高光谱成像技术预测农用乙烯–乙酸乙烯酯塑料棚膜撕裂强度

Application of Hyperspectral Imaging Technology on Predicting Tearing Strength of Agricultural Ethylene-Vinyl Acetate Plastic Greenhouse Films

结合近红外高光谱成像技术和化学计量学方法快速预测农用乙烯–乙酸乙烯酯塑料(EVAC)棚膜的撕裂强度保持率。采用高光谱仪获取EVAC棚膜在988.44～1 679.68nm波长范围内的近红外高光谱曲线原始特征数据,利用冲击试验机获取棚膜样本的撕裂强度保持率,采用Savitzky-Golay、变量正态化法、一阶导数法(1st-Deriv)和多变量漂移矫正法对采集到的高光谱特征数据进行预处理,然后采用随机蛙跃和连续交替投影算法从中选择特征波长作为变量,最后通过支持向量机回归算法建立反映棚膜样本内部结构特性的特征光谱曲线与撕裂强度保持率相关性模型并通过该模型对棚膜的撕裂强度保持率进行了预测和验证。实验结果显示,基于1st-Deriv预处理的连续交替投影–支持向量机回归模型获得最优的预测效果,对应的建模集和预测集的相关系数平方百分数(r2)分别为93.57%和88.61%,比仅用支持向量机回归模型的r2分别提高了1.57%和1.24%,相应的均方根误差分别下降了0.116和0.134。该模型的建立为快速且准确地监测农用EVAC棚膜品质提供了重要的技术手段。

The combination of near infrared hyperspectral imaging technology and chemometrics algorithms was employed to predict the tearing strength retention rate of agricultural ethylene-vinyl acetate plastic(EVAC)greenhouse films quickly.The hyperspectral imaging device was used to gather the raw characteristic data from near infrared hyperspectral curves of EVAC films within the scope of wave length between 988.44 nm and 1 679.68 nm.The tearing strength retention rates of EVAC films were evaluated by using impact testing machine.The raw characteristic data was preprocessed by using the Savitzky-Golay,standard normal variate,first derivative(1st-Deriv)and multiplicative scatter correction algorithms.The algorithms of random frog and successive projections algorithm(SPA)were used to select the sensitive bands from the preprocessed data as variable.A correlation model between the characteristic spectroscopic curves which can reflect the inner structure characteristic of the EVAC films and the tearing strength retention rates was built up by using the support vector machine(SVM)regression algorithms,and the prediction and verification for the tearing strength retention rate of the EVAC films were done through the model.The results show that,the SPA–SVM model based on 1st-Deriv preprocessing generates the best prediction results,the square percentage of correlation coefficient(r2)of corresponding modeling set and prediction set are 93.57%and 88.61%respectively,which increase by 1.57%and 1.24%compared with r2 of simple SVM model and the root mean square error reduce by 0.116 and 0.134.The building of the model can provide an important means for monitoring the quality of agricultural EVAC greenhouse films quickly and precisely.