Optimization of open-pit mining is one of significant tasks to date, with the blasting quality estimation being a key factor. The blasting quality is determined through evaluating the number of fragments and block size distribution, the so-called fragmentation task. Currently, computer vision-based methods using instance or semantic segmentation approaches are most widely applied in the task. However, in practice, such approaches require a lot of computational resources. Because of this, the use of alternative techniques based on algorithms for the real-time object detection is highly relevant. The paper studies the use of YOLO family architectures for solving the task of the blasting quality assessment. Based on the research results, YOLOv7x architecture is proposed as a baseline model. The proposed neural network architecture was trained on a dataset selected by the present authors from digital images of blasted open-pit block fragments, which consisted of 220 images. The obtained results also allow one to suggest the geometrical size of rock chunks as a measure of blasting quality.