基于深度学习的玉米叶片病害识别方法研究

Research on corn leaf disease recognition method based on deep learning

玉米叶片病害的及时识别和防治对保障玉米产量和质量具有重要意义。提出一种基于深度学习的玉米叶片病害识别方法,并开发了相应的移动应用。使用YOLOv5s、YOLOv5m、YOLOv5l、ResNet18、ResNet34、ResNet50、EfficientNet_b0、EfficientNet_b1、EfficientNet_b2共9个深度学习模型,对包含普通锈病、枯萎病、灰色叶斑和健康叶片四类的玉米叶片病害图像进行训练和识别,比较不同模型在识别准确率、识别速度和模型大小方面的性能,结果表明EfficientNet_b0模型具有最高的平衡准确率(96.3%)和较低的模型复杂度。基于EfficientNet_b0模型开发了安卓平台的移动应用,该应用能够实现对玉米叶片病害的实时识别或加载相册图片识别,并给出相应的类别名称和置信度、病害介绍以及建议防治措施,为玉米种植者提供了一种便捷、高效和智能的玉米叶片病害诊断工具,有助于提高玉米生产的效率和质量。

Timely identification and control of maize leaf diseases is of great significance to ensure maize yield and quality.A method of maize leaf disease recognition based on deep learning was proposed and the corresponding mobile application was developed.Nine deep learning models were used:YOLOv5s,YOLOv5m,YOLOv5l,ResNet18,ResNet34,ResNet50,EfficientNet_b0,EfficientNet_b1,and EfficientNet_b2.The results showed that the EfficientNet_b0 model had the highest balance accuracy(96.3%)and low model complexity,and the performance of recognition accuracy,recognition speed and model size were compared among the four types of maize leaf disease images,including common rust,blight,gray leaf spot and healthy leaf.The mobile application on Android platform was developed based on the EfficientNet_b0 model,which could realize realtime identification of maize leaf diseases or recognition of loaded photo albums,and provide corresponding category names and confidence levels,disease introduction and suggested prevention and control measures,providing a convenient,efficient and intelligent maize leaf disease diagnosis tool for maize growers.It is helpful to improve the efficiency and quality of corn production.