Multi-Objective Pavement Distress Detection with CNN and Transformer Deep Learning Architectures

Abstract: Abstract: The traditional methods for obtaining pavement condition data include manual and semiautomated surveys, which involve significant human intervention, are time-consuming and variable in repeatability and accuracy. Global public and private initiatives have often underestimated the complexities of developing fully automated pavement distress detection tools, primarily due to the challenge of replicating human cognitive abilities in computational processes. Recent advancements in deep learning (DL) provide promising solutions for true automated condition evaluation. While DL has proven effective for detecting pavement cracking, simultaneous detection of other distress types - such as potholes, patching, and sealed cracks - remains limited. This study utilizes 3D laser imaging data to extract scale- and mode-specific features for multi-objective DL detection across various pavement distresses. A diverse, high-resolution, manually annotated image library labeled with multiple distress categories was created for training, accompanied by intelligent data augmentation methods. Both traditional convolutional neural networks (CNN) and new-generation Transformer-architecture were developed for pixel-level semantic segmentation. Model performance was assessed and compared using accuracy, F1 score, and Intersection over Union (IoU) metrics. This multi-objective DL approach promises to revolutionize pavement condition surveys by enabling swift, one-pass assessments of multiple pavement distresses at highway speeds.

Keywords: Multi-object detection, Deep learning, convolutional neural networks (CNN), Transformer Architecture