Pothole Detection and Classification Using Enhanced EfficientNet Optimized by Advanced Manta-Ray Algorithm

Abstract

Object detection powered by neural networks has transformed artificial intelligence applications, achieving notable advancements in numerous domains, including the automated detection of road potholes. This research introduces a novel methodology to improve pothole detection and classification accuracy and efficiency by integrating the Manta Ray Foraging Optimization (MRFO) algorithm into EfficientNet. The MRFO, inspired by the collective foraging behavior observed in manta rays, is implemented as a replacement for the conventional Adam optimizer in EfficientNet. This integration strengthens feature extraction by effectively managing the trade-off between global exploration and local exploitation. The developed model successfully overcomes prominent optimization challenges, leading to a substantial gain in classification accuracy  from 84% to 93%. Comprehensive experiments were conducted across the B0, B1, and B2 configurations of EfficientNet, comparing the performance of MRFO against traditional optimization methods. Results consistently demonstrate that MRFO significantly enhances pothole image detection and classification capabilities. This study highlights the potential of MRFO as a robust optimization tool for real-world object detection tasks that can further improve its broader applications in intelligent transportation and beyond.