复杂自然环境下的棉花病叶分割与病害识别的鲁棒性方法研究

项目名称： 复杂自然环境下的棉花病叶分割与病害识别的鲁棒性方法研究

项目编号： No.31501229

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 农业科学

项目作者： 张建华

作者单位： 中国农业科学院农业信息研究所

项目金额： 20万元

中文摘要： 快速准确的病害识别是病害成功防治的关键环节与前提条件。在实际大田环境条件下，一日之内照明变化剧烈、不同背景下的反光、不同气象条件等因素，使获取的棉花病害图像具有背景复杂、颜色不均匀、噪声繁杂的特点，因此，叶片、病斑的分割与病害识别需要特别鲁棒的方法。本项目以对棉花生长危害较为严重的黄萎病、炭疽病、黑斑病、褐斑病、角斑病、盲椿象等6种病害为研究对象，从分析病叶与背景在不同光照、气象条件下的纹理、颜色特点入手，结合多种图像分割方法，研究高鲁棒性棉花病叶分割模型；探索病斑的旋转、缩放不变性特征提取方法，并研究病斑边缘的晕圈与中心区的洞孔信息作为病斑局部特征的可行性，以提取病害高可分性与不变性的特征；在稀疏表示理论基础上，分析自动学习字典大小与稀疏度的关系，探究在光照不均或部分遮挡情况下的病害识别方法，提高病害识别的鲁棒性与准确率。项目研究成果将为自然环境下的棉花病害预测预警提供技术支持。

中文关键词： 病害诊断；图像处理；复杂自然环境；鲁棒性；棉花病害

英文摘要： Fast and accurate identification of diseases is a key link of disease prevention and precondition of success. While in the actual field environment, dramatic lighting changes within one day, Reflection under different background, different weather conditions and other factors, these make the cotton disease image acquisition with complex background noise, uneven color and complex characteristics. Therefore, leaves, lesion segmentation and disease recognition methods require more robust. This project takes 6 kinds of diseases of verticillium wilt, anthracnose, black spot, brown spot, angular leaf spot, blind stinkbug as the research object. We analyze diseased leaves and background in texture, color under different light and weather conditions, and research high robust segmentation model for cotton diseased leaves with a variety of image segmentation method. Exploring rotation, scale invariant feature extraction method for disease spots, analyzing the feasibility of lesion edge halos and Holes in the central area as Local lesion characteristics, is to extract high divisibility and invariance characteristics. After analyze the relationship between automatic learning dictionary size and sparsity, explore disease identification under the light of uneven or partial occlusion based on sparse representation theory, for Improve the robustness and accuracy of disease identified. Project research results will provide technical support for cotton disease prediction and warning under natural conditions.

英文关键词： diseases diagnosis;image processing;complex natural environment;robustness;cotton diseases