Long, Zekun, Zia, Ali, Nelis, Jordi, Rolland, Vivien, and Zhou, Jun (2024) A New Hyperspectral Unmixing Benchmark for Weak Signal Meat Contamination Detection. In: Proceedings of the International Conference on Digital Image Computing: Techniques and Applications. pp. 569-576. From: DICTA 2024: International Conference on Digital Image Computing: Techniques and Applications, 27-29 November 2024, Perth, WA, Australia.

PDF (Published Version) - Published Version Restricted to Repository staff only

Abstract

This study introduces the first hyperspectral image unmixing benchmark for weak signal detection, focusing on real meat contamination captured by hyperspectral cameras. We developed a real dataset and a synthetic dataset to evaluate the performance of various unmixing algorithms, including traditional methods (H2NMF and Hyperweak) and advanced deep learning techniques (DeepTrans and MiSiCNet). Our comprehensive assessment covers different concentrations of (E. coli) in sirloin steak samples, providing an indepth performance analysis of the tested models. Although no algorithm consistently outperforms all others, the experimental results indicate that DeepTrans performs particularly well in the conventional unmixing of fat and muscle. For weak signals such as saline solution or E. coli solution, Hyperweak produced better results on both datasets. In the synthetic dataset, Hyperweak achieved aSAD=0.0060 and aRMSE=0.0167, while in the real dataset, it reached state-of-the-art performance for weak signals in most scenarios. The scarcity of research on weak signal unmixing under challenging real-world conditions underscores the importance of this study, establishing a framework for future technological advancements in food safety.

Actions (Repository Staff Only)

Item Control Page