Personality Recognition Based on Palmistry Using Deep Learning YOLOv5 and YOLO-NAS

Abstract

Personality recognition through palmistry patterns often faces challenges in objectivity and interpretation accuracy. This study aims to identify personality based on the Big Five Personality psychology standards, using two advanced Deep Learning architectures: YOLOv5 and YOLO-NAS, for automatic palmistry pattern identification. Palmistry patterns can be used to represent personality based on both Palmistry and the Big Five Personality. The palmistry patterns used are palm line patterns, hand types, mount, and mercury/little finger. With image augmentation and machine learning methods, palm patterns can be recognized through object detection and classification using Deep Learning YOLOv5 and YOLO-NAS. The palm images data labeled with classes based on palm line patterns, hand types, mount, and fingers. The experimental results show that YOLOv5 is superior, achieving a Precision level of 0.800 compared to 0.104 in YOLO-NAS, and a Mean Average Precision (mAP) of 0.846 compared to 0.603 in YOLO-NAS. However, YOLO-NAS shows a higher Recall value of 0.894 compared to YOLOv5's 0.767. The unbalanced Recall and Precision results in YOLO-NAS are influenced by the use of the Quantization Technique, which transforms model weights to int8, resulting in lighter weights but potentially reducing precision.