Peng Liu, Changlin Song, Junmin Li, Simon X. Yang, Xingyu Chen, Chuanfu Liu, and Qiang Fu
YOLOv3, transmission line, engineering vehicle, deep learning, neural networks
With the continuous acceleration of the infrastructure construction process, the emergence of a large number of engineering vehicles poses a great threat to the power transmission lines, and the detection of engineering vehicles around transmission lines has become one of the essential measures to guarantee the safety of transmission lines. Therefore, an engineering vehicle detection model around the transmission lines is proposed based on the improved YOLOv3. First of all, the training dataset is enhanced through data augmentation, and the number of images between classes is more balanced. Next, the elbow method is used to determine the number of anchors, and the k-means++ algorithm is used to cluster the dataset to determine the size of anchors. Then, to strengthen the fusion of features, the high-level features are cascaded to the low-level features, while the low-level features are also cascaded to the high-level features. Finally, the channel attention mechanism and the spatial attention mechanism are cascaded to enhance the information of the features. Experiments show that the improved model is better than the Faster-RCNN, single shot multibox detector (SSD) and YOLOv3. Its mean average precision (mAP) value is improved by nearly 7% compared with YOLOv3. This method can be used well for the detection of engineering vehicles.
Important Links:
Go Back
