure the safety and quality of the cutting operation. The temperature of the equipment and carcass during the cutting process can be effectively monitored and adjusted to prevent the reduction of cutting quality and equipment damage due to overheating.

[0266] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0267] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. Computer vision-assisted segmentation system for beef cattle carcasses, It is characterized in that It includes visual acquisition and preprocessing module, feature extraction module, segmentation decision module, dynamic target tracking module, execution control module and temperature monitoring and feedback module, among which; The visual acquisition and preprocessing module acquires image data of the beef cattle carcass and preprocesses the acquired image data, including noise filtering, image enhancement and preliminary identification of segmented areas; The feature extraction module extracts feature points and edge information from the preprocessed image data; The segmentation decision module analyzes the extracted feature points and edge information to generate a segmentation path; The dynamic target tracking module tracks the position and posture changes of the beef carcass in the segmentation process in real time and corrects the generated segmentation path, specifically including: Target recognition: Use image recognition algorithms to detect key parts of beef cattle carcasses in real time; Position prediction: Predict the movement trajectory of beef carcasses based on the Kalman filter algorithm; Posture analysis: adjust the segmentation path by analyzing the posture changes of the carcass; Path correction: According to the results of position prediction and posture analysis, the segmentation path is dynamically corrected in real time; The execution control module receives the corrected segmentation path and controls the segmentation equipment to perform a cutting operation on the beef carcass according to the corrected segmentation path; The temperature monitoring and feedback module monitors the temperature of the cutting equipment and the carcass cutting parts in real time.

2. The beef cattle carcass computer vision-assisted segmentation system according to claim 1, It is characterized in that The visual acquisition and preprocessing module includes: Multi-angle image acquisition: Use multiple cameras, arranged at different angles, to collect image data of beef carcasses in real time; Noise filtering: Adaptive median filtering algorithm is used to filter noise from the collected image data; Image enhancement: The histogram equalization algorithm is used to enhance the image data after noise filtering; Preliminary identification of segmented areas: The Otsu threshold segmentation algorithm is used to perform preliminary identification of segmented areas on the enhanced image data.

3. The beef cattle carcass computer vision-assisted segmentation system according to claim 1, It is characterized in that The feature extraction module comprises: Feature point extraction: The scale-invariant feature transformation algorithm is used to extract feature points from the preprocessed image data, which is expressed as: D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))*I(x,y); Where D(x, y, σ) is the difference function in the scale space, G(x, y, σ) is the Gaussian function in the scale space, I(x, y) is the input image, and k is a constant; Edge information extraction: The Canny edge detection algorithm is used to perform edge detection on the preprocessed image data.

4. The beef cattle carcass computer vision-assisted segmentation system according to claim 1, It is characterized in that The segmentation decision module comprises: Feature map generation: Convolutional neural network is used to generate feature maps of input feature points and edge information, which can be expressed as: y=f(W*x+b); Among them, y is the output feature map, W is the convolution kernel, x is the input feature point and edge information, b is the bias, and f is the activation function; Data Analysis: Time series data is analyzed using a recurrent neural network to generate a dynamic segmentation path, represented as: h t =tanh(W h h t-1 +W x x t +b); Among them, h t is the hidden state at the current moment, h t-1 is the hidden state at the previous moment, W h and W x are weight matrices, x t is the input at the current moment, b is the bias; y t =W y h t +b y ; Among them, y t is the output at the current moment, W y is the output weight matrix, b y is the bias of the output layer; Where n is the number of samples, y i is the actual value, is the predicted value; Segmentation path generation: Combining the results of feature map generation and data analysis, the optimal segmentation path is generated through an optimization algorithm, which is expressed as: Among them, L is the loss function, y i is the actual output, is the predicted output, λ is the regularization parameter, W j is the weight matrix.

5. The beef cattle carcass computer vision-assisted segmentation system according to claim 4, It is characterized in that The image recognition algorithm adopts a fast regional convolutional neural network algorithm, and the fast regional convolutional neural network algorithm includes: Region Proposal Network: The candidate regions are generated using the region proposal network, combining multi-scale features to adapt to different sizes of beef carcass parts, expressed as: Among them, p i is the category prediction probability of the i-th anchor point, is the actual category of the i-th anchor point, t i is the regression prediction of the i-th anchor point, is the actual regression value of the i-th anchor point, L cls is the classification loss function, L reg is the regression loss function, N cls and N reg are the normalization factors for classification and regression, respectively, and λ is the weight balancing coefficient; Multi-scale feature fusion strategy: A multi-scale feature fusion strategy is introduced, and a feature pyramid network is used to process different scales of beef carcass parts, which can be expressed as: P i =Conv(C i )+Up-sample(P i+1 ); Among them, P i is the i-th layer pyramid feature map, C i is the i-th layer convolution feature map, Conv is the convolution operation, Up-sample is the up-sampling operation, P i+1 is the i+1th layer pyramid feature map; ROI pooling: Map the candidate region to the feature map and unify it to a fixed size through the ROI pooling layer, expressed as: y roi (i,j)=max (m，n)∈R(i，j) y(m,n); Among them, y roi is the output after ROI pooling, R(i, j) is the area mapped to the feature map, and y is the output feature map; Classification and regression: Combined with multi-task learning optimization, the features after ROI pooling are classified and bounding box regressed, expressed as: L cls (p, u)=-logp u ; Among them, p is the predicted category probability vector, and u is the index of the true category; Among them, t u is the predicted bounding box parameter, v is the actual bounding box parameter, For smoothing L 1 Loss function, x is the horizontal coordinate of the center of the bounding box, y is the vertical coordinate of the center of the bounding box, w is the width of the bounding box, and h is the height of the bounding box.

6. The beef cattle carcass computer vision-assisted segmentation system according to claim 5, It is characterized in that The position prediction includes: State initialization: Initialize the state vector x 0 and the covariance matrix P 0 ; State prediction: based on the state vector x at the previous moment k-1 And the state transfer matrix F is used for state prediction; Covariance prediction: based on the covariance matrix P of the previous moment k-1 And the process noise covariance matrix Q is used for covariance prediction, which is expressed as: Where Q is the process noise covariance matrix; Kalman gain calculation: Calculate the Kalman gain K k , expressed as: Among them, H is the observation matrix, R is the observation noise covariance matrix; State update: based on observation z k Update the state vector x k , expressed as: Among them, z k is the observed value at the current moment; Covariance update: Update the covariance matrix P k , expressed as: Where I is the identity matrix.

7. The beef cattle carcass computer vision-assisted segmentation system according to claim 6, It is characterized in that The posture analysis includes: Posture initialization: Initialize the carcass posture vector θ 0 , including angles in three directions; Posture estimation: Use the inertial measurement unit to obtain the posture change data of the carcass in real time, expressed as: i k =θ k-1 +Δθ k ; Among them, θ k is the posture vector at the current moment, Δθ k is the increment of posture change; Posture change matrix calculation: Calculate the posture change matrix R(θ k ), describing the rotation transformation of the carcass posture, is expressed as: R(θ k )=R yaw (i yaw,k )R pitch (i pitch,k )R roll (i roll，k ); Among them, R yaw , R pitch and R roll are the rotation matrices of yaw, pitch and roll angles respectively; Segmentation path adjustment: According to the posture change matrix R(θ k ) adjusts the current segmentation path, expressed as: P.S adjusted NR(θ k )P current 100. Among them, P adjusted is the adjusted segmentation path, P current The current segmentation path.

8. The beef cattle carcass computer vision-assisted segmentation system according to claim 7, It is characterized in that The path correction includes: Get the position prediction result: Get the current position prediction result p of the beef carcass based on the Kalman filter algorithm k ; Get posture analysis results: Get the current posture change result R(θ k ); Initial segmentation path acquisition: Get the initial segmentation path P initial , make adjustments before making corrections; Path correction calculation: Calculate the corrected segmentation path P based on the position prediction results and posture analysis results adjusted , expressed as: P.S adjusted NR(θ k )·(P initial +p k )4 Among them, P adjusted is the corrected segmentation path, R(θ k ) is the posture change matrix, p k is the location prediction result; Dynamic path adjustment: real-time update and adjustment of the segmentation path, expressed as: P final =αP adjusted +(1-α)P current ; Among them, P final is the final adjusted segmentation path, P current is the current segmentation path, and α is the adjustment coefficient.

9. The beef cattle carcass computer vision-assisted segmentation system according to claim 8, It is characterized in that The execution control module comprises: Receive segmentation path: Receive the corrected segmentation path P final ; Generate control instructions: Based on the received modified segmentation path P final , generate the corresponding motion control instruction C i , including position command, speed command, acceleration command, path command, stop command, start command and emergency stop command; Send control command: Send the generated control command C i sent to the controller of the segmentation device; Real-time monitoring and adjustment: Real-time monitoring of the execution of the cutting equipment, obtaining feedback information during the cutting process, including current position, speed, acceleration, cutting path deviation, cutting quality, equipment status, error signal, execution log and alarm information; Adjust control parameters: Dynamically adjust control instructions C according to feedback information i .

10. The beef cattle carcass computer vision-assisted segmentation system according to claim 9, It is characterized in that The temperature monitoring and feedback module includes: Temperature sensor placement: Place temperature sensors on the splitting equipment; Temperature data collection: real-time collection of temperature data of cutting equipment and carcass cutting parts; Temperature data processing: process and analyze the received temperature data, calculate the average temperature and temperature change rate; Temperature anomaly detection: detect temperature anomalies according to preset temperature thresholds and generate alarm information; Feedback control: According to temperature data and alarm information, the control instructions of the cutting equipment are adjusted in real time.

Images