Building Detection in Remote Sensing Image Based on Improved YOLOV5

Abstract

Due to the different seismic design of high and low-rise buildings, the seismic levels are often different in the same intensity area. Using remote sensing to distinguish various height buildings, and then further make targeted measures to reduce earthquake losses at all seismic levels, which can reduce the secondary disasters of earthquakes. Traditional architectural target detection models use artificially designed target features based on architectural contours, which not only require a lot of time and manpower to process the data, but also cause low detection accuracy due to inconsistent manual standards. The target detection model based on deep learning has high detection accuracy. However, for unmanned aerial vehicle (UAV) remote sensing applications where real-time and accuracy requirements are relatively high, the acquired image information is less and the target features are not obvious, cause target detection accuracy to decrease. In order to improve the accuracy and speed of target detection, a building target detection model with Convolutional Block Attention Module (CBAM) is proposed based on the YOLO V5 model, which improves the target classification accuracy and detection speed of building detection. The model is deployed on an automatic disaster monitoring platform to conduct a house environment target detection experiment. The comprehensive detection accuracy can reach 88.5%, and the detection speed is 26 frames/s, which meet the real-time and accuracy requirements of high-rise and low-rise building target detection.