Pavement raveling inspection using a new image texture-based feature set and artificial intelligence

Abstract

Enhancing pavement distress inspection is crucial within a pavement management system. Recognizing the advantages of automated pavement condition inspection, considerable efforts have been invested in developing more efficient systems for this purpose. Raveling, a common asphalt pavement distress, significantly impacts both passenger comfort and pavement durability. Given its connection to pavement texture and the limited effectiveness of earlier inspection systems, improving automatic raveling inspection has become a challenging endeavor. This study proposes an image-based system for raveling detection, employing a new image processing-based feature set. This approach generates enhanced inputs for artificial intelligence models, thereby improving the efficiency of raveling inspection while optimizing resource allocation. To accomplish this, an image dataset is created by combining newly collected images with an existing collection. Novel combinations of features are then engineered to extract meaningful texture-based image features. Subsequently, three feature selection algorithms are employed to identify the most pivotal features. Machine Learning (ML) models are then trained for effective raveling detection based on the proposed feature set, and their performance is evaluated across various performance indices. The achieved performance metrics, reaching nearly 96 % on the test datasets, underscore the efficacy of the proposed method in automating the inspection of texture-based pavement distresses.