The Multiple Inputs Ensembling for Face Mask Classification Using CNN

Object Detection can be used in many applications. Face mask classification for wearing a mask correctly, wearing a mask incorrectly, and not wearing a mask is a very important application for the recent situation. This research proposes the algorithm to ensemble multiple inputs for better decision making. The processes for this method are 1) face, nose, and mouth detections using Multi-task Cascaded Convolutional Networks (MTCNN), 2) feature extraction for face, nose, and mouth figures, and 3) face classification with multiple inputs ensembling using CNN. The datasets for this experiment are from Face Mask Label Dataset (FMLD) and Andrewmvd Face Mask Detection Kaggle (AFMDK). For both datasets, the experimental results show that the purposed model for Multiple Inputs Ensembling (MIE) has higher evaluation measurement of accuracy, precision, recall, and F1-score than those of the single input for both datasets.