高光谱图像技术检测梨表面农药残留试验研究

Detecting Pesticide Residue on Crystal Crown Pear Surface by Hyperspectral Imaging Technology Combined with Artificial Neural Network

以滴有不同浓度毒死蜱和炔螨特农药的水晶皇冠梨为研究对象,探讨高光谱图像技术结合人工神经网络方法检测水果表面农药残留量的可行性.分别配制不同浓度分布的毒死蜱水溶液和炔螨特水溶液样本各20个,按100μL和150μL取农药溶液滴在梨表面,在835.467 8～1 648.356 8 nm范围采集高光谱图像,提取感兴趣区域数据得到4组样本数据,每组20个,每组随机抽取5个样本作为预测集,利用BP人工神经网络对每组数据分别建立数学模型.滴有150μL和100μL农药溶液区域的建模结果为:两种农药的残留样本相关系数分别都大于0.99和0.95;RMSEC和RMSEP的最大值分别为0.634 9,1.323 9和1.742 5,3.441 7.结果表明:150μL农药样本区优于100μL农药样本区的建模结果,高光谱图像技术结合人工神经网络法检测梨表面农药残留量是可行的,为水果表面农药残留量检测提供了新方法.

The hyperspectral imaging technology combined with BP neural network was used to detect pesticide residues on surface of fruit,crystal crown pears,such as chlorpyrifos and propargite pesticide.Twenty samples were prepared for both chlorpyrifos and propargite with different concentrations.Dropping the 100 μL and 150 μL pesticide solution droplets in pear surface,the hyperspectral images were scanned in 835.467 8~1 648.356 8 nm band range.Twenty regions of interest（ROI） were available for four sample groups,from which five samples were randomly selected as predictor set.The results of models built by BP neural network for 100 μL and 150 μL pesticide solution region showed that correlation coefficient of the samples was greater than 0.99 and 0.95 respectively.RMSEC and RMSEP maximum value was respectively 0.634 9,1.323 9 and 1.742 5,3.441 7.The results indicated that 150 μL pesticide solution region was better than 100 μL in the modeling.Hyperspectral image technology combined with BP neural network method is feasible for detection of pesticide residues on pear surface.which provide a new method for detection of pesticide residues on fruit surface.