Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review
Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review
}
TY - GEN
T1 - Real-Time YOLO-family Comparison for Blast Quality Estimation in the Open Pit Conditions
AU - Ronkin, Mikhail
AU - Reshetnikov, Kirill
N1 - This research was supported by the Russian Science Foundation and Government of Sverdlovsk region, Joint Grant No 22-21-20051, https://rscf.ru/en/project/22-21-20051/.
PY - 2023/5/15
Y1 - 2023/5/15
N2 - Conducting blasting in an open pit environment is an essential step in the technology of rock processing. The subsequent stages of transporting and processing the resulting rock fragments critically depend on the quality of the blast. As a result, the control of blast quality in real time is necessary during blasting in the open pit. This paper compares the quality of detection and segmentation of rocks obtained after blasting using the YOLO model family. A series of neural network training experiments were conducted and compared using the metrics precision, recall, average precision, and inference time per image. Based on experimental results, it is shown that the updated YOLOv8X model performs better for both detection and segmentation tasks. For the object detection task, YOLOv8X shows an average precision of 76%. This outperforms the YOLOv7X model by 1% on Average precision and the YOLOv5X model by 5%. In the segmentation task, the YOLOv8X model outperforms the YOLOv7X model by 5% on Average precision, and the YOLOv5X model by 7%.
AB - Conducting blasting in an open pit environment is an essential step in the technology of rock processing. The subsequent stages of transporting and processing the resulting rock fragments critically depend on the quality of the blast. As a result, the control of blast quality in real time is necessary during blasting in the open pit. This paper compares the quality of detection and segmentation of rocks obtained after blasting using the YOLO model family. A series of neural network training experiments were conducted and compared using the metrics precision, recall, average precision, and inference time per image. Based on experimental results, it is shown that the updated YOLOv8X model performs better for both detection and segmentation tasks. For the object detection task, YOLOv8X shows an average precision of 76%. This outperforms the YOLOv7X model by 1% on Average precision and the YOLOv5X model by 5%. In the segmentation task, the YOLOv8X model outperforms the YOLOv7X model by 5% on Average precision, and the YOLOv5X model by 7%.
UR - http://www.scopus.com/inward/record.url?partnerID=8YFLogxK&scp=85164925076
U2 - 10.1109/USBEREIT58508.2023.10158813
DO - 10.1109/USBEREIT58508.2023.10158813
M3 - Conference contribution
SP - 254
EP - 257
BT - Proceedings - 2023 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology, USBEREIT 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2023 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT)
Y2 - 15 May 2023 through 17 May 2023
ER -
ID: 41985443