【摘要】 在马铃薯的生长发育过程中，马铃薯会遭受到各种各样病害的侵染，这些病害会严重影响马铃薯的产量和质量。针对人工识别植物病害方法效率低下、识别精度低和处理范围有限的问题，提出了一种基于YOLOv8的马铃薯病害识别模型。本文以背景复杂的马铃薯叶片图像为研究对象，在原YOLOv8的基础上，在YOLOv8的Backbone层和Neck层中加入了CBAM注意力模块和更适应图像边缘变化的DCN可变形卷积，以提高模型对病害叶片的识别能力。实验结果表明，优化后的YOLOv8模型的全类平均准确率mAP@0.5、mAP[0.5:0.95]和准确率P比原YOLOv8模型分别提高了4.32%、5.48%、7.72%。实验证明，优化后的模型比原模型有更好的性能表现。更多还原

【Abstract】 During the growth and development of potatoes, they are susceptible to various diseases which seriously affect the yield and quality. A potato disease recognition model based on the YOLOv8 was proposed to address the issues of low efficiency, low recognition accuracy, and limited processing range in the manual identification process of plant diseases. The potato leaf images with complex backgrounds were taken as the research object, based on the original YOLOv8, CBAM attention modules and DCN deformable convolutions which are more adaptable to image edge changes were added to the Backbone and Neck layers of YOLOv8 to improve the model’s recognition ability for diseased leaves. The experimental results indicate that the average accuracy of all classes mAP@0. 5, the mAP [0. 5:0. 95], and the P accuracy of the optimized YOLOv8 model were improved by 4. 32%, 5. 48%, and 7. 72% respectively compared to those of the original YOLOv8 model. The experimental results demonstrated that the optimized model had better performance than the original model.更多还原