基于深度学习的农作物病害检测算法研究

打开文本图片集

中图分类号：S126 文献标识码：A 文章编号：2095-5553（2025）10-0230-06

Abstract：To address the ineficiency incrop disease detection，several improvements were made to theYOLOv5 deep learningalgorithm.TheK—means algorithm was applied toadjust thelabelanchorboxes，enhancing themodel’sadaptability tothevaryingsizesofdisease-affectedareasinthetrainingdataset.Additionally，theGhostNetmodulewas integratedinto theYOLOv5 backbone network，reducing computational complexityand the numberof parameters，while maintaining the qualityoffeature extraction.Furtherenhancements wereachieved by introducing parallelconnection structuresandresidual conections basedonthechanneland spatialattentionmechanisms inthe CBAMatention module.These modifications， incorporatedintotheNeckpartof thenetwork，significantlyboosted itsfeatureextractioncapabilies.Asaresultof these improvements，the model size of the enhanced YOLOv5 was reduced from 13.8MB to11.1 MB.The modified model achieved a mean average precision （mAP@0.5） of 90.7% ，representing a 3.2% improvement over the original YOLOv5. Furthermore，the F1 score increased by 2.1% . These enhancements provide a practical and efficient solution for crop disease detection，enabling accurate classification and localization with high precision and fast detection speed.

Keywords：crop；disease detection；attention mechanism；deep learning；dataaugmentation；clustering algorithms

0 引言

每年由于农业病害、自然灾害等原因，经济损失可达数十亿元。（剩余9967字）