Construction Vehicle Detection Method Based on Multi-Scale Residual Network

In the process of urbanization, construction vehicles such as cranes pose a significant threat to the safe operation of overhead transmission lines. It is helpful to reduce the risk of lines by collecting images in the construction area through cameras, monitors, and other equipment, and detecting the construction vehicle targets that may pose a threat in real-time. Aiming at the problems of large scale variation and poor detection accuracy of construction vehicles, a multi-scale residual construction vehicle detection network (MSRCVD-Net) is designed based on the residual network. By introducing a hierarchical residual connection structure into the residual blocks in the feature extraction network, the model can express coded multi-scale features at a more refined level, enhance the receptive field range of each network layer, and improve the feature extraction capability. In addition, by adding a receptive field enhancement network after the feature extraction network, the problem of information loss caused by the network in the process of sampling under feature extraction is alleviated, and the detection capability of construction vehicles is further improved. The experiment shows that the MAP value of MSRCVD-Net is 71.8%. The effect of detection is effectively improved.