Abstract: To explore a new method for identifying heavy metal contaminated clams, a hyperspectral spectrometer was used to collect reflectance spectra in the range of 450－900 nm of normal and heavy metal contaminated clams. The Multiple Scatter Correction(MSC) method was employed to eliminate interference factors in the spectra. Six methods, including Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Local Linear Embedding (LLE), Independent Component Analysis (ICA), Multidimensional Scaling (MDS), and Isometric Mapping (ISOMAP), were used to reduce the dimensionality of the data. Four classifiers, namely K-Nearest Neighbors (KNN), LogitBoost, Support Vector Machine (SVM), and GradientBoosting, were applied to classify 800 contaminated clams with heavy metals(cadmium (Cd), copper (Cu), lead (Pb) and zinc (Zn)) and normal clams. The results showed that all four classifiers performed well on the spectra reduced by LDA, with the LogitBoost classifier achieving an averageAccuracyof 99.40% and anF-measureof 97.99%, outperforming the other classifers. Furthermore, under imbalanced sample class sizes, classifying each type of heavy metal contaminated and normal clams separately further confirmed the robustness of the MSC-LDA-LogitBoost identification model. This study confirmed the feasibility of using hyperspectral technology combined with machine learning method to identify heavy metal contaminated clams.

Key words: 高光谱成像, 重金属污染鉴别, 蛤仔, LogitBoost分类器, 光谱降维