A multitude of applications in engineering, ore processing, mineral exploration, and environmental science require grain recognition and the counting of minerals. Typically, this task is performed manually with the drawback of monopolizing both time and resources. Moreover, it requires highly trained personnel with a wealth of knowledge and equipment, such as scanning electron microscopes and optical microscopes. Advances in machine learning and deep learning make it possible to envision the automation of many complex tasks in various fields of science at an accuracy equal to human performance, thereby, avoiding placing human resources into tedious and repetitive tasks, improving time efficiency, and lowering costs. Here, we develop deep-learning algorithms to automate the recognition of minerals directly from the grains captured from optical microscopes. Building upon our previous work and applying state-of-the-art technology, we modify a superpixel segmentation method to prepare data for the deep-learning algorithms. We compare two residual network architectures (ResNet 1 and ResNet 2) for the classification and identification processes. We achieve a validation accuracy of 90.5% using the ResNet 2 architecture with 47 layers. Our approach produces an effective application of deep learning to automate mineral recognition and counting from grains while also achieving a better recognition rate than reported thus far in the literature for this process and other well-known, deep-learning-based models, including AlexNet, GoogleNet, and LeNet.

Accurately describing grain shapes is crucial in geology, mineral exploration, civil engineering, and other sciences. Advances in image analysis now allow for easy object separation and quantitative shape description. However, despite extensive applications in sedimentology, chemistry, and civil engineering, there is no consensus on the use of shape descriptors, and their meanings often remain unclear. This article presents a method for quantitatively describing grain shapes at a micrometer-to-centimeter scale using various image analysis techniques. Our approach selects the most appropriate combination of quantitative descriptors to describe grain shape. This work is based on an extensive literature review across many scientific fields to extract multiple quantitative shape measurements. This paper focuses on size, orientation, and form descriptors. A total of 51 descriptors, including elongation and Fourier amplitudes, were extracted, compiled, and computed using Python. The descriptor computation code is provided as a library with this article. We use principal component analysis to select the most significant descriptors and use multiple descriptors without losing clarity. We validated our approach on generated images. Using this combination of principal component analysis and image-based descriptors, we could discriminate 8 of the 13 ideal forms (ranging from a circle to a dodecagon), showcasing the potential precision when running noiseless data. The process was then applied to a sample of 584 galena grains, and we successfully described quantitatively the shape tendencies within this galena grain population. Our results, accompanied by noisy generated images, highlight the strong influence of roundness, roughness, and form descriptors on each other, which explains the challenges in identifying the best descriptors. This work demonstrates the potential of combining PCA with image-based shape analysis to improve the quantitative description of grains, with implications for sedimentology, mineral exploration, and other fields.

Identifying minerals is essential for geology, mineral exploration, engineering, and environmental sciences. Recent advances in machine learning have illustrated its potential as a fast, cost-effective, and reliable tool for identifying minerals from photographs or photomicrographs. However, in the recent literature, few studies have been dedicated to image acquisition. Machine learning generally requires reproducible, high-quality data to perform complicated tasks such as mineral identification to avoid common pitfalls. In this paper, we propose a practical image acquisition protocol for optical microscopes. This protocol focuses on ensuring reproducibility and enhancing image quality. To favor reproducibility, we detail dealing with camera errors, using reference color gauges, and establishing experimental parameters such as the external light source and temperature. For image enhancement, we explain the importance of lighting and its impact on machine learning precision, selection of the objective, and white balance calibration. In addition, we trialed the protocol on heavy mineral concentrate from till samples (20 species) with a typical deep learning model and it revealed that minor lighting modification (<5% difference in one channel) significantly increased misclassification rates: kyanite from 6.4% to 24.9% and monazite from 6.5% to 42.9%.

Geological processes are recorded in grain shape and geochemistry. However, grains are often described with minimal quantification. These descriptions are generally textual and can vary in their precision and accuracy. Historically, detailed studies of crystal size distribution have provided valuable insights into petrogenesis. A thorough analysis of numerous computable grain descriptors will offer even more significant information. Despite extensive literature on shape descriptors in fields like sedimentology, chemistry, and civil engineering, there is no consensus on their use, and their meanings often remain unclear. This article proposes a quantitative grain description method ranging from micrometers to centimeters using various image analysis techniques. Our approach consists of combining multiple quantitative descriptors to describe grain shape. This work is based on a comprehensive literature review across multiple scientific fields to extract numerous quantitative shape measurements. This paper focuses on roundness and roughness descriptors. A total of 25 descriptors, including Waddell roundness and fractal dimension, were extracted, compiled, and computed using Python. The descriptor computation code is provided as a library with this article. We use principal component analysis (PCA) to combine all descriptors in the same category without losing clarity and validated our approach on both generated and real grain images. For both roundness and roughness descriptors, the generated images and real grain images results are in accordance and could be summarized as follows. 1) The roundness descriptors PCA effectively distinguish grain shapes, performing comparably to form descriptors. However, it struggles to differentiate high degrees of roundness, and roughness significantly influences these results. 2) The roughness descriptors PCA excels at discriminating roughness intensity, despite the influence of form and roundness. These results align with our previous study on form descriptors and lead us to a new understanding of shape description: shape description includes both large-scale phenomena (“form”) and small-scale phenomena (“roughness”). And roundness is a specific case of shape description where various shapes transition into a circle. This study highlights the potential of using PCA alongside image-based shape analysis to enhance the quantitative description of grains, offering valuable implications for volcanology, planetary sciences, petrology and other fields.