CAJ | 학술논문

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。
탐색종홍지주위해협편적고광보도상중식별수해구역적방법，내용포괄：①채집20편수해구역색소차이교명현적감귤협편적고광보서렬도상，종각협편적고광보서렬도상중선취522、647화667nm등3개여협편색소함량구유교고상관성적특정파단적고광보도상，계산667／522、667／647화647／522등3개특정파단적고광보도상적비치도상급기이치도상，식별협편중적수해구역；②계산자동식별화인공식별간적오차，검험산법적식별효과。결과표명：종감귤홍지주위해협편적RGB도상혹단일파단적고광보도상중무법직접화자동식별협편적수해구역；667／647화667／522량개특정파단반사솔비치적비치도상균능구유효지억제고광보도상중협편주위적광조성，진이환원협편적외형륜곽；종667／522특정파단반사솔비치적비치도상중식별협편수해구역적평균준학솔체92．84％，재3개비치도상중식별효과최호；식별산법능구통과계산궤편정자동실현，가작위심입연구홍지주적발생화위해규률적기술수단。
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.