近红外高光谱技术鉴别长枣表面的农药种类

Discrimination of Different Pesticides on Long Jujubes' Surface by Near-Infrared Hyperspectral Technique

探索一种能对长枣表面农药种类进行准确鉴别的方法。利用900nm^1700nm的高光谱成像系统对喷洒不同农药的长枣进行原始光谱数据采集,利用Savitzky-Golay卷积平滑的方法对900nm^1700nm波段范围内的原始光谱进行预处理,运用偏最小二乘回归系数选择7个特征波长(SevenWavelengths,SW),采用对农药种类进行赋值的方法建立偏最小二乘(PLS)模型进行鉴别分析,结果表明,不同种类的农药在一定程度上能被有效的鉴别,鉴别正确率达到88.75%,但是在鉴别吡虫啉与哒螨灵两种农药时容易出现鉴别错误。进一步采用SW-LDA对不同农药种类进行鉴别,建立7个特征波长下的线性判别分析(LDA)模型,研究结果显示,不同农药种类的鉴别准确率达到90%以上。说明近红外高光谱联合SW-LDA能够准确地鉴别长枣表面不同的农药种类,为长枣表面农药种类的鉴别探索出了一种新方法。

In this paper, a new non-destructive method for discriminating varieties of Pesticides on Lingwu long jujubes＇surface by near-infrared hyperspectral technique was developed. Raw spectral of long jujubes which have been sprinkled by different pesticides was gathered by hyperspectral imaging system between 900 nm and 1 700 nm. Savitzky-Golay smoothing was used to deal with the raw spectral between 900 nm and 1 700 nm and 7 optimal wavelengths （SW） were selected by PLS. Different numbers was assigned to different Pesticides that PLS was established to identify. The results indicate that different Pesticides can be discriminated in a sense and discrimination accuracy is 88.75 %. However, the identification error can occur between Bichonglin and Damanling. Different Pesticides were discriminated by SW-LDA model. The correct answer rate of LDA model based on 7 wavelengths is more than 90%. The results indicate that variety discrimination of Pesticides can be achieved accurately by SW-LDA model based on near-infrared hyperspectral technique. SW-LDA model based on near-infrared hyperspectral technique is a new method for discriminating different Pesticides on long jujubes＇surface.