Fingertip Blood Collection Point Localization Research Based on Infrared Finger Vein Image Segmentation

In the automatic collection of blood from the fingertip, it is necessary to squeeze the finger to increase bleeding due to the relatively tiny blood vessels and the comparatively small blood volume. Thus, this will cause cell fluid to enter the blood extracted and lead to inaccurate test results. If the vein-dense area of the finger is selected as the blood collection point, the amount of blood that can be extracted will be greatly increased. Due to the poor contrast of the infrared image of finger veins, it is difficult to effectively distinguish the vein area with the traditional segmentation method. Deep learning has made great achievements in image segmentation. However, due to the insufficient dataset of infrared image segmentation of finger veins, deep learning has hardly been applied in the image segmentation of finger veins. In order to solve the above problems, we created the dataset of the infrared image segmentation of finger veins and investigated how to use pyramid structure and mechanism of attention to segment the image of finger veins. We used a locally similar pyramid module to quantify a pixel to a similar degree of other pixels, adopted pyramid fusion module to enhance the obtained multiscale features, and proposed a network integrating distinct features of different weights, the details of which were highlighted. Based on the experimental results, this network shows a better performance compared with other advanced methods. It can better segment the finger vein region and, thereby, ensure the accuracy of blood sampling location.