Pesticide Residue Detection Using Grating Spectroscope and GWO–CNN–BiLSTM Method

Abstract

Pesticide residues represent a globally significant issue due to their high toxicity and broad dissemination. Thiamethoxam (TMX), commonly applied during the cultivation of vegetables like spinach, has its principal metabolite, clothianidin, which can accumulate in crops, posing significant long-term dietary risks. To accurately detect TMX residues in spinach, this study developed a portable detection device integrating a grating spectrometer (GS) with a smartphone and introduced image processing techniques alongside deep learning detection methods. This method employs a ResNet50 model enhanced by Squeeze-and-Excitation Networks (SE) attention mechanism to extract key features, which are subsequently input into a hybrid model combining a convolutional neural network (CNN) and a bidirectional long short-term memory (BiLSTM) network, optimized using the gray wolf optimization (GWO) algorithm. The results demonstrate that the method achieves a root mean square error (RMSE) of approximately 0.220, a mean absolute error (MAE) of about 0.060, a mean bias error (MBE) of about 0.002, and a coefficient of determination (R²) of approximately 0.960. The R² increased by 0.049 compared to pre-optimization values and by 0.060 relative to the top traditional machine learning models, thereby enhancing the precision of detection. This technology promises to be a vital tool in the field of pesticide residue detection, offering robust support for ensuring food safety and public health.