基于高光谱成像的玉米收获后根茬行分割方法  

Segmentation Method for Maize Stubble Row Based on Hyperspectral Imaging

作　　者：王春雷 陈婉芝 卢彩云[1,2] 王庆杰 WANG Chunlei;CHEN Wanzhi;LU Caiyun;WANG Qingjie(College of Engineering,China Agricultural University,Beijing 100083,China;Scientific Observing and Experimental Station of Arable Land Conservation(North Hebei),Beijing 100083,China)

机构地区：[1]中国农业大学工学院,北京100083 [2]农业农村部河北北部耕地保育农业科学观测实验站,北京100083

出　　处：《农业机械学报》2020年第S02期421-426,共6页Transactions of the Chinese Society for Agricultural Machinery

基　　金：现代农业产业技术体系建设项目(CARS03);中国农业大学基本科研业务费专项资金项目(2020RC025)。

摘　　要：在华北一年两熟区,利用联合收获机留茬收获玉米后,玉米根茬行与行间秸秆及裸露地表颜色相近,采用传统的图像检测方法对其进行分割比较困难。针对该问题,采集了利用联合收获机留茬收获玉米后的根茬行高光谱图像,以根茬顶端切口为目标,提出了一种玉米根茬行高光谱图像的分割方法。首先,对黑白校正后的全波段图像进行主成分分析,根据主成分图像权重系数优选出3个特征波长,分别为1260、1658、2131 nm;然后,对3个特征波长处的图像再次进行主成分分析,并对所得到的PC2图像进行单阈值分割;最后,通过中值滤波、形态学开运算、根茬行区域外噪声滤除对分割结果进行优化。为验证该分割方法的效果,利用采集的50幅玉米根茬行高光谱图像进行试验,并选取分割准确率、召回率和F1值对分割结果进行定量评价。结果表明:该分割方法下的玉米根茬行图像分割效果较好,分割准确率、召回率和F1值分别为91.85%、90.49%和91.16%。研究结果表明基于高光谱成像技术可对玉米根茬行进行分割。Traditional vision detection methods generally have better image segmentation effect when image foreground and background have obvious chromaticity difference.However,for the maize stubble field harvested by combine harvester,there are other backgrounds besides maize stubble row such as maize residues,naked land surface,and their color are very similar.Therefore,traditional image processing methods are not suitable for the segmentation of maize stubble row.In order to achieve precise and rapid segmentation of maize stubble row,a segmentation method for maize stubble row based on hyperspectral imaging technology was put forward.Firstly,the original hyperspectral image of maize stubble row was corrected by black and white correction algorithm.Then,the principal components analysis algorithm(PCA)was used to analyze the hyperspectral image.And the feature wavelengths(1260 nm,1658 nm and 2131 nm),which could maximum highlight the stubble tip incision and lighten the backgrounds,was selected according to the weight coefficient curve.In addition,the PCA algorithm was widely used in hyperspectral image analysis because of its effective dimension reduction effect and convenience.Secondly,images at three wavelengths were extracted and analyzed by PCA.After that,the PC2 image was convert into binarization image via single threshold method.Thirdly,the median filtering algorithm,morphological open operation,and edge noise removing algorithm were applied to ensure the precision and integrity of the maize stubble row.Totally 50 test images were collected to verify the segmentation effect of the presented method.At the same time,the segmentation precision rate,recall rate,and F1 value were calculated.The results revealed that the method proposed had good segmentation effect,and the segmentation precision rate,recall rate,and F1 value were 91.85%,90.49%,and 91.16%,respectively.Therefore,the developed method realized good performance in maize stubble row segmentation and can provide great help for detection of navigation line in m

关 键 词：玉米根茬行 高光谱图像 光谱分析 主成分分析 图像分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]