基于YOLOv7的苹果叶片病理检测改进方法

Improved Pathologic Detection Algorithm of Apple Leaves Based on YOLOv7

我国是传统的农业大国,先进的苹果叶片病理检测方法可以有效减少林地的农药喷洒,减轻农民经济负担,同时对保护环境,减少土壤污染有着重要的意义,但是目前的苹果叶片病理虫害的很多检测模型都无法准确、及时、有效地检测到不同的虫害表面,这便阻碍了机器学习模块在苹果等林作物上的发展。在基于YOLOv7的基础上添加卷积注意力机制模块(CBAM)和轻量模型(MobileNet),建立一个新的检测模型Pest-Net。并采用CBAM注意力机制,兼顾空间和通道,多方位提高苹果叶片的缺陷检测精度,但会增加冗余的参数,需要更多的计算资源,因而结合MobileNet轻量模型,减少计算量,提高检测速度。引入SIOU损失函数提高该模型的鲁棒性。实验结果表明:与其他四种常见的目标检测模型相比,Pest-Net在mAP值(97.1%)和精确度(94.5%)上均取得了更高的准确率,有着更好的应用前景。

China is a traditional agricultural country,The advanced pathological detection method of apple leaves can effectively reduce the pesticide spraying of forest land,reduce the economic burden of farmers,and play an important role in protecting the environment and reducing soil pollution.However,many current detection models of pathological insect infestation on apple leaves are unable to accurately,timely and effectively detect different insect surfaces,which hinders the development of machine learning modules on forest crops such as apples.This paper adds convolutional attention mechanism module(CBAM)and lightweight model(MobileNet)based on YOLOv7,so a new detection model Pest-net is established,CBAM attention mechanism is adopted,taking into account space and channel,to improve the defect detection accuracy of apple leaves in all directions,but at the same time,redundant parameters will be added and more computing resources will be required.Therefore,combined with MobileNet lightweight model,reduce the amount of calculation,improve the speed of detection.Finally,SIOU loss function is introduced to improve the robustness of the model.The experimental results show that compared with other four common target detection models,Pest-net have achieved higher accuracy rates on mAP(97.1%)and precision(94.5%).Therefore,Pest-net studied in this thesis have better application prospects in the pathological detection of apple leaves.