Aplicação de inteligência artificial para detecção e segmentação em imagem de digitais de recém-nascidos

Abstract

The biometric identification of newborns is crucial for security and effective record-keeping in hospital environments, with papilloscopy being the most reliable technique. Fingerprint collection in newborns is a recurring challenge due to the difficulty in obtaining high-quality images, as a result of the very small ridge patterns and the low resolution of conventional scanners. Although high-resolution scanners represent the ideal solution, their financial impracticality justifies the proposal of using smartphones, widely accessible tools, as capture devices. This work proposes the development and evaluation of a low-cost scanner prototype that uses a smartphone and a prism to collect multiple fingers simultaneously, focusing on fingerprint identification and separation through a neural network. The methodology involved image acquisition for dataset creation and the training of a YOLOv8 neural network for accurate detection and segmentation of regions of interest in each image. The results demonstrated high segmentation performance, achieving an mAP50 of 100% and an mAP50–95 of 95% in validation, confirming the method’s effectiveness. It is concluded that this low-cost solution is a viable and efficient proof of concept. The main limitation lies in the need for field validation with a newborn dataset to confirm the final suitability of the prototype and the identification system.