Research on segmentation and reconstruction of overlapping ore contours based on EAM-SOLOv2 and convex hulls

In order to solve the problem of inaccurate segmentation of overlapping ores, a method for ore image segmentation based on the improved SOLOv2 model is proposed in this study, and convex hulls are used for reconstructing the contours of overlapping ores. An edge-aware module (EAM) that enhances the edge information is introduced to address the problem of mis-segmentation of ore contours due to the overlap problem. This enables the model to make full use of the edge information of the ore image in the feature fusion process. In this study, the masking of overlapping ores is also processed by using the convex hull contours of the ores to establish a formula for the masking relationship between the ores, and then reconstructing the contours of the occluded ores using a cubic B-spline curve. The experimental findings show that the improved SOLOv2 model is more effective in segmenting adherent and overlapping ores. Additionally, after fitting and reconstructing the occluded ore contours, the overlap ratio between the reconstructed contours and the real contours is significantly increased, which makes the contours closer to the real contours of the ore. Therefore, this study can provide a basis for the automatic detection of ore particle size.