An efficient and lightweight YOLOv8s strawberry maturity detection model

Manual strawberry picking is inefficient and costly, limiting the scalability and economic benefits of the industry. In this study, the problems of low accuracy and high complexity in the existing ripeness detection models are solved by introducing the improved YOLOv8s model, which meets the real-time and precision requirements. The main improvements of this model include replacing the necked Bottleneck structure with the FasterNet structure, incorporating an efficient multi-scale attention (EMA) mechanism, and utilising the Ghost module to reduce the computational load without sacrificing performance. In addition, Wise-IoU (WIoU) improves the accuracy of bounding box regression compared to traditional CIoU. The model performs well in complex environments with 93.8% mAP, 36.8% parameter reduction, 34.6% reduction in complexity (GFLOPs), and 12.7% improvement in FPS, along with a 37.7% reduction in model size. Overall, the YOLOv8s-FEGW model significantly enhances strawberry detection and supports automation and intelligence in agriculture.