| With the progress of social economy and the improvement of people ’s living standards,consumers pay more and more attention to the high quality,diversity and refinement of food.However,the existing livestock quality detection and weight grading devices in China are low in intelligence,low in accuracy and slow in efficiency,and lack of large-scale pipeline produ ction devices.Therefore,through market research and literature review,this study summarize s the current situation of quality detection,dynamic weighing and quality grading devices at h ome and abroad.Taking chicken wings as the research object,the research content is related t o the non-destructive detection and weight grading technology of chicken wings,and the nondestructive detection and weight grading device of chicken wings based on deep learning algo rithm is designed.The main research work includes the following aspects :(1)A non-destructive testing and weight grading device for chicken wings was designed.The device is divided into four parts : non-destructive testing device,dynamic weighing devi ce,flipping device and intelligent cross sorting device.Solid Works is used to build the overall three-dimensional model of the device,design its size and selection,and perform static stress analysis on key parts to verify its rationality.(2)Chicken wing quality detection based on YOLOv5 s algorithm.Firstly,the data set of chicken wings is established by industrial camera,then the data set is expanded by data enhan cement and the model is trained.The model will detect whether the skin has tears,bruises,stai ns,feather residues,blood spots and other defects to determine the quality of chicken wings b ased on the integrity of the carcass surface.Experiments show that the recognition accuracy a nd detection accuracy of the model need to be improved.(3)An improved algorithm YOLO-CD is proposed.Firstly,for the backbone network,th ree lightweight networks are selected for experimental comparison and analysis.Finally,Shuf fle Netv2 is used to replace the backbone network to speed up the detection algorithm.Secondl y,for the backbone network,three different types of attention mechanisms are used for experi mental comparative analysis to enhance the feature expression ability,improve the algorithm positioning ability,and select the CA attention mechanism to improve the detection accuracy of the algorithm.Finally,the improved algorithm YOLO-CD is successfully deployed on the Jetson Nano embedded platform.Compared with different one-stage algorithms,YOLO-CD p erforms better in detection accuracy,detection speed and model complexity.(4)Hardware and software design of device control system.In order to effectively reduc e the weighing error caused by mechanical vibration,the dynamic weighing module uses ’ am plitude limiting debounce filtering ’ for digital filtering.At the same time,a human-computer i nteraction interface is designed to display the weighing value in real time,realize parameter se tting and fault warning.Experiments show that the device meets the actual production require ments.(5)Carry out relevant test verification.The accuracy of non-destructive testing,the accur acy of dynamic weighing and the accuracy of intelligent grading were tested respectively.The white feather chicken wings were selected as the experimental object.According to the Chine se national standard of livestock meat quality grading,the detailed parameters of the prototyp e were tested. |
