A dataset of AI-Ready standardized images for road surface conditions

<p indent="0mm">In response to the global scarcity of standardized datasets characterizing road weather conditions (RSCs), especially the lack of RSC images in extreme weather, this study develops an AI-ready dataset covering six categories of RSCs, designed to provide a high-quality and reliable training foundation for deep learning models. Based on statistical patterns of traffic impacts under extreme weather, the dataset categorizes snow and ice hazards into three subtypes: road ice formation, heavy snowfall, and snow blowing. Through systematic integration of highway surveillance footage, mobile device recordings, and open-source web data, we developed a standardized image repository covering six typical RSCs: dry road, fully snowy road, icy road, snow-blowing road, snow-melting road, and wet road. To mitigate subset correlation interference caused by data augmentation, we implemented a ″split-then-augment″ strategy. The original data was independently partitioned into training (60%), validation (20%), and test sets (20%), followed by separate augmentation operations for each subset, including random flipping, rotation, translation, and Gaussian noise addition. This methodology ultimately expanded the dataset to 9,000 images. The dataset provides preprocessed parameters optimized for deep learning training, with RGB channel means and standard deviations measured at [0550, 0565, 0568] and [0082, 0082, 0085] respectively. Zero-mean unit-variance normalization has been demonstrated to significantly enhance model training efficiency. This comprehensive dataset systematically covers multiphase characteristics of snow-ice affected road surfaces, establishing a standardized benchmark for development and evaluation of road state recognition algorithms in intelligent transportation systems.