Abstract: Joint fissures are one of the significant factors that influence the stability of open-pit mine slopes. With advancements in image processing and machine vision technology, the applications of intelligent algorithms for identification have attracted significant attention. Therefore, this paper proposes a method for identifying slope fissures in open-pit mines and deciphering geometric parameters, modernizing the U-net backbone network using residual network (ResNet) series algorithms for fast acquisition of joint fissure geometric information. The high-resolution images of open-pit mine slope fissures are collected using drones by considering factors such as viewpoint, distance, overlap rate, and flight speed. The images are subjected to preprocessing using the global threshold segmentation technique, and data augmentation is performed via random rotation, brightness, and contrast adjustment. The fissure image dataset then undergoes operations such as grayscale, threshold segmentation, dilation, hole filling, and the removal of small connected domain areas to eliminate the influence of background noise. Then, the U-Net network backbone is improved using five types of ResNet models: ResNet 18, 34, 50, 101, and 152. This led to the proposed slope fissure recognition model based on the improved U-net network, which uses the pixel binary classification problem’s accuracy, Intersection over Union (IoU), and F1 Score as evaluation indicators. In addition, the proposed model is trained and assessed using the fissure image dataset. The fissure binary image output is compared with that of traditional fissure recognition methods. The Res101-Unet algorithm achieved accuracy (Pa) and IoU of 96.23% and 62.13%, respectively, offering finer and more extensive fissure recognition results than other methods. Geometric parameter information, such as fissure length and width distribution rules and parameters, is calculated from the fissure binary image. The results show an improvement in the model evaluation indicators owing to the enhancement of the INet model by the ResNet model. Furthermore, the accuracy of the index evaluation increases with the depth of the network layers. The Res101-Unet model reached its highest evaluation index when the number of network layers reached 101, with accuracy, IoU, and F1 scores reaching 95.12%, 60.13%, and 79.53%, respectively. This scenario significantly improves the recognition of simple and complex fissures. As network layers deepen, fissure features can be captured from higher dimensions without substantially increasing network parameters. Thus, comprehensive and structurally distinct fissures can be obtained. The trained Res101-Unet model achieves the highest evaluation index upon reaching 101 network layers. Moreover, the number of recognized fissures on the target slope is consistent with the results obtained using the on-field measuring line method, confirming that the recognition results of this model are consistent with the actual engineering data.