基于高光谱成像技术应用光谱及纹理特征识别柑橘黄龙病（英文）

发布时间：2017-12-27 17:28

本文关键词：基于高光谱成像技术应用光谱及纹理特征识别柑橘黄龙病（英文）　出处：《光谱学与光谱分析》2016年07期 　论文类型：期刊论文

【摘要】：讨论了基于高光谱成像技术光谱及纹理特征在识别早期柑橘黄龙病中的应用。使用一套近地高光谱成像系统采集了176枚柑橘叶片的高光谱图像作为实验样品,其中健康叶片60枚,黄龙病叶片60枚,缺锌叶片56枚。手工选取每幅叶片高光谱图像的病斑位置作为样品感兴趣区域(regions of interest,ROI),计算其平均光谱反射率,并以此作为样品的反射光谱,光谱范围为396~1 010nm。样品光谱分别经过主成分分析(PXA)及连续投影算法(SPA)进行数据降维,再结合最小二乘支持向量机(LS-SVM)分类器建立分类模型。相比原始光谱,由PCA选取的前四个主成分及SPA选取的一组最佳波长组合(630.4,679.4,749.4和899.9 nm)建立的模型拥有更好的分类识别能力,其对三类柑橘叶片平均预测准确率分别为89.7%和87.4%。同时,从被选四个波长的每幅灰度图像中提取6个灰度直方图的纹理特征以及9个灰度共生矩阵的纹理特征再次构建分类模型。经SPA优选的10个纹理特征值进一步提高了分类效果,对三类柑橘叶片的识别正确率达到了100%,93.3%和92.9%。实验结果表明,同时包含光谱信息及空间纹理信息的高光谱图像在柑橘黄龙病的识别中显示了很大的潜力。
[Abstract]:The application of spectral and texture features based on hyperspectral imaging technology in the identification of early orange yellow dragon disease was discussed. A collection of 176 hyperspectral images of citrus leaves was collected by using a near earth hyperspectral imaging system as experimental samples, including 60 healthy leaves, 60 Huanglong disease leaves and 56 zinc deficient leaves. Manual selection of each leaf hyperspectral image spots position as the region of interest in the sample (regions of interest, ROI), calculate the average spectral reflectance, and as the reflectance spectra of the samples, the spectral range of 396~1 010nm. Sample spectra were dimensionality reduced by principal component analysis (PXA) and continuous projection algorithm (SPA) respectively, and then the classification model was established by combining least squares support vector machine (LS-SVM) classifier. Compared with the original spectrum, the first four principal components selected by PCA and the best wavelength combination selected by SPA (630.4679.4749.4 and 899.9 nm) have better classification and recognition ability, and the average prediction accuracy of three kinds of citrus leaves is 89.7% and 87.4%, respectively. At the same time, from the selected four wavelengths of each gray image, extract 6 grayscale histogram texture features and 9 gray level co-occurrence matrix texture features, and build the classification model again. The classification results were further improved by 10 texture features selected by SPA, and the recognition accuracy of three types of citrus leaves was 100%, 93.3% and 92.9%. The experimental results show that high spectral images, including spectral information and spatial texture information, have shown great potential in the identification of Citrus Yellow Dragon disease.
【作者单位】：中国农业大学现代精细农业系统集成研究教育部重点实验室;河南科技大学农业工程学院;Department
【基金】：The Citrus Research and Development Council,USA,National Natural Science Foundation for Young Scholars of China(31301240)
【分类号】：S436.66;O657.3
【正文快照】：

【相似文献】