| 摘 要: 草莓作为高营养价值的经济型浆果，病害严重影响其产量与果实品质。针对传统病害识别方法难以适应规模化种植中对病害快速精准防控等问题，本文提出一种基于改进YOLOv5s的草莓病害轻量化识别算法。该算法以YOLOv5s目标检测模型为基础，首先采用Mosaic增强、HSV色彩调整及Cutout等数据扩增方法，提升模型泛化能力；接着利用K-means聚类算法优化先验框尺寸，增强检测针对性；再引入Ghost卷积模块，再保证模型检测精度基本不变的前提下，有效降低了模型的计算量和参数量，提升检测速度；最后模型在PyTorch框架下进行训练与调优。实验结果显示，系统在测试集上的平均精度均值（mAP@0.5）达到99%，精确率与召回率分别为95.4%和98.5%，F1分数为97%。在多场景验证中，系统在近距离无遮挡条件下识别准确率为100%，在强光、光照不均及局部遮挡等复杂环境下仍保持92%以上的识别率，表现出良好的鲁棒性与实用价值。实验结果表明，该算法能实现早期诊断与针对性防治，减少农药滥用，为草莓种植的智能化与可持续发展提供技术依据。 |
| 关键词: 草莓病害识别 YOLOv5s算法 目标检测 深度学习 数据增强 轻量化模型 |
|
中图分类号:
文献标识码:
|
|
| Lightweight Recognition Algorithm for Strawberry Diseases Based on Improved YOLOv5s Model |
|
Xiezeqi
|
Zhengzhou SIAS University
|
| Abstract: As an economical berry with high nutritional value, strawberry diseases seriously affect its yield and fruit quality. Aiming at the problem that the traditional disease identification method is difficult to adapt to the rapid and accurate prevention and control of diseases in large-scale planting, this paper proposed a lightweight identification algorithm for strawberry diseases based on improved YOLOv5s. Firstly, data amplification methods such as Mosaic enhancement, HSV color adjustment and Cutout were used to improve the generalization ability of the model. Secondly,K-means clustering algorithm was used to optimize the size of prior boxes, so as to enhance the detection pertinence. Thirdly,the Ghost convolution module was introduced to effectively reduce the computational cost and parameters of the model while basically maintaining its detection accuracy, thereby improving the detection speed.Finally, the model was trained and optimized under the PyTorch framework. The experimental results showed that the average accuracy mean ( mAP@0.5 ) of the system on the test set reaches 99 %, the precision and recall rates were 95.4% and 98.5%, and the F1 score was 97%. In the multi-scene verification, the recognition accuracy of the system was 100% under the condition of no occlusion at close range, and it still maintains more than 92% recognition rate in complex environments such as strong light, uneven illumination and partial occlusion, showing good robustness and practical value. The experimental results showed that the algorithm could realize early diagnosis and targeted prevention and control, reduce pesticide abuse, and provide technical basis for the intelligent and sustainable development of strawberry planting. |
| Keywords: Strawberry Disease Identification YOLOv5s Algorithm Object Detection Deep Learning Data Augmentation Lightweight Model |
