The ore fragment size on the conveyor belt of concentrators is not only the main index to verify the crushing process, but also affects the production efficiency, operation cost and even production safety of the mine. In order to get the size of ore fragments on the conveyor belt, the image segmentation method is a convenient and fast choice. However, due to the influence of dust, light and uneven color and texture, the traditional ore image segmentation methods are prone to oversegmentation and undersegmentation. In order to solve these problems, this paper proposes an ore image segmentation model called RDU-Net (R: residual connection; DU: DUNet), which combines the residual structure of convolutional neural network with DUNet model, greatly improving the accuracy of image segmentation. RDU-Net can adaptively adjust the receptive field according to the size and shape of different ore fragments, capture the ore edge of different shape and size, and realize the accurate segmentation of ore image. The experimental results show that compared with other U-Net and DUNet, the RDU-Net has significantly improved segmentation accuracy, and has better generalization ability, which can fully meet the requirements of ore fragment size detection in the concentrator.

Detection of the loading volume of mining trucks is an important task in open pit mining. Aiming at the addressing the current problems of low accuracy and high cost of the detection of the loading volume of mining trucks, this paper proposes a mining truck loading volume detection model based on deep learning and image recognition. The training and test data of the model consists of 6000 sets of images taken in a laboratory environment. After image preprocessing, the VGG16 network model is used to pre classify the ore images. The classification results are displayed and the possibility of each category is determined. Then, the loading volume of mining trucks is calculated by using the classification results and the least squares algorithm. By using the labeled image data of five kinds of mining truck loading volume, the arbitrary loading volume detection of mining trucks is realized, which effectively solves the problem of a lack of labeled data types caused by the difficulty in obtaining mine data. Root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the fitting effect of the model. The experimental results show that the model has high prediction accuracy. The average absolute error is 17.85 ${\mathrm{cm}}^3$. In addition, this paper uses 400 real mining truck images of open-pit mines to verify the model and the average absolute error is 2.53 ${\mathrm{m}}^3$. The experimental results show that the model has good generality and can be applied well to the actual production of open-pit mines.

Accurate road information is crucial for the effective planning and management of open‐pit mines. However, the parameters related to the roads in open‐pit mines change constantly, and their shapes are often complicated. These factors indicate the difficulties in updating road information in open‐pit mines. This study reports a method for updating the road information in open‐pit mines using the trajectories of trucks. The method employs an algorithm for compressing trajectory data, an algorithm for extracting new roads, and an algorithm for old‐road classification. Data from the Fushun open‐pit mine is used to validate this method, and our experiments show that it can identify more than 90% of new roads and the majority of disappearing roads. In addition, we analyze several factors affecting the results obtained using this approach and provide a detailed description of the limitations of this method. Overall, our findings indicate that this is a promising approach to road‐information retrieval.

The safety of blasting site in open-pit mine can be greatly improved by risk early warning. Therefore, an intelligent real-time risk early warning method of blasting site in open-pit mine based on deep learning was proposed in this research. The mobile wireless webcams, H.264 video compression algorithm, and real-time transport protocol were applied to achieve real-time video acquisition and transmission of blasting site in open-pit mine. A single-stage deep neural network DG-YOLOv3 was proposed in this research. DG-YOLOv3 is an improvement of Gaussian YOLOv3, among which Darknet41 is used to improve the model’s detection speed and detection accuracy of small targets. To further improve the performance (i.e., speed and accuracy) of risk early warning, the surveillance videos were first split into pictures by frame. Then, the pictures were processed by weighted average grayscale and contrast limited adaptive histogram equalization. Experiments show that the mean average precision of DG-YOLOv3 proposed in this paper reaches 87.45 and the detection speed reaches 56.82 frames per second, which has better accuracy and speed compared with other algorithms. In addition, DG-YOLOv3 has good robustness in complex scenarios. Based on the detection results, the intelligent real-time risk early warning of the blasting site in open-pit mine is achieved finally.

：In order to further improve the truck dispatching theory of open pit mine, an open pit mine truck dispatching equilibrium distribution model based on traffic flow planning was put forward. The distribution model is based on the principle of Wardrop, and the distribution of the road resistance is used as the basis of the distribution. Verified by the case , the road traffic flow of the model in the truck dispatching process is more optimized as the research object than the result of traffic planning , and achieves the purpose of further optimization of truck dispatching process.

Given a point set of servers and a point set of users, where each server has a capacity (i.e., a maximum number of users it can serve), the capacity constrained assignment (CCA) is to assign every user to one server subject to the capacity constraints. To offer the highest quality service, we expect to minimize the average distance between users and their assigned server, which is termed optimal assignment. Since users are mobile, it is necessary to carry out real-time reassignment to ensure the high-quality services. Meanwhile, the distribution of users in road-networks is non-uniform. To improve the efficiency of resource utilization, our task is to continuously select k servers from all servers to provide services for all users, which is termed continuous k-selection capacity constrained assignment (C-ACCA) problem. We propose an incremental algorithm Ä-RAIA that checks the server set based on a Voronoi diagram and uses a heuristic method to solve this problem. The experiments on real datasets demonstrate the efficiency and effectiveness of the proposed algorithm.