Utilize Data Augmentation And Flexi Corner Block Diagram For Road Damage Detection

Abstract

This paper presents a novel and highly robust framework for automated road damage detection, a critical component in maintaining intelligent transportation systems and ensuring public safety. Traditional inspection methods are often labor-intensive, subjective, and prone to environmental inconsistencies. To address these limitations, we introduce an advanced deep learning architecture that integrates a customized Data Augmentation pipeline with an innovative Flexi Corner Block Diagram topology. The proposed Data Augmentation strategy specifically targets the inherent scarcity and class imbalance of real-world road distress datasets by synthetically generating diverse environmental conditions, illumination variations, and complex artifact occlusions. This ensures that the model generalizes effectively across heterogeneous pavement surfaces without overfitting. Crucially, the backbone architecture is enhanced by the Flexi Corner Block, a structural mechanism engineered to dynamically capture multi-scale spatial features and intricate boundary geometries of irregular pavement distresses such as longitudinal cracks, potholes, and severe alligator cracking. By adaptively adjusting its receptive field at the structural corners of the feature maps, the Flexi Corner Block preserves fine-grained topological details that standard convolutional layers typically smooth out or omit. Experimental evaluations conducted on benchmark pavement distress datasets demonstrate that our combined approach achieves superior localization accuracy and a significantly higher Mean Average Precision (mAP) compared to state-of-the-art object detection baselines. Furthermore, the lightweight design of the Flexi Corner component ensures high computational efficiency, making it highly viable for real-time deployment on edge devices mounted on municipal inspection vehicles. Ultimately, this research bridges the gap between highfidelity feature extraction and data-starved deep learning applications, offering a scalable, reliable, and automated solution for modern infrastructure asset management and predictive road maintenance scheduling.