Research on Soybean and Maize Seeds Counting Method Based on Computer Vision
[Objective]The weight of crop seeds is one of the important factors in yield composition,and the traditional calculation process for hundred-grain weight/thousand-grain weight is time-consuming and laborious,urgently requiring a fast method for measuring the number of crop seeds and calculating weight.[Method]This article focuses on soybeans and maize as research objects.Firstly,addressing is-sues such as complex seed counting environments,small targets,and high density,the Albumentations library is used to enhance the dataset;Then,by comparing the five sub-models of YOLOv8,the best performing YOLOv8n model was selected.Based on this,Focal-IOU was used to replace the CIOU loss function,resulting in an improved model;Finally,the improved model was compared with various clas-sic object detection models.[Result]The results showed that the average accuracy mAP50-95 of the im-proved models for soybean and maize seeds datasets reached 88.78%and 86.89%,respectively,which was 1.29%and 0.51%higher than the original model.The performance was significantly better than other object detection models such as YOLOv5 and SSD.In addition,the Mean Absolute Percentage Error(MAPE)of the improved model on the two crop test sets were 0.035%and 0.045%,respectively,and the frame rates per second respectively reached 70.17 and 100.41,respectively.[Conclusion]In conclu-sion,the difference between the results of the improved model and the actual quantity in soybean and corn seeds counting is not significant.This model has a fast-real-time processing speed for data.The re-search results can meet the needs of seed counting in the calculation of hundred-grain weight/thousand-grain weight in seed testing.
NETL
NSTL
万方数据
