基于近红外光谱和系统聚类法的蝗虫识别模型研究

Locusts Detection Model Based on Near Infrared Spectra and Cluster Analysis

以蝗虫的近红外光谱图作为分析的对象,采用聚类分析方法对蝗虫进行快速的识别研究。结果表明:矢量归一化预处理后的光谱数据经最长距离法进行聚类分析,分类结果正确率为100%。此法可为蝗虫自动侦测系统中识别蝗虫提供一种可靠、简便的手段,盲样检测的准确率可达91.67%。

The objective of this study was to develop a rapid, real-time method for the detection of locusts. Near infrared spectra （NIR） in the range of 3996- 12489.5cm^-1 were recorded and the range of 3996 - 11502cm^-1 was selected. Models were established by using the cluster analysis method. Different methods were compared. The study showed that complete-linkage model gave better results than single-linkage and other methods with the eorrecdy classification rate of 91.67% for the test data set. Results indicate that NIR method can be used to on-line detect the locusts.