The rapid and accurate detection and identification of coal gangue is one of the premises and key technologies of the intelligent separation of coal gangue, which is of considerable importance for the separation of coal gangue. Focusing on the problems in the current deep learning algorithms for the detection and recognition of coal gangue, such as large model memory and slow detection speed, a rapid detection method for lightweight coal gangue is proposed. YOLOv3 is taken as the basic structure and improved. The MobileNetv2 lightweight feature extraction network is selected to replace Darknet53 as the main network of the detection algorithm to improve the detection speed. Spatial pyramid pooling (SPP) is added after the backbone network to convert different feature maps into fixed feature maps in order to improve the positioning accuracy and detection capability of the algorithm, thereby obtaining the lightweight network MS-YOLOV3. The experimental equipment was set up and multi-condition coal and gangue datasets were constructed. The model was trained and the identification and positioning results of the model were tested under different sizes, illumination intensities and various working conditions, and compared with other algorithms. Experimental results show that the proposed algorithm can detect the coal gangue quickly and accurately, with an mAP of 99.08%, a speed of 139 fps and a memory occupation of only 9.2 M. In addition, the algorithm can effectively detect mutually stacking coal and gangue of different quantities and sizes under different lights with high confidence and with a certain degree of environmental robustness and practicability. Compared with the YOLOv3, the performance of the proposed algorithm is significantly improved. Under the premise that the accuracy is unchanged, the FPS increases by 127.9% and the memory decreases by 96.2%. Therefore, the MS-YOLOv3 algorithm has the advantages of small memory, high accuracy and fast speed, which can provide online technical support for the detection and identification of coal and gangue.

The development of surveillance video vehicle detection technology in modern intelligent transportation systems is closely related to the operation and safety of highways and urban road systems. Yet, the current object detection network structure is complex, requiring a large number of parameters and calculations, so this paper proposes a lightweight network based on YOLOv5. It can be easily deployed on video surveillance equipment even with limited performance, while ensuring real-time and accurate vehicle detection. Modified MobileNetV2 is used as the backbone feature extraction network of YOLOv5, and DSC “depthwise separable convolution” is used to replace the standard convolution in the bottleneck layer structure. The lightweight YOLOv5 is evaluated in the UA-DETRAC and BDD100k datasets. Experimental results show that this method reduces the number of parameters by 95% as compared with the original YOLOv5s and achieves a good tradeoff between precision and speed.