The remote sensing method for large-scale asphalt pavement aging assessment with automated sample generation and deep learning.

The evaluation of asphalt pavement aging is a critical foundation for road maintenance decision-making and plays a significant role in ensuring traffic safety. Traditional monitoring methods often rely on manual field surveys, which are time-consuming and inefficient. This study proposes an innovative framework that integrates multi-endmember mixed pixel unmixing, automated sample generation, and deep learning. The goal is to enable rapid and accurate assessment of asphalt pavement aging over large areas. Based on WorldView-3 remote sensing data, high-quality training and validation samples were generated using multi-endmember spectral unmixing and neighborhood filtering. The Jeffries-Matusita (J-M) distance values exceeded 1.9 for model training samples and 1.7 for validation samples. In addition, a one-dimensional convolutional neural network (1D-CNN), combined with an unsupervised zero-shot transfer approach, was employed for model training and inference. The proposed framework was applied to several study areas. Overall classification accuracy and Kappa coefficients reached 95.95% and 0.9459 in study area-I (the south-central part of the Wuhan Economic and Technological Development Zone), and 89.70% and 0.8628 in study area-II (the northern part of the zone). This study highlights the effectiveness of the proposed framework for rapidly assessing asphalt pavement aging over large areas. The results are valuable for pavement maintenance and traffic safety warning applications.