Hyperspectral Data Analysis Algorithm Based on Partial Least-squares Regression Model and Its Application in the Identification of Hogwash Oil

摘要

In this paper a hyperspectral based hogwash oil identification method is proposed. The proposed method using near-infrared hyperspectral imaging technology for edible soybean oil adulteration classification among the selected mixed oil samples with different proportions of soybean oil and hogwash oil. First, we capture the mixed oil samples hyperspectral images by using hyperspectral camera whose wavelength range is between 900-1600 nm. Next, some features of the obtained hyperspectral image data are extracted for identify different mixed oil samples. Then we classify the pure soybean oil and adulterated oil by using the Partial Least Squares (PLS) algorithm, Support Vector Machines (SVM) and Neural Network (NN) classification models. Finally, by using Principal Component Analysis (PCA), the important bands of samples are extracted for computing efficiency. Experimental results show that in the full-band space, partial least squares average accuracy rate reached 89.75%, which is higher than the support vector machines and neural networks. While the performance of space partial least squares declined slightly with the only important band, but still achieved good results.