Hyperspectral Image Visualization Through Neural Network For The Food Industry

Hyeok Yoon, Jungi Lee

Abstract: Hyperspectral images (HSI) visualization is important for data annotation, which is used for finding and analyzing foreign objects in food based on machine learning or deep learning techniques. However, HSI visualization is challenging due to the numerous bands compared to RGB images. Recently to address the problem, a lot of methods have been proposed such as color matching function. The color matching function is a representative method based on the human visual system and returns a static value. However, the property sometimes makes results indistinguishable due to the failure to catch the spectrum features. In this paper, we propose a mapping method from HSI to RGB through the neural network that learns the relationships between spectrum and RGB values directly. Our proposed method is evaluated on the dried radish and filefish data obtained from the near-infrared hyperspectral sensor.