A dual-task convolutional neural network for ground cracks detection in the gof of coal mine. Treating the ground crack detection as a segmentation task and boundary extraction task. We designed a dual branch cnn to simultaneously obtain crack segmentation and crack skeleton extraction result. And a crack dataset was constructed using UAV high spatial resolution imagery, which can be used to train single or dual task DL model.