Abstract
: Image segmentation is an important part of ore particle size detection, and the quality of image segmentation directly affects the accuracy and reliability of particle size detection. Due to the poor quality and low efficiency of ore particle image segmentation in ore size detection, developing a fast and accurate algorithm for segmenting ore particle images remains a global challenge. However, the quality of image segmentation is closely related to calculating ore density, improving beneficiation efficiency, and evaluating crushing effectiveness. In this paper, a novel image segmentation algorithm is proposed, combining the K-means algorithm with a hybridized IGK-means. Firstly, the IGWO_SOA, by introducing a nonlinear convergence factor and incorporating the migration and spiral search mechanisms of SOA, is applied to overcome the weakness of being sensitive to initial centroids of the traditional K-means. IGWO_SOA is utilized to iteratively search for the optimal values of the initial cluster centers, which are then output as the results for subsequent clustering segmentation. An industrial experiment was conducted for multiple comparisons, which proved that the IGK-means has the characteristics of better image segmentation quality and being insensitive to illumination. The PSNR of the images segmented by IGK-means can reach up to 24.24 dB, and the FSIM can reach up to 0.2733, which proves the superiority and practicality of the algorithm in this paper.