番茄植株在生长过程中受病虫害的侵染, 将导致番茄减产和种植户的经济效益降低, 该研究用高光谱技术结合化学计量学方法, 实现了番茄叶片斑潜蝇虫害的快速识别。 搭建了简易的高光谱成像系统, 包括光源单元、 高光谱图像采集单元和数据处理单元, 用该系统获取番茄叶片的高光谱图像, 对高光谱图像进行校准, 并从每一幅图像中提取光谱信息。 分别采用了光谱角匹配(SAM)分析方法和光谱红边参数判别分析(DA)方法识别番茄叶片斑潜蝇虫害。 在SAM分析中, 对高光谱数据进行了归一化预处理, 以消除多余信息, 增加样品之间的差异。 比较了以不同番茄叶片样品的反射光谱作为测试光谱时, 虫害识别效果的差异, 当以受到斑潜蝇侵染的番茄叶片的平均反射光谱作为测试光谱时, 虫害识别的正确率较高, 达到96.5%。 在光谱红边参数判别分析中, 从光谱数据中提取了红边位置、 红边振幅、 最小振幅、 红边面积、 红谷位置和红边振幅/最小振幅6组红边信息, 利用判别分析方法建立番茄叶片斑潜蝇虫害的判别模型, 比较了距离判别、 Fisher判别、 Bayes判别分析方法的判别效果, 使用距离判别分析建模的判别正确率最低, 判别正确率为88.0%, 使用Fisher判别分析建模的效果最佳, 判别正确率为96.0%。 研究结果表明, 采用高光谱技术识别番茄叶片斑潜蝇虫害具有可行性。

Tomato yield and farmers’ economic benefits will decrease when insect pest occurs in the growth of tomato plants. This study used hyperspectral technology combined with chemometrics methods to realize fast identification of tomato leaf LiriomyzaSativae Blanchard pest. A simple hyperspectral imaging system was developed, including a light source unit, and hyperspectral image acquisition unit and a data processing unit, and hyperspectral images of tomato leaves were collected through this system. Hyperspectral images were calibrated and spectral information was extracted from each image. Spectral angle mapping (SAM) analysis method and spectrum red edge parameters discriminant analysis (DA) method were adopted to identify tomato leaf Liriomyza Sativae Blanchard pest respectively. In the SAM analysis, normalization algorithm was utilized to preprocess hyperspectral data so as to eliminate redundant information in hyperspectral data and increase the differences between samples. Discriminant effects of tomato leaf pest were compared when different reflective spectrums of tomato leaf samples were used as test spectrums. It was found that when regarding the average reflectance spectrum of 100 tomato leaves infected by Liriomyzasativae Blanchard pest as the test spectrum, the overall recognition accuracy was higher, reaching to 96.5%. In spectrum red edge parameters discriminant analysis, 6 kinds of red edge information that red edge position, red edge amplitude, minimum amplitude, red edge area, location of minimum chlorophyll absorption, and the ratio of red edge amplitude to minimum amplitude were extracted from tomato leaves’ spectral data. Discriminant analysis method was used to develop discriminant model of tomato leaf LiriomyzaSativae Blanchard pest, discriminant effects of distance discriminant analysis, Fisher discriminant analysis, and Bayes discriminant analysis were compared. Comparison results indicated that Fisher discriminant analysis generated the best discriminant effect. The discriminant accuracy was 96.0% for validation set, while distance discriminant analysis produced the worst discriminant effect, with 88.0% discriminant accuracy. Research results showed that using hyperspectral technology to identify Liriomyza sativae Blanchard pest was feasible.