IBR-Yolov5: Illegal Building Recognition with UAV Image Based on Improved YOLOv5

Abstract

The urban illegal buildings have seriously disrupted the urban land planning and development space, even having serious security risks. How accurately and comprehensively identifying the illegal buildings is very important for their management and planning. The traditional method of identifying illegal buildings is inefficient and high consumption. Learning-based methods can improve efficiency and save resources but a high miss rate. In response to the above problem, this paper proposes an illegal building recognition method based on UAV (Unmanned Aerial Vehicle) images: IBR-Yolov5 (Illegal Building Recognition with UAV Image Based on Improved YOLOv5). IBR-Yolov5 appends SPP (Spatial Pyramid Pooling) based on stochastic pooling to the Backbone module. Meanwhile, CBAM(Convolutional Block Attention Module) is used to highlight the main features and suppress irrelevant features, which can ultimately improve the detection accuracy of IBR-Yolov5. In addition, IBR-Yolov5 combined with BiFPN (Bidirectional Feature Pyramid Network) can fuse features extracted by different layers of networks to reduce feature loss. Experimental results show that the miss detection rate of the proposed improved model is lower in comparison with the original YOLOv5, and detection precision has been improved.