Safe operation of the conveyor belt is a necessary condition for the safe and efficient transport of coal production. In the process of transferring large pieces of coal, Due to the fast operation of conveyor belts in underground coal mines and the height difference at the interface, the impact force of large pieces of coal is large, which will cause damage to the conveyor belt. Aiming at the above problems, proposed a novel approach for real-time surveillance of the conveyor belt operation in large coal management. This method leverages the YOLOv5 algorithm to achieve efficient monitoring, which firstly preprocesses the data through adaptive histogram equalization, and secondly, introduces a lightweight neural network to reduce the volume and complexity of the model. Finally, the fusion of the attention mechanism improves the detection performance of the model, so as to realize the real-time monitoring of bulk coal. The experimental results show that the number of model parameters and floating-point operations of the improved algorithm are reduced by 62.8% and 86.4%, respectively, and the detection speed reaches 96.77fps, which has an obvious lightweight advantage, while the detection precision is 60.48%. Therefore, the real-time monitoring method of mining conveyor belt based on lightweight neural network and attention mechanism can better cope with the dim and complex environment of underground coal mine.
Due to the high channel dimensionality caused by the large number of antennas, the beamspace channel estimation is challenging in millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) uplink systems. To enhance the channel estimation accuracy, a novel beamspace channel estimation method is presented by employing the benefit from the compressed sensing (CS). Specifically, the problem of beamspace channel estimation is firstly transformed into an underdetermined optimization problem based on the sparse characteristic of the beamspace channel. Then the approximate message passing (AMP) based on neural network is presented, which is called AMP-Net. When the handcrafted optimal transform is orthogonal in the AMP algorithm, a balanced convolutional neural network (CNN) is designed to select the beamspace channel estimation results under the guidance of the AMP algorithm. Finally, the iterative process is expanded into a deep network form. In the network, controllable parameters are introduced to increase the flexibility of the beamspace channel estimation process. Simulation results demonstrate that the proposed algorithm has better the estimation accuracy than conventional CS based algorithms and other network models.
In the practical application of traffic target detection, it is difficult to ensure the detection accuracy of multi-scale traffic signs in real-time detection because the scale of the target varies greatly. In this paper, an improved method based on YOLOv5s is proposed to address the accuracy problem of multi-scale targets in real-time traffic detection by adapting its data enhancement at the input and localisation loss at the output respectively. The original data enhancement method is improved to solve the problem of target loss in a small range to a certain extent. In addition, an α-FEIOU algorithm is proposed to replace the original CIoU algorithm to improve the accuracy of real-time target detection. The results show that the proposed method shows good performance in different scenarios and its effectiveness is confirmed. At the same time, it has strong robustness and provides a certain reference for the application of deep learning in the field of traffic target detection.