Abstract: In the research of image segmentation methods for open-pit mine blast piles, aiming at the problems of low accuracy and poor generalization ability of traditional segmentation methods caused by the characteristics of blasting pile image, such as complex background, large target scale difference and fuzzy edge, an open-pit mine blast piles image segmentation methods based on RPD-UNet is proposed to realize high-precision segmentation of blast piles. The RPD-UNet model takes u-net of encoder decoder structure as the basic framework, and introduces the pre trained resnet34 as the backbone feature extraction network. Its deep residual structure can alleviate the difficulties of deep feature extraction and gradient dissipation caused by the complex background and texture diversity of block image. Aiming at the problems of key feature ambiguity and noise interference caused by complex lighting, shadow occlusion and similar foreground and background, a parallel CBAM module operating on both spatial and channel dimensions is proposed. The parallel processing can simultaneously process the channel and spatial information, enhance the focus of the model on the key areas, and make the model more accurately lock the blast pile and detail features. Aiming at the problems of high computational complexity and low computational efficiency of the model, the deep separable convolution (DSC) is used to replace the standard convolution operation, which solves the problem of low computational efficiency of the model without losing or even improving the ability of the model. In order to verify the effectiveness of the proposed RPD-UNet model, the results of explosive pile segmentation were compared with the results of mainstream segmentation models. The comparison results showed that the mIoU and mPA and Precision and F₁-score of RPD-UNET were 89.90%、94.54%、94.52% and 94.51%, respectively; In terms of reasoning efficiency, the processing time of a single image is 91 ms, which is better than the mainstream segmentation model. The RPD-UNet model not only improves the segmentation accuracy and generalization ability, but also ensures the computational efficiency of the model. The model provides a reference for the intelligent identification of blast pile in open-pit mines and the assessment of blasting quality.