Abstract: This paper proposes a K-means based algorithm to identify the interface of coal and rock. Firstly, we use wavelet transform to extract large-scale features in coal-rock images, which eliminates spurious textures and imaging noise and thus facilitates the subsequent clustering process. Then, the K-means algorithm is applied to complete the clustering of coal-rock interface distribution. Finally, image edges are extracted from the clustered binary image using the Canny operator, and two image morphological operators, erosion and dilation, are used to connect adjacent segments and smooth the boundaries. The experimental results of simulated and real images show that our algorithm is more accurate to extract the true coal-rock boundaries, compared with the direct K-means and Mean-shift image segmentation algorithms. The proposed algorithm is more promising to the autonomous long arm mining applications.

Key words: coal-rock interface detection, K-means, Canny edge detection, erosion and dilation