Tobacco adulteration recognition study by hyperspectral data processing under machine learning

Traditional tobacco adulteration manual detection methods have problems such as complex operation and low precision, which cannot meet the increasingly refined needs of tobacco adulteration detection. In order to realize the discrimination of adulteration of different varieties of tobacco, this study combines hyperspectral imaging technology with deep learning to design a method that can quickly and accurately identify tobacco adulteration. Hyperspectral images of tobacco were collected, and the region of interest for collecting image data of tobacco samples was selected as a means of obtaining spectral data of tobacco leaves. Then the original hyperspectral data were preprocessed using the SNV method, followed by the application of PCA processing method for dimensionality reduction of the hyperspectral data. Finally, the last 5 layers of GoogLeNet were removed and 15 new layers were added, and the ReLU activation function was changed to LeakyReLU activation function in order to improve the CNN model and enhance its expressiveness. Data preprocessing, data dimensionality reduction, and hyperspectral tobacco adulteration recognition method based on improved convolutional neural network were established. The experimental study of tobacco adulteration recognition was carried out. The results show that the recognition accuracy of the improved CNN model in this paper reaches 97.56% after PCA processing. The modeling time was reduced by 30% after data dimensionality reduction by PCA method. It shows that the tobacco adulteration recognition scheme combining the PCA preprocessing method and the improved CNN model of this paper can effectively recognize tobacco adulteration. And it can also play an excellent performance in the problem of tobacco adulteration origin identification, which indicates that the combination of hyperspectral and machine learning technology can provide a powerful means for the field identification of adulterated tobacco, and has a wide range of application prospects in the field of tobacco adulteration identification.