Hyperspectral Deep Learning for Food Freshness Assessment and Spoilage Detection

Abstract

Reliable food freshness assessment is essential for improving quality control and reducing losses throughout food production and distribution systems. Conventional inspection methods often rely on destructive laboratory testing or subjective visual evaluation, limiting their suitability for continuous monitoring. This study proposes a hyperspectral deep learning framework for non-destructive food freshness classification and spoilage detection. The proposed method integrates spectral feature extraction with a lightweight convolutional attention network to characterize chemical and structural changes associated with food deterioration. The illustrative dataset consists of 38,600 hyperspectral observations obtained from five categories of perishable food products under varying storage temperatures and humidity conditions. The proposed model achieves an overall classification accuracy of 95.1%, a macro-averaged F1-score of 0.937, and a spoilage detection sensitivity of 94.6%. Compared with a conventional support vector machine baseline, classification accuracy improves by 7.8 percentage points. The average processing time is reduced to 42 milliseconds per sample, enabling near-real-time assessment. These findings highlight the potential of hyperspectral artificial intelligence for scalable and non-destructive food quality monitoring.