Phase diagram combined with improved fuzzy support vector machine for rapid and nondestructive detection of Diarrhetic shellfish poisoning

Background. The diarrhoeal shellfish poisoning (DSP) toxin is a powerful marine biological toxin. Eating DSP toxin-contaminated mussels will lead to serious gastrointestinal diseases. To this end, a method for the detection of DSP toxins using near-infrared reflectance spectroscopy combined with pattern recognition is proposed. Material and methods. In the range from 950−1700 nm, spectral data of healthy mussels and DSP-contaminated mussels were obtained. To select the optimal band subsets, a band selection algorithm based on model cluster analysis was applied. As distinguishing DSP toxin-contaminated mussels from healthy mussels is a classification problem of imbalanced data, an improved fuzzy support vector machine-based recognition method was proposed. The influence of the parameters of the band selection algorithm and the fuzzy support vector machine on the model performance was analyzed. Results. Compared with the traditional support vector machine, the proposed model has better performance in detecting DSP toxins and is not affected by the imbalance ratio. Its geometric mean value can reach 0.9886 and the detection accuracy can reach 98.83%. Conclusion. The results show that as an innovative, fast and convenient analytical method, near-infrared spectroscopy is feasible for the detection of DSP toxins in mussels.