Rapid detection method for edible oil adulteration based on hyperspectral imaging and transformer

Background: Hyperspectral images, as a form of imaging with a three-dimensional data structure, combine spatial and spectral information and are widely applied in fields such as remote sensing, geological exploration, and environmental monitoring. Objective: This paper aims to explore the application of hyperspectral imaging data processing and classification techniques, analyzing the potential of SVM and Transformer algorithms in hyperspectral data classification, with the goal of improving the accuracy and efficiency of hyperspectral data processing. Method: Data preprocessing methods such as SNV and MSC were applied, and the classification of hyperspectral data was studied using SVM and Transformer models to evaluate their classification accuracy and applicability. Results: Experimental results show that the combination of SVM and Transformer algorithms can effectively enhance the accuracy of hyperspectral data classification. The Transformer model performed the best in the adulteration detection of sesame oil, especially with accuracy and F1-score in the Set T sample, outperforming the SVM and KNN models. Therefore, the Transformer model demonstrates strong recognition ability in this task and is recommended as the preferred model. Innovation: This paper proposes a hyperspectral data processing method combining SVM and Transformer algorithms, further exploring its application prospects in improving classification accuracy and optimizing algorithm efficiency, providing theoretical support for the further development of hyperspectral imaging technology.