Rapid detection of adulterated lamb meat using near infrared and electronic nose: A F1-score-MRE data fusion approach

Abstract

Individual detection techniques cannot guarantee accurate and reliable results when combatting the presence of adulterated lamb meat in the market. Here, we propose an approach combining the electronic nose and near-infrared spectroscopy fusion data with machine learning methods to effectively detect adulterated lamb meat (mixed with duck meat). To comprehensively analyse the data from both techniques, the F1-score-based Model Reliability Estimation (F1-score-MRE) data fusion method was introduced. The obtained results demonstrate the superiority of the F1-score-MRE method, achieving an accuracy rate of 98.58% (F1-score: 0.9855) in detecting adulterated lamb meat. This surpasses the performance of the traditional data fusion and feature concatenation methods. Furthermore, the F1-score-MRE data fusion method exhibited enhanced stability and accuracy compared with the single electronic nose and near-infrared data processed by the self-adaptive BPNN model (accuracy: 94.36%, 93.66%; F1-score: 0.9435, 0.9368). This study offers a promising solution to address concerns regarding adulterated lamb meat.

Keywords: Data fusion; Electronic nose; F1-score-MRE; Machine learning; Near-infrared spectroscopy.