Abstract:  Fusarium head blight is one of the main diseases affecting wheat yield and quality. In order to monitor the occurrence of fusarium head blight in wheat rapidly and effectively, RGB images were acquired by using a digital camera in wheat fields with artificially inoculated Fusarium graminearum in this study, and Deeplabv3+ network model parameters were adjusted and trained on the basis of image preprocessing. Using the lightweight network MobileNet V2 as the network coding module, an identification, and detection model of the wheat ear caused by fusarium head blight was established using the hollow convolution technology based on a deep learning network, and the model was verified and evaluated using the measured data. The results showed that the average accuracy of the model was 0969 2, the Loss value of loss function was 0.103 0, and the mean intersection over union (MIoU) was 0793. The model has a good recognition and detection effect. These results provide a new method for the detection and identification of wheat fusarium head blight.

Key words: wheat, Fusariumhead blight, deeplabv3+ model, deep learning, image recognition

摘要： 赤霉病是影响小麦产量和品质的主要病害之一。为快速、有效地监测小麦赤霉病的发生情况,利用数码相机对人工接种赤霉病菌的小麦田进行RGB图像获取,在图像预处理基础上,对Deeplabv3+网络模型进行调参和训练。以轻量化网络MobileNet V2为网络编码模块,利用空洞卷积技术建立基于深度学习网络的小麦赤霉病发病麦穗的识别与检测模型,并用实测数据对模型进行验证和评价。结果表明,该模型的平均精度为0.969 2,损失函数Loss为0.103 0,平均交并比MIoU为0.793,模型识别与检测效果较好。上述结果为小麦赤霉病的检测与识别提供新的手段。

关键词: 小麦, 赤霉病, Deeplabv3+模型, 深度学习, 图像识别