基于改进Xception的玉米大斑病识别

利用无人机平台进行作物病害识别时，由于其拍摄图像分辨率高、目标病斑占比小，现有检测方法需要对图像进行多步骤处理，费时费力且检测效果不稳定。为减少图像处理步骤，提高检测准确率，文章以无人机拍摄图像中玉米大斑病为检测对象，首先将图像按照一定的比例进行缩小和裁剪，利用2种不同分辨率的图像重构公开数据集；然后对Xception网络进行改进，通过增加密集连接减少病斑特征信息丢失，以提高特征信息融合能力，增加注意力模块调整图像通道，以抑制无效信息；最后训练模型完成对玉米大斑病的识别并进行性能评估。实验结果表明，所提模型识别准确率达到了95.23%，单张图片识别时间减少到了0.5476 s。该模型能够有效检测无人机拍摄的图像中的玉米大斑病。

Identification of Corn Leaf Spot Disease Based on Improved Xception

When using the UAV platform to identify crop diseases, due to the high resolution of the captured images and the small proportion of the target disease spots, the existing detection methods need to process images in multiple steps, which is time-consuming, laborious and low robust in detection effect. In order to reduce the image processing steps and improve the detection accuracy, this paper takes the maize large spot disease in the image captured by the drone as the detection object. First, the image is reduced and cropped according to a certain ratio, and the public data set is reconstructed using two different resolution images. Then an improved Xception network is adopted to reduce the loss of lesion feature information and improve the ability of feature information fusion by adding dense connections, and an attention module is integrated to adjust the image channel and suppress invalid information. Finally, the training model completes the identification of corn leaf spot and performs performance evaluation. The experimental results show that the recognition accuracy rate of the network proposed in this paper reaches 95.23%, and the recognition time of a single image is reduced to 0.5476 seconds, and the proposed model can effectively identify corn lesions in images taken by drones.