Ractopamine concentration in pork prediction network based on improved one-dimensional convolutional autoencoder

Ensuring food safety necessitates the establishment of a universal and efficient method for detecting ractopamine residues in pork. However, accurate quantitative detection of complex samples using Surface-Enhanced Raman Scattering (SERS) presents challenges, particularly in mitigating interferences from sample matrix mixtures and addressing variations in hotspot distribution. This study proposes a ractopamine concentration prediction network based on an improved one-dimensional convolutional autoencoder (AE). By multi-layer convolution, pooling structure, incorporating attention mechanisms, and skipping connections, prediction accuracy is significantly enhanced. SERS-AE proposed in this paper, coupled with a custom loss function and evaluation metrics, substantially improves the model's prediction performance. SERS-AE demonstrates a higher coefficient of determination (R²), reaching 0.9792. In the comparison between predicted and actual values, SERS-AE exhibits superior performance, underscoring the potential and practical applicability of this method in ractopamine concentration detection.