An unmanned aerial vehicle-assisted visual measurement framework for forest road condition assessment with overlapping distress quantification

Routine forest-road maintenance is essential for road safety, accessibility, and serviceability, yet current inspections remain labor-intensive. Existing visual methods usually quantify pavement distresses independently, which may introduce area-estimation errors and affect condition assessment when multiple distresses spatially overlap. To address this engineering challenge, this study proposes an engineering artificial intelligence framework integrating Transformer-based semantic segmentation and diffusion-based data augmentation for unmanned aerial vehicle-assisted multi-distress measurement, overlapping-distress quantification, and forest-road condition assessment. This framework incorporates a segmentation model that converts aerial images into pixel-level distress masks for area measurement. The segmentation model employs a dynamic-gated convolutional backbone to integrate local textures with global context and a frequency-modulated spatial attention decoder to enhance small-scale and weakly visible distresses. A two-stage overlapping-distress identification strategy is developed to localize overlapping regions, identify their constituent distress categories, and support overlap-aware area quantification and condition assessment. A multi-condition guided diffusion model is further introduced to enrich inspection-image diversity, and a Forest Road Daily Inspection dataset is constructed for model training and evaluation. For engineering assessment, a Forest Pavement Condition Index is developed by incorporating measured areas of multiple distress types and overlapping regions, enabling section-level condition evaluation and maintenance prioritization. Experiments on the constructed forest-road inspection dataset and a public benchmark demonstrate that the proposed segmentation model improves multi-distress segmentation, overlapping-region identification, and measurement-oriented condition assessment, supporting its potential for forest-road condition measurement and maintenance-oriented assessment under the tested conditions.