Table './tdc/tbl_dc_meta_control_57' is marked as crashed and last (automatic?) repair failed
Tanjira Sirisut. Development of segmentation algorithm using color scale image analysis for aggregates. Master's Degree(Mining and Georesources Engineering). Chiang Mai University. Library. : Chiang Mai University, 2025.
Development of segmentation algorithm using color scale image analysis for aggregates
Abstract:
Currently, there are numerous innovations that utilize images analysis for various types of analysis. In the mining industry, image analysis-based innovations have been adopted to assist in analyzing the size of rocks, which are then used to determine particle size distribution. Blasting rock piles and crushing plants commonly apply this approach. Analyzing the particle size distribution(PSD) can contribute to improving the efficiency of the blasting. The Department of Mining and Petroleum Engineering, Faculty of Engineering at Chiang Mai University, has developed an Automated Aggregate Size Distribution Analyzer (AASDA). This system was developed to analyze images in order to identify rocks that oversize of the primary crusher. The original AASDA system employed image segmentation techniques based on grayscale images, achieving approximately 80-90% accuracy in size analysis. This study, the researchers proposed enhancing the system by employing color images as the basis for image segmentation analysis, resulting in the development of the Automated Aggregate Size Distribution Analyzer-Color (AASDA-C). Experiments were conducted to identify optimal environmental conditions and factors affecting color-based image segmentation performance. These factors included material pile arrangements, light intensity, direction of lighting, camera angle, and opacity levels. Four limestone size ranges2135 mm, 1323 mm, 1018 mm, and 1149 mmwere used in the experiments. The experimental results showed that the AASDA-C system could achieve measurement accuracy in the laboratory environment ranging from 78.19% to 96.11%. However, significant variation in counting accuracy was observed when environmental conditions changed. Optimal operational conditions were identified with a light intensity range of 250300 lux, light angles 90 degrees, camera angles between 6590 degrees, and material opacity levels not exceeding 10.5%. These conditions minimized segmentation errors arising from color similarity between opaque materials and the background, clearly highlighting limitations of the K-means clustering method, which is the primary segmentation technique used in both AASDA and AASDA-C systems. Further field tests were conducted by photographing limestone samples alongside a colored reference ball with a known diameter of 22.86 cm. The field experiments showed an average size measurement accuracy exceeding 80%. Additionally, statistical agreement between manual measurements and automated AASDA-C measurements was evaluated using Kappas Coefficient, resulting in a coefficient value of -0.05 , indicating pool agreement. Because AASDA-C is unable to analyze every rock particle captured in the sample limestone images from the crushing plant. Nevertheless, the accuracy in counting aggregate particles according to measured sizes remains significantly affected by environmental factors and obstruction due to overlapping aggregates. This results in considerable variance in the system 's counting performance due to overlapping and similar color in segmentation. Thus, further studies and development of advanced segmentation and analytical techniques are necessary to overcome these limitations in the future.