摘要
为正确分割田间籽棉图像,将棉花与背景视为两个类别,在典型的未成熟籽棉图像和不同质量等级的成熟/过熟籽棉图像中,用肉眼选取20000个白棉、黄染棉和污染棉等棉花像素以及20000个棉株、土壤等背景像素,在RGB、HSI、La*b*和Hunter颜色空间下获取两类像素之间的颜色阈值,基于阈值进行图像分割,选取噪声较少的HSI和La*b*颜色空间,进一步基于形态学滤波器去噪,实验结果表明,907幅籽棉图像分割的准确率为87.21%和86.33%。HSI颜色空间更适合分割成熟籽棉图像,La*b*颜色空间则适合未成熟籽棉;颜色阈值覆盖范围广,基于速度的阈值分割法能够适应田间籽棉环境。
The goal of cotton production in China is to improve corresponding rate of cotton quality grade, foreign fibers, adulteration, and cotton baling inconsistent phenomenon to decrease continuously. With the background, machine vision and pattern recognition technologies are introduced into traditional picking task to discriminate maturity degree and grade of quality of field cotton, which will solve the problem of picking cotton by the way from source, so that various cotton varieties can be adapted, pollution caused by agriculture chemicals can be avoided, labor cost can be reduced and agriculture cost can be decreased. In order to segment field cotton images exactly, we regarded cotton and its background as two classes and segmented them based on their color threshold. A total of 20 000 white, yellow, and stain cotton pixels and 20 000 background pixels of soil and cotton plant, including cotton bracteole, leaf, and branch, were extracted from typical under-ripe cotton images and ripe/over-ripe cotton images with various quality grades from 1 to 7. Color threshold of two classes of cotton and its background pixels were obtained in RGB, HSI, La＊b＊, and Hunter color space respectively; on the basis of which cotton regions were segmented from images; and HSI and La＊b＊ color spaces were selected respectively by using S below 28, I over 108, L over 118, a＊ from 123 to 134, b＊ below 136 with less segmentation noise which would be removed based on morphological filter. The experiment results showed that 907 cotton images were segmented with an accuracy of 87.21% and 86.33% in HSI and La＊b＊ color space respectively. The front images were segmented with an accuracy of 90.83% and 89.98% and the side images with an accuracy of 83.33% and 82.42%. Ripe cotton images were segmented perfectly in HSI color space while under-ripe cotton images in La＊b＊ color space, and the speed-based segmentation method with threshold covering a wide area was preferable for field cotton surroundings.
出处
《作物学报》
CAS
CSCD
北大核心
2010年第3期502-507,共6页
Acta Agronomica Sinica
基金
国家高技术研究发展计划(863计划)项目(2006AA10Z259)资助
关键词
田间棉花
图像分割
颜色阈值
去噪
Field cotton
Image segmentation
Color threshold
Removing noise
作者简介
第一作者联系方式：E-mail：lingw@njau．edu．cn；Tel：13913306944
通讯作者（Corresponding author）：姬长英，E-mail：chyji@njau．edu．cn；Tel：13951994628
