YOLO-NAS Based Low-Power CNN Hardware for Digital Number Recognition: Design, Optimization, and Implementation

This study aims to identify the most accurate and reliable model for digit recognition in photographs. The models were tested using various metrics such as classification loss, accuracy, recall, mean average accuracy (mAP), and F1 score. YOLO-NAS was found to be the most effective, with a classification loss of 1.2, accuracy of 0.85, recall of 0.90, and mean absolute performance of 0.80. This indicates that YOLO-NAS is valid and competent for digit identification tasks. However, YOLOv8 and YOLOv5 showed significant deficiencies in precision and overall accuracy, indicating a need for further optimization in digit recognition applications.