Application Of YOLOv5 Algorithm In Sweet Potato Surface Defect Detection

Sweet potato is one of the main food crops in China,which has high nutritional value,anti-inflammatory and anti-oxidation effects,and has gradually become a key-keeping food that people increasingly favor in recent years.The quality of sweet potato mainly includes nutritional quality,appearance quality,etc.The appearance quality of sweet potato can be divided into shape,size,color,surface defects,etc.The most important division of appearance quality is surface defects,and the most common surface defects are breakages and rots.In the process of harvesting,transportation and storage,surface defects such as breakages and rots will inevitably appear,which will affect the quality of sweet potato.At present,the surface defects of sweet potato are detected by manual sorting method,which has some problems such as high work intensity,low efficiency and inconsistent standards,which is not conducive to the development of sweet potato processing automation industry.In this study,YOLOv5 target detection algorithm is applied to the detection of surface breakages and rots of sweet potato for the first time,and a system for surface defect detection and appearance quality grading of sweet potato is developed to provide a new method for the development of sweet potato processing automation industry.YOLOv5 algorithm is chosen mainly because of its high speed,small volume and high accuracy,which can be deployed in practical application scenarios in the future.The main contents of this study are as follows:(1)Construction of data set.The images of sweet potato with two surface defects,breakage and rot,were collected by ourselves,and the data set needed for this experiment was made through data labeling and data amplification.(2)Training and selection of YOLOv5 model for surface defect detection of sweet potato.The made data sets are trained by four models: YOLOv5 s,YOLOv5m,YOLOv5 l and YOLOv5 x,and the performance evaluation data of the four models after training are obtained.After comparative analysis,YOLOv5 s model is selected for the subsequent system development.(3)The design and implementation of sweet potato surface defect detection and appearance quality grading system.Combining the trained YOLOv5 s model with the backend frame Flask and the front-end frame Vue,a sweet potato surface defect detection and appearance quality grading system was designed and developed.