摘要
The manual picking of strawberries is inefficient and costly,limiting scalability and economic benefits.Mechanizing this process reduces labor demands,improves working conditions,and modernizes the strawberry industry.Target detection technology,crucial for mechanized picking,must accurately determine strawberry maturity.This study presents an enhanced YOLOv8s model addressing current machine learning issues like accuracy,parameters,and complexity.The improved model replaces the Bottleneck structure in C2f with the FasterNet network,integrates an efficient multi-scale attention mechanism,and uses the Ghost module in the backbone to reduce computational load while maintaining performance.It also introduces Wise-IoU for bounding box regression loss,improving recognition accuracy.The YOLOv8s-FEGW model achieves a 93.8%mAP in detecting strawberry ripeness,with significant reductions in parameters(36.8%),complexity(34.6%),and model size(37.7%),alongside a 12.7% Frames Per Second(FPS)boost.These enhancements result in excellent detection capabilities,supporting agricultural automation and intelligence.
基金
funded by the National Engineering Research Center of Special Equipment and Power System for Ship and Marine Engineering and the Shanghai Engineering Research Center of Ship Intelligent Maintenance and Energy Efficiency Control(20DZ2252300).
