Identification of Thawed Fish Meat Based on Near-Infrared Spectroscopy Combined with DAG-LSTMSVM Model

Background.Rapid and accurate identification of frozen-thawed fish remains a challenge in food quality assessment.This study proposes a qualitative detection model based on near-infrared spectroscopy combined with a Directed Acyclic Graph Least Squares Twin Mahalanobis Distance-based Support Vector Machine (DAG-LSTMSVM).The model is designed to effectively classify fish meat subjected to different thawing methods. Material and methods.A near-infrared reflectance spectroscopy system was employed to collect spectral data from fresh fish meat, ultrasonic-thawed fish meat, and microwave-thawed fish meat across the wavelength range of 950-1700 nm.Various pre-processing methods were applied to the spectral data to evaluate their influence on model performance. Results.The DAG-LSTMSVM model combined with Savitzky-Golay (SG) smoothing pre-processing achieved a detection accuracy of 97.8%, outperforming conventional support vector machine models.Notably, this high accuracy was maintained even when the sample size was substantially reduced, indicating strong robustness.Conclusion.The integration of near-infrared spectroscopy with the DAG-LSTMSVM model provides a rapid and reliable approach for identifying thawed fish meat, offering a promising method for fish quality evaluation.