Paper Key : IRJ************552
Author: Md. Nahid Sultan,S M Sojib Ahamed,Hossain Imrn,Md. Arman Hossain,Md Nadiruzzaman Nahid
Date Published: 13 Jul 2024
Abstract
The task of detecting abnormal objects is of utmost importance and finds applications in diverse sectors, ranging from security and surveillance to industrial quality control. Deep learning models have significantly transformed the computer vision field, demonstrating significant potential in real-world anomaly detection. This study utilizes YOLOv5, a cutting-edge real-time object detection model, to create a high-performing and precise abnormal item detection system. This paper discusses the challenges of detecting aberrant objects, emphasizing the need for real-time, resilient solutions, and introduces the YOLOv5 model for its speed and accuracy. Additionally, we outline its modification specifically tailored for the purpose of abnormal object detection. By doing fine-tuning on the YOLOv5 model using a specific dataset that consists of both normal and abnormal objects, we are able to customize the network to achieve superior performance in accurately identifying anomalies. In order to improve the efficacy of the model, we introduce a unique loss function that integrates both classification and localization losses. This approach aims to optimize the model's capacity to accurately identify and precisely determine the location of anomalous items. In addition, we investigate the potential of transfer learning to enhance the model's abilities in effectively addressing a wide range of scenarios and object categories. The empirical findings of our study provide evidence supporting the efficacy of the YOLOv5-based deep learning framework for the timely identification of anomalous items in both dynamic video streams and stationary photos. The model integrates YOLOv5 with deep learning techniques for accurate object detection, minimizing false alarms and ensuring prompt anomaly identification, making a significant contribution to computer vision. The proposed system, utilizing deep learning and YOLOv5, is suitable for various applications like security, surveillance, industrial automation, and healthcare. It addresses the need for efficient object detection in complex environments, contributing to cost-effective, digitally fortified facilities.
DOI Requested