Research On Quality Detection And Weight Grading Of Chicken Wings Based On Deep Learning

With the development of social economy and the improvement of people's living standards,consumers' demand for food tends to be high quality,diversified,and refined.However,at present,our country's quality detection and weight grading for livestock and poultry mainly adopts manual auxiliary pipeline operation,with low level of mechanization and intelligence.In view of the low intelligence level of the existing livestock and poultry weight classification equipment in our country,I take the chicken wings as the research object and make the research on the non-destructive testing and weight grading technology of chicken wings based on deep learning.At first I design the double-sided non-destructive testing model of chicken wing quality based on Mobile Netv2-YOLOv4 target detection algorithm and apply it to the Jetson Xavier NX embedded platform.At the same time,the mathematical model of dynamic weighing and intelligent grading is established for chicken wing flip device with adaptive conveyor belt speed,device of dynamic weighing and intelligent grading device.At last,the prototype is optimized and tested.The main work of this study includes the following aspects:(1)Mechanical key parts design and analysis.According to the requirement of doublesided detection,chicken wing flip device which adapts to the speed of conveyor belt,was designed to realize the online flip with high-throughput.Step motors,sliding tables and push rod were used to optimize the chicken wing automatic grading device,which greatly improved the response speed of grading system.At the same time,the static analysis of the key components of the flipper and push rod was carried out.(2)A non-destructive testing model of chicken wing double-sided quality based on Mobile Netv2-YOLOv4 target detection algorithm was designed.According to the characteristics of carcass integrity,abnormal color spots and feather residual state,I built the chicken wing data.The test results showed that the model had higher recognition accuracy and faster recognition speed,and the model was deployed on the Jetson Xavier NX embedded platform.(3)Algorithm optimization.The dynamic weighing module adopted “the limiting filtering method” and “the maximum and minimum sliding window mean filtering method” for digital filtering.And the effectiveness of the filtering method was proved as valid by dynamic test,which effectively reduced the weighing error caused by mechanical vibration.At the same time,the chicken wing grading control model based on PID algorithm,was designed to improve the response speed of the grading system and the anti-interference ability of the control system.(4)Design for the human-computer interaction interface.The dynamic weighing value and the total amount of chicken wing grading were displayed in real time by the host computer.The functions of parameter setting and fault warning were realized.The results of prototype test showed that the parameters such as grading standard and grading speed can be set by the host computer in the production process.The accuracy of wing turnover was above 99.2%.The average accuracy of the non-destructive testing model was94.80%.The average recall rate was 96.86%,and the missed detection rate was 0%,which showed good robustness.In order to take the accuracy and grading efficiency into account,when the grading speed was set as 4 per second,the mixed grading accuracy reached 98.0%,and 14400 chicken wings were graded per hour,which can meet the market demand of rapid industrial development and food quality safety.