Hyperspectral Imaging Combined with Deep Learning for Food Composition Analysis: Enhancing Intelligent Engineering Systems through Cyber-Physical Integration

Abstract

Combining hyperspectral imaging (HSI) with deep learning (DL), we can quickly measure the levels of sugar, protein, fat and counterfeit products without destroying food. This paper reviews what happened in system engineering, network physical system (CPS) and opinions based on information theory from 2019 to 2023. We regard HSI-DL as an important module to observe and make decisions in the intelligent food system. According to this theory, spectral data is like an information source with a large number of dimensions, while DL encoder needs a lot of information, which is helpful for closed-loop control in CPS architecture. We studied the methods of extracting spectral and spatial features, reducing the size and inferring nearby equipment, and the integration of online detection and reinforcement learning of computer vision, and planned in an adaptive way. The application of strawberry maturity assessment, apple internal quality grading, grain protein prediction and meat adult detection shows great improvement in accuracy and throughput, but there are also limitations in domain generalization, model interpretability and real-time calculation constraints. The original idea provides an information-optimized perceptual layer in digital dual-enabled CPS, which maximizes mutual benefit between spectral signatures and variable composition of information, while respecting delay and energy limitations. In conclusion, interdisciplinary verification is needed to make flexible and sustainable food engineering systems using AI.