Road Detection in Open-Pit Mining Areas Using D-DMCATNet Network Based on UAV Images

With the growth of the mining industry, efficient resource management and exploration have become key tasks. Traditional methods, relying on manual inspections or satellite remote sensing, are inefficient, costly, and limited by factors such as weather and time. Due to complex terrain and interference from various cover types, traditional road identification methods face challenges. A deep learning method, a double convolution multiscale attention network (D-DMCATNet), is introduced based on the U-Net structure. By combining multiscale convolutional blocks and channel attention mechanisms, the model captures features at different scales and levels of detail. Adaptive attention blocks and dropout mechanisms improve the model’s generalization and robustness. First, UAV images of mining roads were annotated to create a training dataset, and then different network architectures and optimizers were compared. The experimental results showed that D-DMCATNet achieved the best performance with an accuracy of 92.92%. Accurate road identification helps managers better plan and monitor resource development and provides data for other land classification tasks. This study is of great significance for mine resource management and exploration.