改进的YOLOv7茶叶病害识别模型

Improved Tea Disease Recognition Model for YOLOv7

摘要: 针对茶叶病害图像背景复杂、目标小、易漏检等问题，提出一种改进YOLOv7的茶叶病害识别模型。该模型首先引入混合注意力模块ACmix加强对小目标的敏感度，解决茶叶病害目标小，易漏检的问题。其次，采用C3模块替换Neck部分的ELAN-W模块以提高网络性能。最后使用Alpha-IoU损失函数优化原YOLOv7模型中的CIoU损失函数，提升模型对检测目标的定位能力。实验结果表明，改进后模型的平均检测精度mAP达到93.3%，比YOLOv7模型提高了1.8%，在FPS增加的同时模型参数量降低了3.5 M。该研究内容可以为茶园病害的智能化监控设备提供支持。

Abstract: An improved YOLOv7 model is proposed to address the challenges of complex background, small target size, and easy omission in tea disease image recognition. Firstly, the model introduces the Hybrid Attention module ACmix to enhance sensitivity towards small targets and solve the problem of easily missing tea disease targets. Secondly, the C3 module is used to replace the ELAN-W module in the Neck part to improve network performance. Lastly, the Alpha-IoU loss function is used to op-timize the original YOLOv7 model’s CIoU loss function, enhancing the model’s ability to locate detec-tion targets. Experimental results show that the improved model achieves an average precision of 93.3%, a 1.8% improvement over the YOLOv7 model, while reducing model parameters by 3.5 M, without decreasing the FPS. This research can provide support for intelligent monitoring devices for tea garden diseases.