A Non‐Destructive Detection Method for Heavy Metal‐Contaminated Shellfish Using the CARS – LDA – KNN Model

ABSTRACT Eating shellfish contaminated by heavy metals is harmful to human health, so it is imperative to detect such contamination. In this paper, a new model for rapid and nondestructive detection of heavy metal‐contaminated shellfish, based on hyperspectral image technology and a machine learning algorithm, is proposed. First, hyperspectral images of shellfish samples are collected and the Competitive Adaptive Reweighted Sampling (CARS) algorithm is used to select wavelength variables. Then, the Linear Discriminant Analysis (LDA) algorithm is used to reduce the dimension of hyperspectral image data. Finally, the K‐nearest Neighbors (KNN) classification model is applied to classify and detect heavy metal‐contaminated shellfish. For binary classification of single heavy metal‐contaminated and healthy samples, the accuracy of the CARS–LDA–KNN model for detecting heavy metal contaminated samples exceeds 99.91%. For multi‐classification of cadmium‐, copper‐, lead‐ and zinc‐contaminated and healthy samples, the accuracy reaches 100%. Moreover, the model's performance is virtually unaffected by the number of samples or the proportion division of the training set and test set, and the model exhibits strong robustness and generalization ability. For these reasons, the CARS–LDA–KNN method was established for an accurate recognition model of shellfish contaminated by heavy metals, thus it provides a new means for rapid detection of heavy metal‐contaminated shellfish.