A method for detecting the dissemination size of metal minerals under the microscope based on deep learning

Abstract

Objective The embedded granularity of target minerals refers to their particle size and distribution in ore, which directly influences the design and effectiveness of ore beneficiation processes. Therefore, the measurement of embedded granularity of target minerals is a crucial task in process mineralogy research. In traditional process mineralogy research, the main method for observing and analysing ore samples is to use a polarized light microscopy. However, this approach generally suffers from problems such as long processing times, susceptible to subjective factors, and difficulties in achieving automation and large-scale applications. To overcome these limitations, this paper proposes a method for detecting the embedded granularity of metal minerals under the microscope based on based on deep learning. Methods This method takes specimens from the Shuichang magnetic ore deposit in Tangshan City, Hebei Province as the object. We take photographs under the reflected light conditions using a polarized light microscope to create a dataset, and design a mineral recognition network model based on the Deeplabv3+ network. Thereby it enables automated feature extraction and intelligent recognition of the target metal minerals. This method achieves segmentation of the target minerals by generating binary images of the desired metal minerals. Finally, the analysis and measurement of the embedded granularity of the target metal minerals are completed by combining the maximum Feret diameter. Results Compared with traditional manual microscopic measurement methods, the application of image measurement based on deep learning in mineral particle analysis has increased the processing speed by approximately 119.8 times and the measurement accuracy 169.5 times when measuring the same mineral particles, demonstrating its significant improvement in terms of processing efficiency and measurement accuracy. Conclusion The deep learning-based method for detecting the embedded granularity of metal minerals under a microscope significantly reduces the detection time and enhances the detection accuracy for mineral embedded granularity. Moreover, it eliminates the influence of subjective factors, which is of great significance for promoting the intelligent development of process mineralogy.

Keywords

• deep learning
• recognition of mineral
• particle size detection
• process mineralogy
• feret diameter
• network model
• automated feature extraction
• intelligent recognition