IDENTIFICATION OF SURFACE STRUCTURAL-LAYER IN ROAD CONSTRUCTION USING LOW-ALTITUDE UAV IMAGERY

Monitoring and assessing the progress and quality in road construction are crucial aspects of infrastructure development.Among these, the identification of surface structural-layer serves as a key task for progress control and quality management.Traditional manual inspection methods are inefficient and costly, especially given the spatially linear and narrow nature of road construction areas, making them inadequate for smart construction management.The application of low-altitude UAV imagery offers a new opportunity to address these challenges.However, its effective utilization hinges on the precise identification of surface structural-layer during construction.Conventional segmentation techniques often struggle with noise interference, leading to limited accuracy in identifying surface structural-layer.To overcome this, this paper proposes an improved DeepLabV3+ model for the identification of surface structurallayer in road construction.The model enhancement involves integrating a ResNet-50 backbone to bolster feature representation and extraction capabilities, thereby mitigating overfitting risks.Additionally, a CBAM (Convolutional Block Attention Module) is incorporated into the ASPP (Atrous Spatial Pyramid Pooling) module to enhance model performance, particularly in capturing fine details and boundary information.A self-constructed dataset is utilized, partitioned into training, testing, and validation sets in an 8:1:1 ratio.The performance of the proposed model is evaluated using four key metrics: overall accuracy, mean accuracy, frequency-weighted accuracy, and mean intersection over union.Comparative analyses demonstrate the superiority of the improved DeepLabV3+ model over the baseline in terms of segmentation accuracy.This paper provides a new method for the monitoring and assessment of road construction progress and quality, advancing the application of low-altitude UAV technology in intelligent construction practices.