Authors

Abstract

As an economic crop of Rosaceae family, strawberry has the advantages of short reproductive cycle, wide ecological adaptability and significant economic benefits, and its planting industry has been rapidly developed in recent years. Aiming at the low efficiency and high labor cost of traditional manual picking detection methods in the intelligent transformation of strawberry industry, this study innovatively proposes a strawberry fruit intelligent detection system based on YOLOV8N. By introducing RFAConv dynamic sensory field convolution, SENet channel attention mechanism and InceptionNeXt lightweight structure, combined with Wise-IoU loss function and DIoU-NMS post-processing algorithm, the synergistic enhancement of detection accuracy and computational efficiency is realized. The ablation experiments show that the improved model has a precision rate of 95.92%, a recall rate of 95.45%, and a mAP50 of 98.29% on the strawberry dataset, which are 4.14%, 3.31%, and 1.55% higher than that of the baseline model, respectively, while the number of model parameters is compressed to 5.17 M (a reduction of 12.96%). This research can provide technical support for intelligent strawberry picking.

Abstract in Chinese

草莓作为蔷薇科经济作物，具有生育周期短、生态适应性广及经济效益显著等优势，其种植产业近年来得到了快速发展。本研究针对草莓产业智能化转型中传统人工采摘检测方法存在的效率低下、人工成本高等痛点，创新性提出一种基于YOLOV8N的草莓果实智能检测系统。通过引入RFAConv动态感受野卷积、SENet通道注意力机制及InceptionNeXt轻量化结构，结合Wise-IoU损失函数与DIoU-NMS后处理算法，实现了检测精度与计算效率的协同提升。消融实验表明，改进后模型在草莓数据集上精确率达95.92%、召回率为95.45%、mAP50达98.29%，较基线模型分别提升4.14%、3.31%和1.55%，同时模型参数量压缩至5.17M（减少12.96%），该研究可为草莓智能化采摘提供技术支持。