Enhanced Real-Time Pothole Detection Using YOLOv8 with EfficientNet Backbone: A Deep Learning Approach for Road Safety

Potholes present substantial problems to road safety and infrastructure maintenance, frequently resulting in accidents, vehicle damage, and increased repair costs. Manual inspection methods are inefficient and unfeasible for large roadway networks. This paper presents a new method to real-time pothole identification based on a customized YOLOv8 segmentation model that replaces the usual backbone with EfficientNet-B0. EfficientNet-B0, noted for its remarkable feature extraction and parameter efficiency, has been integrated to improve the precision of detection while retaining computational feasibility. The approach suggested outperforms the baseline YOLOv8 architecture in terms of mAP, precision, and recall, with a fitness score of 0.976. Extensive testing demonstrates the approach's resilience as well as utility in identifying various pothole scenarios. This study highlights the potential of deep learning-driven breakthroughs for automated transportation maintenance systems.