基于改进YOLOv5s对苹果叶片病害的检测识别

Detection and Identification of Apple Leaf Disease Based on the Improved YOLOv5s

摘要: 苹果是一种备受欢迎且营养丰富的水果，市场需求也在持续增长。然而，苹果叶片病害严重威胁着苹果的产量和质量。准确、快速地识别病害对苹果产业的发展尤为重要，因此，本文利用YOLOv5s模型构建了一个改进的病害检测识别模型，通过引入SE注意力机制、Res2Net和解耦检测头结构，改进了模型的注意力、感受野和检测效果。探索了在苹果叶片病害检测中提高准确性和性能的方法。构建出YOLOv5s-SRD模型，准确率、召回率、平均检测精度分别达到84.8%、83.3%、86.0%，相比原模型分别提高了1.3%、1.5%、1.8%，提升效果显著，模型检测效果较好。有助于苹果叶片病害的检测识别，从而提高苹果产量。

Abstract: Apples are a popular and nutritious fruit, and the market demand continues to grow. However, apple leaf disease seriously threatens the yield and quality of apples. Accurate and rapid identification of diseases is particularly important for the development of apple industry. Therefore, this paper uses YOLOv5s model to build an improved disease detection and identification model. By introducing SE attention mechanism, Res2Net and decoupling detection head structure, the attention, receptive field and detection effect of the model are improved. Methods to improve accuracy and performance in apple leaf disease detection were explored. The YOLOv5s-SRD model was constructed, and the accuracy, recall rate and average detection accuracy reached 84.8%, 83.3% and 86.0% respectively, which were 1.3%, 1.5% and 1.8% higher than the original model respectively, with the significant improvement effect and the model detection effect was good. It is helpful for the detection and identification of apple leaf disease, so as to increase the apple yield.