基于融合小波的高光谱生菜农残梯度鉴别研究  

Study on hyperspectral lettuce pesticide residues gradient identification based onwavelet fusion

作　　者：周鑫[1] 孙俊[1,2] 武小红[1] 杨宁[1] 李青林[2] 路新男 

机构地区：[1]江苏大学电气信息工程学院,江苏镇江2120131 [2]江苏大学现代农业装备与技术教育部重点实验室,江苏镇江212013

出　　处：《中国农机化学报》2016年第8期80-86,共7页Journal of Chinese Agricultural Mechanization

基　　金：国家自然科学基金(No.31471413);江苏高校优势学科建设工程资助项目PAPD(苏政办发2011 6号);江苏省六大人才高峰资助项目(ZBZZ-019);江苏大学大学生创新创业训练计划项目(No.32);江苏大学大学生科研立项(Y14A094)

摘　　要：为有效地实现光谱信息预处理,本文将类内距离和类间距离法引入小波阈值、小波分段预处理算法中,提出WBWT和WB-PWT两种融合小波预处理算法。利用WT、PWT、WB-WT和WB-PWT预处理算法对相同的生菜农药残留高光谱数据进行预处理。通过连续投影法对预处理后光谱进行特征选取,利用支持向量机对特征选取的光谱数据进行分类鉴别。结果表明,WB-WT和WB-PWT算法较传统的WT和PWT预测准确率有了较大的提高。其中,以db4、db6、sym5函数为小波基函数和WB-WT、WB-PWT算法预处理对应的模型预测准确率分别为75.00%、84.38%、87.50%和84.38%、90.63%、93.75%,它们的预测准确率均优于WT与PWT算法分别对应的模型预测准确率57.58%、62.50%、69.70%和72.73%、87.88%、90.63%,表明融合小波预处理算法能有效地提高分类建模预测精度。In order to improve the reliability and integrity of hyperspectral information,wavelet threshold（WT）and piecewise wavelet（PWT）were used to preprocess spectroscopic data,within-class distance and between-class distance（WB）were introduced into the two preprocessing algorithm as it could define the sample,then WB-WT and WB-PWT,the fusion of wavelet preprocessing algorithms,were proposed.WT,PWT,WB-WT and WB-PWT were used to preprocess the hyperspectral data of four kinds of pesticide residues in lettuce,and the feature extraction was carried out by the successive projections algorithm and the support vector machine classification models were established.The results show that the accuracies of the models were 75.00%,84.38%,87.50%and84.38%,90.63%,93.75% respectively,by using db4,db6 and sym5as the WB-WT,WB-PWT basis functions.However,the accuracies of the models were 57.58%,62.50%,69.70% and 72.73%,87.88%,90.63%,respectively,by using db4,db6 and sym5as the WT and PWT basis functions.The fusion of wavelet preprocessing algorithm can effectively realize the pretreatment of sample sets and spectral sets,and can improve the late qualitative classification modeling accuracy.

关 键 词：预处理 融合小波 农残检测 高光谱 

分 类 号：S481.8[农业科学—农药学] O657.3[农业科学—植物保护]