A lamb freshness detection model using a flexible optoelectronic in-situ sensing system and multi-input multi-label causal ensemble learning

Wenhao He; Wentao Huang; Yunpeng Wang; Zhigang Li; Tobolková Blanka; Xiaoshuan Zhang

doi:10.1016/j.foodchem.2025.142803

A lamb freshness detection model using a flexible optoelectronic in-situ sensing system and multi-input multi-label causal ensemble learning

Food Chem. 2025 Jan 7:471:142803. doi: 10.1016/j.foodchem.2025.142803. Online ahead of print.

Authors

Wenhao He¹, Wentao Huang¹, Yunpeng Wang¹, Zhigang Li², Tobolková Blanka³, Xiaoshuan Zhang⁴

Affiliations

¹ College of Engineering, China Agricultural University, Beijing 100083, PR China.
² College of Information Science and Technology, Shihezi University, Shihezi 832000, PR China.
³ Department of Chemistry and Food Analysis, National Agricultural and Food Centre-Food Research Institute, 82475 Bratislava, Slovakia.
⁴ College of Engineering, China Agricultural University, Beijing 100083, PR China. Electronic address: zhxshuan@cau.edu.cn.

PMID: 39799687
DOI: 10.1016/j.foodchem.2025.142803

Abstract

Efficient, non-destructive and real-time meat freshness assessment has always been a hot research topic. This paper presents a novel approach for detecting lamb meat freshness using a flexible optoelectronic sensing system combined with an integrated learning model. We developed a flexible impedance sensing system and a flexible optical sensing system through laser direct writing and transfer technology. Freshness of lamb samples was evaluated under various storage conditions (0 °C, 4 °C, and 8 °C). Our results demonstrated that the system offers notable portability, stability, and accuracy compared to traditional methods. Statistical analysis revealed a moderately strong correlation (r > 0.86) among physicochemical properties, impedance measurements, and spectral data of lamb meat. The Granger causality test indicated a causal relationship between impedance and spectral data (p < 0.05). Data were analyzed using our 1DCNN-BiLSTM-ATT model for freshness grading, achieving an accuracy of 94.57 %, significantly outperforming traditional algorithms. This paper provides an innovative solution for the efficient and accurate prediction of lamb freshness.

Keywords: Deep learning; Flexible optoelectronic sensing system; Lamb freshness; Multi-input; Non-destructive detection.