Deep Learning Approach: YOLOv5-based Custom Object Detection

Abstract

Object detection is of significant importance in the field of computer vision, since it has extensive applications across many sectors. The emergence of YOLO (You Only Look Once) has brought about substantial changes in this domain with the introduction of real-time object identification with exceptional accuracy. The YOLOv5 architecture is highly sought after because of its increased flexibility and computational efficiency. This research provides an in-depth analysis of implementing YOLOv5 for object identification. This research delves deeply into the architectural improvements and design ideas that set YOLOv5 apart from its predecessors to illuminate its unique benefits. This research examines the training process and the efficiency of transfer learning techniques, among other things. The detection skills of YOLOv5 may be greatly improved by including these features. This study suggests the use of YOLOv5, a state-of-the-art object identification framework, as a crucial tool in the field of computer vision for accurate object recognition. The results of the proposed framework demonstrate higher performance in terms of mAP (60.9%) when evaluated with an IoU criterion of 0.5 and when compared to current methodologies in terms of reliability, computing flexibility, and mean average precision. These advantages make it applicable in many real-world circumstances.