Cybersecurity Considerations in Hyperspectral Imaging and AI for Food Quality Assessment

A recent study demonstrates the use of visible-near infrared (VIS-NIR) hyperspectral imaging and deep learning (DL) techniques to evaluate the quality of dried squid, a highly valued marine product in Eastern cuisines. The research involved acquiring and preprocessing hyperspectral reflectance images (400-1000 nm) from 93 dried squid samples. Key wavelengths were selected using competitive adaptive reweighted sampling (CARS), principal component analysis (PCA), and successive projections algorithm (SPA). A novel 1D-KAN-CNN network, based on the Kolmogorov-Arnold Network (KAN), was introduced for non-destructive measurement of fat, protein, and total volatile basic nitrogen (TVB-N) content. While the primary focus of the study is on food quality assessment, the integration of advanced imaging and AI technologies in this domain raises important cybersecurity considerations. The reliance on deep learning models for critical quality assessments introduces vulnerabilities to adversarial attacks, where manipulated inputs could deceive the system into accepting substandard products. Additionally, the integrity of hyperspectral data and AI models must be protected to prevent tampering that could compromise food safety. The adoption of such technologies in food processing and supply chains also expands the attack surface, necessitating robust cybersecurity measures to safeguard against both digital and physical threats. This study highlights the importance of securing advanced imaging and AI systems in critical applications to ensure their reliability and integrity.