从红蜘蛛为害叶片的高光谱图像中识别受害区域  

In-leaf Affected Area Identification from Hyper-spectral Image of Citrus Red Mite Infected Leaf

作　　者：王建[1,2] 李震[1,2,3] 洪添胜[1,2,3] 倪慧娜[2] 邓小玲[1,2,3] 郑建宝[2] 

机构地区：[1]南方农业机械与装备关键技术教育部重点实验室,广州510642 [2]华南农业大学工程学院,广州510642 [3]国家柑橘产业技术体系机械研究室,广州510642

出　　处：《农机化研究》2015年第7期18-22,共5页Journal of Agricultural Mechanization Research

基　　金：国家自然科学基金项目(31101077);现代农业产业技术体系建设专项(CARS-27);广东省科技计划项目(2011B020308009)

摘　　要：探索从红蜘蛛为害叶片的高光谱图像中识别受害区域的方法,内容包括:1采集20片受害区域色素差异较明显的柑橘叶片的高光谱序列图像,从各叶片的高光谱序列图像中选取522、647和667nm等3个与叶片色素含量具有较高相关性的特征波段的高光谱图像,计算667/522、667/647和647/522等3个特征波段的高光谱图像的比值图像及其二值图像,识别叶片中的受害区域;2计算自动识别和人工识别间的误差,检验算法的识别效果。结果表明:从柑橘红蜘蛛为害叶片的RGB图像或单一波段的高光谱图像中无法直接和自动识别叶片的受害区域;667/647和667/522两个特征波段反射率比值的比值图像均能够有效地抑制高光谱图像中叶片周围的光噪声,进而还原叶片的外形轮廓;从667/522特征波段反射率比值的比值图像中识别叶片受害区域的平均准确率达92.84%,在3个比值图像中识别效果最好;识别算法能够通过计算机编程自动实现,可作为深入研究红蜘蛛的发生和为害规律的技术手段。The paper explores the methods of recognizing the citrus red mite infected leaf area based on the hyper-spec-tral image technology, the main content is as follows：1） collect 20 hyper-spectral citrus leaves image with large different pigment concentration firstly, then three bands namely 522, 647 and 667nm as the characteristic bands are chosen which are highly related to the amount of citrus leaf pigment concentration to detect the situation of citrus red mite infected leaf, and the ratio images of characteristic bands of 647/522, 667/647 and 667/522 and their black and white images are calculated and the affected leaf areas are dejected.2） The test error is obtained by means of automatic computer recogni-tion and man-made recognition perspectively to test the detection algorithm.The test shows that the monochrome images transformed from the two ratio images at the band ratios of 667/647 and 667/522 can restore the shape of the leaf finely. Within the two ratio images, the 667/522 image’ s monochrome image of which the highlighted areas match highly with the damage leaf areas.Hereby, this reflectance ratio of the characteristic bands can effectively restrain the noise intro-duced by taking the hyper-spectral image and it can be used as an technique means for studying the citrus red mite infec-tion happening and regularity.

关 键 词：柑橘 红蜘蛛 高光谱图像 特征波段 机器视觉 

分 类 号：S123[农业科学—农业基础科学] TP391.41[自动化与计算机技术—计算机应用技术]