Spectral Sensing Detection Technique Combined With an Adaptive Deep Learning Method (SFCC-Net) to Identify the Mung Beans of Different Origins

Mung beans from diverse geographical origins exhibit significant variations in nutritional composition and quality attributes due to the combined effects of geographical conditions, climatic patterns, and soil characteristics. Effective origin tracing techniques play a pivotal role in safeguarding consumer rights and combatting market adulteration. In this work, we propose a spectral detection system integrated with an adaptive deep learning method for geographical identifying of mung beans. First, a hyperspectral system is used to collect mung bean samples from six distinct production regions. Subsequently, we design a Spectral Features Calculation Module (SFCM) that implements deep spectral feature computation through synergistic integration of lightweight operations and attention mechanisms, specifically tailored for spectral information characteristics. Building upon this foundation, we construct a Spectral Features Calculation and Classification Network (SFCC-Net) to attain the origin identification of mung beans. Comprehensive evaluation through structure optimization, ablation studies, and comparative analysis with state-of-the-art spectral classification methods demonstrate the superior performance of SFCC-Net, achieving the best performance, with an accuracy of 98.42%, a precision of 98.69%, and a recall of 98.40%. In conclusion, the integration of spectral sensing detection technique with SFCC-Net establishes a reliable, accurate, rapid, and nondestructive method for geographical authentication of mung beans, which is of great significance for ensuring product quality and promoting fair trade in agricultural markets.