A Comparative Analysis of Deep-Learning-Based YOLO Models (V8n and V8s) for Object Detection Using GSV Images

This study presents a comparative analysis of deep-learning-based YOLO (You Only Look Once) models, namely YOLOv8n and YOLOv8s, for detecting power poles along a street in Nellore, Andhra Pradesh. The objective is to assess how well and efficiently these models accurately detect power poles in Google Street View (GSV) images. The study utilizes a dataset consisting of street view images that are annotated and used for training the YOLOv8n and YOLOv8s models, which are then tested on a set of different images. To verify the models' accuracy and effectiveness in recognizing power poles, evaluation criteria like precision, recall, and F1 score are used. The results indicate that both the YOLOv8n and YOLOv8s models are effective in detecting power poles along the street in Nellore. However, the YOLOv8s model has a greater accuracy of 81% compared to the YOLOv8n model's accuracy of 78%. The findings of this work demonstrate the potential of deep-learning-based YOLO models for power pole detection using GSV images. The comparative analysis offers valuable insights for researchers and individuals in the GIS and computer vision field, contributing to the development of efficient and accurate methods for infrastructure monitoring and management.