Abstract: The joints and fissures produced by the surface deterioration of rock slopes in open pit mines are the key factors inducing landslides, and timely and accurate fissure detection is an important prerequisite for mining safety. Aiming at the problems of single fissure morphology generated by traditional data enhancement algorithms and the existing slope fissure monitoring methods being limited to local image analysis, a method for detecting fissuress in rock slopes of large-scale open-pit mines using Copy-Paste algorithm has been proposed, in which the fissure morphology is augmented by the Copy-Paste algorithm, and then a three-phase system workflow is established with sliding window algorithm-U-Net accurate segmentation-panoramic feature stitching to overcome the problems of lost and false detection of fissure features in large-scale scenes. The experimental results show that the proposed scheme results in a significant increase in the diversity of cleft samples, and the algorithm achieves 0.22% and 0.85% improvement in the segmentation IoU metrics when compared to the Mask R-CNN and U-Net models using traditional data enhancement, respectively. The slope project of Luoyang Luanchuan Longyu Molybdenum Mine is used for validation, showing the proposed three-phase system workflow has a positive role in promoting the identification and segmentation effect of large-scale fissures, with the mIoU value significantly increasing by 7.09% compared to that the three-phase system workflow is not used. The proposed scheme provides a new technical approach for disaster warning in steep rock slope engineering.