关键词: 玉米叶部病斑, 全卷积神经网络, 图像分割, 反卷积

Abstract: Accurate segmentation of corn leaf diseases is the key of recognizing the types of corn leaf diseases and achieving the precise application of crop medicine. In order to segment the diseased area of corn leaf accurately, this paper proposes a method for the segmentation of corn leaf diseases based on improved Fully Convolutional Neural Network（FCN）. The network structure of the proposed method mainly includes an encoding network and corresponding decoding network, and a pixel-level classifier is added behind the decoding network. The structure of the encoding network is improved on the basis of the traditional VGG-16 network. The decoding network is mainly used for the deconvolution operation of the lower sampling layer in the encoding network. By training the decoding network continuously, the resolution of the output characteristic graph of the encoding network can be recovered, and more accurate segmentation effect can be obtained. The results show that this method has good segmentation performance and can accurately segment the diseased spot area of corn leaf, compared with FCN, DeepLabV3, PSP Net and other image segmentation network models on different evaluation indexes.

Key words: corn leaf disease, fully convolutional neural network, image segmentation, deconvolution