Segmentation of ore and waste rocks in borehole images using the multi-module densely connected U-net

During exploitation of metal mines, the delineation of orebody boundary is of great importance to control the dilution rate, ensure the mining efficiency and improve the economic benefits. At present, the delineation of orebody boundary is usually achieved by conventional methods, such as laboratory tests on rock powder, identification of ore samples under microscope and spectrum analysis etc. However, as the actual condition of orebody occurrence is extremely complex, the conventional delineation methods can hardly meet the requirements of full-length real-time efficient analysis for boreholes. As a result, the blasting design is out of step with the blasting mining process, leading to great economic loss. In this paper, a method for intelligent identification and automatic delineation of ore and waste rock boundary is proposed, based on the borehole imaging technology and the image segmentation method. The digital borehole imaging system is employed to obtain the borehole images in mine roadways. The Inception and DenseNet blocks are integrated to develop the multi-module densely connected U-net (MMDC-Unet), which can realize pixel-level semantic segmentation of borehole images by segmenting the borehole images of highly similar and highly randomly distributed ore according to the global and local image features. Based on the image segmentation results, the number of pixels for ore is counted to quantitatively analyze the distribution of ore and realize real-time rapid delineation of ore and waste rock boundary. By comparing the intelligent identification results with the chemical test results, it is found that the proposed method can greatly enhance the identification efficiency while satisfying the engineering requirements. •Proposes a method of ore and rock recognition based on borehole images and deep learning.•Propose a new network for ore image segmentation.•Based on the results of image segmentation, the distribution of ore and rock can be described quantitatively by quantifying the number of ore and rock pixels.Automatic detection of ore can be realized by calculating the density of ore pixels.•This method is more convenient and efficient than chemical analysis, microscopic image and spectral analysis.