Acquisition of binary images and calculation of texture parameters for asphalt pavements based on the HTC-net model

Numerous testing techniques currently exist for evaluating pavement skid resistance, yet each faces specific application limitations and inherent shortcomings. To address this, this study proposes an innovative approach: performing semantic segmentation on asphalt pavement surface images using the HTC-Net model to obtain binary results of surface texture, thereby enabling precise quantification of pavement texture distribution parameters. First, a dataset was established and the images were preprocessed. Subsequently, an HTC-Net semantic segmentation network was constructed and trained using the preprocessed images to effectively distinguish aggregate particle regions from void regions. Finally, utilising the high-precision binary image generated from segmentation, multiple pavement texture feature parameters are calculated and correlated with the mean texture depth (MTD) measured by the manual sand spreader method. The study demonstrates that the extracted texture indicators—including the mean texture depth (MTD) and fractal dimension—exhibit excellent predictive accuracy and can serve as an effective basis for indirectly evaluating the skid resistance of asphalt pavements.