Conv-trans dual network for landslide detection of multi-channel optical remote sensing images

Landslide detection is crucial for disaster management and prevention. With the advent of multi-channel optical remote sensing technology, detecting landslides have become more accessible accurate. Although use convolutional neural network (CNN) has significantly increased accuracy landslide on images, most previous methods using CNN lack ability to obtain global context information due structural limitations convolution operation. Motivated by powerful modeling capability Swin transformer, we propose a new Conv-Trans Dual Network (CTDNet) based Swin-Unet. First, dual-stream module (CTDBlock) that combines advantages ConvNeXt which can establish pixel-level connections dependencies from hierarchy enhance model extract spatial information. Second, apply an additional gating (AGM) effectively fuse low-level extracted shallow high-level deep minimize loss detailed when propagating. In addition, We conducted extensive subjective objective comparison ablation experiments Landslide4Sense dataset. Experimental results demonstrate our proposed CTDNet outperforms other models currently applied in experiments.