Similarity Measurement Method of Near Infrared Spectrum Based on Adaptive Geodesic Local Tangent Space Alignment Algorithm

Traditional approaches for recipe development, raw material understanding, and quality assessment face significant challenges. Experts in recipe design find it difficult to rely solely on sensory judgments such as visual observation, tactile perception, and olfactory assessment to discriminate raw material quality. Therefore, it becomes crucial to study similarity measurement methods for raw materials and assist experts in quickly identifying suitable ingredients. Commonly used methods based on distance, cosine angle, or correlation are not well-suited for judging similarity among high-dimensional and nonlinear data. Hence, this paper proposes a similarity measurement method of near-infrared spectrum based on adaptive geodesic local tangent space alignment algorithm (AGLTSA). By utilizing geodesic distance, this algorithm can better measure the similarity between near-infrared spectral data. Additionally, the use of an adaptive neighborhood selection strategy improves the fitting of the local space for high-dimensional data, thus enhancing the effectiveness of similarity measurement for near-infrared spectral data. This study designs experiments to validate the effectiveness of the proposed algorithm in dimensionality reduction, similarity detection, and classification tasks related to real-world raw materials such as beer malt in industrial production. The experimental results demonstrate that this algorithm are superior to those obtained by PCA, ISOMAP, LLE, LTSA, and UMAP algorithms. It can effectively handle the classification and substitution tasks of recipe ingredients in industrial production, thus assisting in maintaining product quality in enterprise.