Application of concave point matching algorithm in segmenting overlapping coal particles in X-ray images. (1st September 2021)

• Record Type: Journal Article
• Authors: Sun, Aiyun; Jia, WenBao; Hei, DaQian; Yang, Yeyu; Cheng, Can; Li, JiaTong; Wang, ZhaoLian; Tang, Yajun
• Is Part Of: Minerals engineering. Volume 171(2021)
• Publication Date: 2021-09-01
• Subjects: XRT -- Coal-gangue Sorting -- Concave Point Matching -- Image Segmentation; Mines and mineral resources -- Periodicals; Ressources minérales -- Périodiques; Mines and mineral resources; Periodicals; Electronic journals; 622
• DOI: 10.1016/j.mineng.2021.107096
• Languages: English
• ISSNs: 0892-6875

Abstract

Highlights: Concave point matching algorithm (CPM) was applied for a coal-gangue sorting instrument. The area threshold method was developed in the CPM algorithm in order to define individual particles from overlapping images of two adjoining particles to improve ore sorting efficiencies. Only one of 33 tests failed to be segmented correctly using CPM algorithm, for a two-particle cluster. The calculation time of the CPM algorithm is small enough (below 3.06 ms, for two-particle clusters) to meet industrial needs. Abstract: When sorting coal and gangue by X-ray transmission technology, precise and reliable images are key to identify individual particles and calculate their centroids. However, the quality of the images is sometimes affected by adjoining particles causing an overlapping image, which will be handled as one item affecting the ore sorting efficiencies. In this work, we proposed a method for segmenting overlapping images of coal and gangue by using the concave point matching algorithm based on the macroscopic convexity of the ore. The area threshold selection method was improved to match the concave points caused by overlap. To validate the proposed algorithm, a coal-gangue sorter consisting of an X-ray tube and a dual-energy line array detector was developed and used for imaging of various overlapping objects. The algorithm was then applied to segment adjoining coal-gangue images. The results indicate that the segmentation of the overlapping images can be successfully achieved. In addition, the calculation time of this algorithm is shorter than 3.06 ms, which can meet the required time industrially.