Fluorite ore recognition using spectral clustering and smartphone digital images calibrated with a ColorChecker: A case study at the Lújar underground mine, Spain

• Pellets are photographed with a smartphone for fluorite grade recognition. • To get the genuine colors of the pellets, a color calibration chart is used. • Images of 479 pellets and their chemical composition are considered. • Pellets are grouped in three clusters based on color attributes of the images. • The recognition rate in waste in one cluster (C3), is up to 94.7 % by spectral clustering. Identifying and predicting mineral grade is crucial in mining engineering as it directly influences the efficiency and profitability of extraction processes. By accurately determining ore quality, mining operations can optimize resource allocation, reduce waste, and make environmentally responsible decisions. Over the past few years, the image-based methods have been proved to be effective since they can improve efficiency and reduce costs. In this study, pellet samples made from drilling chips are photographed by a smartphone to recognize fluorite grade. To avoid the impact of environmental factors on the image colors, four patches of the ColorChecker are used for color correction. The mean red and blue pixel intensities, uncertainty in the color clouds and texture are processed by a principal component analysis for developing five clustering models. After outlier detection, 479 pellets are retained for the development of clustering models. The resulting three clusters respond to changes in the mineral composition of the pellets, having each a different amount of ferric and aluminum oxides making possible to discriminate fluorite grade. The excellent recognition rate in waste for one of the clusters (C3), up to 94.74 % with the spectral clustering method, enables that new pellets could be deemed waste if the color attributes correspond to this cluster. For other clustering models, the recognition rate for the waste in cluster C3 is also effective but lower, in the range of 89.47 %–92.31 %. This result ensures a cheap and fast screening criterion of samples based on color properties of smartphone-captured images, hence lowering grading costs and increasing equipment availability. Compared with previous studies, this approach does not require expensive photography equipment and highly qualified operators, overcoming at the same time lighting effects. Such characteristics are of particular interest for small and medium-sized mine companies with limited access to expensive equipment. The present work is a step forward to use smartphone images in mine applications, such as modeling of ore deposits, mine planning, mineral processing and product quality assessment.