Design and Implementation of Road Rutting Detection using MAnet with Efficientb0 Architecture

Abstract

Road rutting will become a serious problem in transportation infrastructure causing surface deterioration, safety concern, and increased maintenance expenses. The aim of this study is to establish an automatic and efficient detection model for discriminating road rutting, which can overcome the inconvenience of human-made reading with less errors. The present study develops the state-of-the-art knowledge in real-time and computationally efficient models for road rutting detection, focusing on effective operation of these universal tools under complex environments with different lighting conditions and surface material types. In the proposed approach to detect rutting with high accuracy, MAnet and efficientb0 architectures are used in combination. MAnet is an attention mechanism-aware network developed to extract more useful fine-grained features, by capturing the spatial and channel-wise dependencies between input images. Efficientb0: Efficientb0 which is the least size model and very computational efficient that allows our model to do inferences on real-time keeping accuracy unaltered. The experimental results confirm that the proposed model outperforms current state-of-the-art models (DeeplabV3 and U-Net) in performing semantic segmentation tasks for aerial images, obtaining a test set mIoU of 0.865. The experiment results indicate that the MAnet-Efficientb0 model is suitable for application in road maintenance system with high accuracy and computationally efficient.