A multilevel cooperative attention network of precise quantitative analysis for predicting ractopamine concentration via adaptive weighted feature selection and multichannel feature fusion

Authors: YIN, TIANZHEN - China Agricultural University; PENG, YANKUN - China Agricultural University; LI, YONGYN - China Agricultural University; Chao, Kuanglin; NIE, SEN - China Agricultural University; ZUO, JIEWEN - China Agricultural University

Submitted to: Food Chemistry
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/6/2025
Publication Date: 2/8/2025
Citation: Yin, T., Peng, Y., Li, Y., Chao, K., Nie, S., Zuo, J. 2025. A multilevel cooperative attention network of precise quantitative analysis for predicting ractopamine concentration via adaptive weighted feature selection and multichannel feature fusion. Food Chemistry. earticle: 464. https://doi.org/10.1016/j.foodchem.2024.141884.
DOI: https://doi.org/10.1016/j.foodchem.2024.141884

Interpretive Summary: Ractopamine is a veterinary drug which promotes leaner meat in swine and cattle. The presence of veterinary drug residues in meat and meat products may cause health risks to consumers. In this study, we have developed two novel machine learning models for precise detection of ractopamine residues in pork samples. Compared to traditional statistical methods, the machine learning techniques demonstrated high accuracies for prediction of ractopamine concentrations using spectroscopy. The machine learning techniques proposed in this study effectively address challenges such as nonlinear relationships, noise, and spectral interference in spectroscopic data analysis. The results of this study will serve as an important reference which will benefit researchers who have interest in developing machine learning methods for accurate quantitative analysis other possible veterinary drug residues in meat products.

Technical Abstract: Surface enhanced Raman spectroscopy (SERS) holds great potential for detecting contaminant residues such as veterinary pharmaceuticals and feed additives in foods due to its capability for rapid detection and high sensitivity. However, issues such as signal noise, fluctuations, and spectral shifts can negatively impact its performance in detecting ractopamine in pork. Hierarchical Gradient Aware Spectral Network (HGASNet) was proposed to address these issues. The key innovations implemented are the Spectral Gradient Weighted Attention (SGWA) and Multi-Channel Peak Attention Mechanism (MCPAM) modules within HGASNet. The SGWA module dynamically adjusts feature weights, enhancing sensitivity to critical Raman spectral shifts. Meanwhile, MCPAM leverages multi-channel attention to better capture long range dependencies and fuse global and local information. Additionally, HGASNet's hierarchical structure incrementally extracts and integrates features at various levels, enabling the model to focus on global features in the higher layers while preserving fine-grained local details in the lower layers. Experimental results show HGASNet outperforming existing approaches, achieving R² of 0.9972, RMSRE of 0.0413, RMSLE of 0.0548, sMAPE of 6.47%, and MAPE of 6.72%.