Research And Experiment Of Sheep Carcass Segmentation Path Recognition Method

The vigorous development of the domestic economy has greatly improved the material living standards of the people,among which the increasing demand for high-quality meat and poultry products is particularly prominent.The automation of livestock and poultry processing can effectively accelerate the production efficiency of livestock and poultry meat and improve the production quality of livestock and poultry products,thus meeting the urgent demand of livestock and poultry processing enterprises for carcass segmentation devices and improving enterprise profits.At present,the sheep carcass slaughter equipment in China is outdated,and the vast majority of the processing process relies on manual participation.In view of the insufficient sheep carcass segmentation technology,the low degree of automation of segmentation equipment,and the poor segmentation accuracy and efficiency,this paper proposed to study the critical problem of sheep carcass segmentation path identification in automatic sheep carcass segmentation technology.Based on the image semantic segmentation neural network model and regression prediction model,a method for identifying the segmentation path of sheep carcasses was proposed in this paper.And an image processing system and a stepper motor control system were designed to recognize the segmentation path of sheep carcass.Finally,The feasibility of the proposed sheep carcass segmentation path identification method was verified by knife walking test of the robotic arm.The main research content of this paper is as follows:(1)The recognition of key parts of sheep carcass images(lamb ribs)was completed.Based on the underlying theory and structural features of the image semantic segmentation network Deeplab v3+,the Deeplab v3+ network model was built using the Py Torch deep learning framework,and the sheep carcass image dataset was built and trained.The lightweight network Mobilenet v2 was applied to replace the Xception feature extraction network of the original model to perform an image detection speed optimization test of the sheep carcass key part recognition model.Comparing with other commonly used image semantic segmentation networks and testing the generalization capability of the model,the Deeplab v3+-Mobilenet v2 network model outperformed other models in terms of segmentation accuracy and detection efficiency with PA,MIo U and single image detection times of 98.29%,95.07% and 9.02 ms,respectively.(2)The prediction of sheep carcass segmentation path was completed.The sheep carcass segmentation method was determined by NY/T 1564-2007 "Technical Specification for Lamb Segmentation" and the commonly used sheep carcass segmentation standards.Based on the theoretical basis of Gaussian process regression in machine learning,a Gaussian process regression prediction model was developed using Matlab 2020 b.The obtained lamb rib chops feature information was used as the model feature input value,and the lamb carcass segmentation path was predicted after eliminating the interference information that was not related to the output value.Then a comparison test between the GPR model and other commonly used regression prediction models was conducted.The test results showed that the R2,MAE and RMSE values of the Gaussian process regression model were 0.984900,0.057574 and0.073665,respectively,which had smaller prediction errors,and the model was more suitable for sheep carcass segmentation path prediction than other prediction models.Finally,based on the spatial coordinate conversion theory,the predicted segmentation coordinates were transformed into spatial three-dimensional coordinates,which laid a good foundation for the subsequent validation tests.(3)The control system of the automatic sheep carcass segmentation platform was designed.In the hardware design section,the selection and wiring of cameras,STM32 microcontrollers,stepper motors,motor drivers and other equipment were completed.In the software design part,the Py Qt5-based upper computer interface design and Keil5-based stepper motor control algorithm design were completed.And the information transfer between the upper computer and the microcontroller was realized by using the serial communication.(4)The knife walking test of the automatic sheep carcass segmentation platform was completed.In order to verify the feasibility of the designed image processing system and stepper motor control system,we designed and completed the knife walking test of sheep carcass based on the existing automatic sheep carcass dividing platform.In the experimental results,the PA and MIo U values of the semantic segmentation network model in the image processing system were 98.242% and95.411%,respectively,and the errors of the six feature values output by the regression prediction model were small,and none of the regression loss rate exceeds 2%,and the average time taken by the system was 1.1152 s.The microcontroller in the stepper motor control system could receive the segmentation path coordinate sequence in real time,and the robot arm could complete the knife walking test precisely according to the received coordinate sequence.The test showed that the designed image processing system and stepper motor control system can effectively complete the task of sheep carcass segmentation and achieve the expected goal of the project.