Abstract: Invasive blood glucose measurement has a strong sense of discomfort and risk of infection, so the study of non-invasive blood glucose has a strong practical significance. At present, the optical method is not convenient for practical use, and the energy conservation method requires strict requirements. In view of the above problems, infrared thermography is used to detect blood glucose. After acquiring infrared thermal images of face figure, we extract the gray feature and reduce its dimension. In order to speed up the training and prevent over fitting, depth regression network is improved to model the infrared thermal image gray feature, and the ideal testing results have been achieved in the test set, which provides a new method of research and design for the noninvasive blood glucose detection algorithm research.

Key words: noninvasive blood sugar detection, improved depth regression network, infrared thermography, image feature extraction, pca dimensionality reduction