Efficient Object Detection and Intelligent Information Display Using YOLOv4-Tiny

Abstract

This study aims to develop an innovative image recognition and information display approach based on you only look once version 4 (YOLOv4)-tiny framework. The lightweight YOLOv4-tiny model is modified by replacing convolutional modules with Fire modules to further reduce its parameters. Performance reductions are offset by including spatial pyramid pooling, and they also improve the model’s detection ability for objects of various sizes. The pattern analysis, statistical modeling, and computational learning visual object classes (PASCAL VOC) 2012 dataset are used, the proposed modified YOLOv4-tiny architecture achieves a higher mean average precision (mAP) that is 1.59% higher than its unmodified counterpart. This study addresses the need for efficient object detection and recognition on resource-constrained devices by leveraging YOLOv4-tiny, Fire modules, and SPP to achieve accurate image recognition at a low computational cost.