Advancing Mineral Stone Classification C02.pptx

For ML in mineral stones

1. Advancing Mineral Stone Classification in Afghanistan through Image Processing Techniques By: Iqbal Samsoor Candidate for the Masters Degree scholarship Fall 2023 1 Thursday, November 30, 2023

2. Agenda Introduction Research Questions Research O bjectives Research Motivations Literature Review Methodology Data Collection and Preparation Dataset Preparation Model Development and Training Model Validation and Performance Evaluation Research Contribution Q&A 2 Thursday, November 30, 2023

3. Introduction Afghanistan is endowed with abundant natural resources that remain largely untapped. Afghanistan's these mineral-rich resources are crucial for economic development , but mineral classification faces challenges due to limited technological resources and expertise . Mining sector in Afghanistan  is currently controlling by the Ministry of Mines and Petroleum Afghanistan. T he country has world-class deposits of iron ore, copper , gold, rare-earth minerals, and a host of other natural resources. According to a joint study by The Pentagon and the United States Geological Survey, Afghanistan has an estimated US$1 trillion of untapped minerals . [1 ] This research will harness technological advancements to enhance mineral stones identification processes in Afghanistan, contributing to economic development and geosciences . Thursday, November 30 , 2023 3

4. Intro- Research Questions How can image processing techniques be effectively applied to classify mineral stones in Afghanistan? What is the comparative accuracy and efficiency of the image processing-based classification system versus traditional methods ? What are the potential challenges and solutions in implementing this technology in Afghanistan’s mining sector? Thursday, November 30 , 2023 4

5. Intro- Research Objectives To develop an image processing-based model for accurately classifying different types of mineral stones found in Afghanistan To compare the efficiency and accuracy of the developed model against traditional classification methods To assess the feasibility and scalability of implementing this technology in the Afghan mining sector Thursday, November 30 , 2023 5

6. Intro- Motivations Empowering Afghanistan's resource wealth : to revolutionize mineral stones classification in Afghanistan, as the mining sector in Afghanistan is a significant contributor to the country's economy. Breaking Barriers : I t faces several challenges, including the manual and often inaccurate classification of mineral stones . Unveiling Hidden Treasures : This research will explore recent advances in image processing and machine learning technologies to develop a more reliable and efficient classification system, which is currently a gap in the local mining industry . Driving Economic Growth : Advancing mineral stones classification for a thriving Afghanistan. Thursday, November 30 , 2023 6

7. Literature Review In august 2022 Mauro Tropea and his collogues [2] have written a paper and developed an automatic recognition system using Convolutional Neural Networks (CNNs) for classifying stones . The study proposes a two-stage hybrid approach, utilizing Deep Learning (DL) with CNNs in the first stage and Machine Learning (ML) models ( Softmax , MLR, SVM, kNN , RF, GNB) in the second stage . A work in Feb 2022 by Fatih Akkoyun with his team [3] done for industrial white quartz stone classification using image processing and supervised learning . In this system Four different images are captured under four different angles and processed to extract visual parameters of each stone sample. In training stage 67% of the data were used for training and rest were used for testing process. The method correctly classifies mine stones up to 98 %. In June 2010 M. Lopez and their team [4] developed a system for functional classification of ornamental stone using machine learning techniques . The authors focus on models with functional inputs and utilize Support Vector Machine (SVM) and Neural Network (NN) techniques. Thursday, November 30 , 2023 7

8. Methodology- Data Collection and P reparation Data Collection Sampling Locations: The study will strategically select sampling from the museum of the Ministry of Mines and Petroleum of Afghanistan. besides , we may have direct input from the technical cadre members from the faculty of Geology at Kabul University . Necessary permissions and agreements for data collection will be secured to adhere to legal and ethical standards. Data Preparation Image Acquisition Process The RGB camera used for image acquisition. Specifications such as pixel density and image format will be chosen to optimize image quality for processing. Preliminary Image Analysis A preliminary and initial review of the captured images will be conducted to assess quality and consistency. Any anomalies or issues in imaging will be addressed before proceeding to the dataset preparation phase. Thursday, November 30 , 2023 8

9. Methodology- Dataset Preparation Dataset Preparation Image Preprocessing The initial step involves cleaning the images to remove any artifacts or noise that might affect the analysis. Resizing and Normalization Images will be resized to a uniform dimension suitable for the CNN input . Normalization will be conducted to scale pixel values to a standard range, typically between 0 and 1, enhancing model performance. Training , Validation, and Test Sets The dataset will be divided into training, validation, and test sets. A common split ratio could be 70% for training, 15% for validation, and 15% for testing. Thursday, November 30 , 2023 9

10. Methodology- Model Development and Training CNN Architecture Convolutional Neural Networks (CNNs), recognized for their superior performance in image classification tasks, will be the cornerstone of our model development. Training Process The CNNs will be meticulously trained using the rich mineralogical data obtained , focusing on feature extraction and pattern recognition pertinent to mineral stone classification. Thursday, November 30 , 2023 10

11. Methodology- Model Validation Performance and Evaluation Independent Dataset The developed model will undergo rigorous validation using an independent dataset to ensure unbiased performance evaluation. Evaluation Metrics The effectiveness of the model will be quantitatively assessed based on accuracy, precision, and recall, providing a comprehensive understanding of its classification capabilities. Thursday, November 30 , 2023 11

12. Research Contribution Image processing techniques enhance efficiency by automating the mineral stone classification process, resulting in reduced time and effort for manual inspection. This will enable the Ministry of Mines and Petroleum of Afghanistan to process a larger quantity of mineral stone samples more quickly, thereby facilitating faster exploration and decision-making . Image processing techniques facilitate the classification of a wide range of mineral stones found in Afghanistan. This comprehensive approach enables the Ministry to have a holistic understanding of the country's mineral resources, supporting effective resource management and strategic planning . It empowers the Ministry to make evidence-based decisions, optimize mining operations, and formulate long-term policies for sustainable development . Overall, the adoption of image processing techniques for mineral stone classification in the Ministry of Mines and Petroleum, Afghanistan, offers advantages in terms of efficiency, accuracy, resource allocation, comprehensive management, and data-driven decision making, leading to improved outcomes in the mining sector . Thursday, November 30 , 2023 12

13. https:// en.wikipedia.org/wiki/Mining_in_Afghanistan [1] M . Tropea , G. Fedele, R. De Luca, D. Miriello , "Automatic Stones Classification through a CNN-Based Approach," Sensors, 2022. [2] F . Akkoyun , O. Ekin , O. Sbetici , "Industrial White Quartz Stone Classification Using Image Processing and Supervised Learning," El- Cezerî Journal of Science and Engineering, vol. 9, pp. 801-813, 2022 . [3] M. López , et al, "Functional classification of ornamental stone using machine learning techniques," Journal of Computational and Applied, vol. 234, p. 1338–1345, 2010. [4] References Thursday, November 30 , 2023 13

14. Thank you! Your questions are valuable and can lead to deeper discussions and new insights! Thursday, November 30, 2023 14