การตรวจจับและการจำแนกใบหน้าสวมหน้ากากโดยใช้ข้อมูลเข้าหลายค่ากับการเรียนรู้เชิงลึก

Abstract

This research presents face-mask detection and classification using multiple inputs. The face-mask detection composes of 3 classes including the with_mask class, with_out_mask class, and wear_mask_incorrect class. The differences between these classes are the nose area and mouth area which help in classification. A deep learning multiple input model that can use face images, nose images, and mouth images as inputs was developed. This experiment is tried out with 2 datasets including Face Mask Label Dataset (FMLD) and Andrewmvd Face Mask Detection Kaggle (AFMDK). There are comparison models which are created by using single input and multiple inputs. The study finds that the results are confirmed that the purposed multiple input model has accuracy, precision, recall, and F1 score has higher values than a single input model in both datasets. This research also does an experiment on image enhancement by super-resolution for small image problems. The results increase the resolution at the nose and mouth area.The experiment shows that model trained by image from BSRGANs cannot solves the small image problem but can solve the medium and large images.