Remote Sensing Building Detection Based On Deep Learning

The detection of ground objects in remote sensing images has always been a research hotspot.Among them,buildings are the main place for human activities,so it is very important to realize the detection of such objects.The traditional target detection algorithm that extracts the features of simple objects according to the color,texture and shape of buildings is no longer applicable.The target detection algorithm based on deep learning has a strong ability to adapt to the environment with high background complexity and can extract more detailed features.It is suitable for remote sensing image scenes and has a good performance in robustness.This paper mainly studies remote sensing building detection based on deep learning.Target detection method based on region Mask RCNN and target detection method based on end-to-end YOLO V3 are adopted to study the unsatisfactory segmentation effect of irregular buildings and low detection accuracy of small buildings respectively.The main work and achievements of this paper are as follows:(1)Aiming at the problem that the traditional rectangular recognition frame could not meet the detection requirements due to the irregular shape of the building itself,an improved Mask RCNN target detection algorithm was proposed,which was improved for the main network structure FPN and RPN,and adjust the mask parameters.The experimental results show that the proposed algorithm improves the detection accuracy by 1.54% and 1.65%,respectively,compared with the original Mask RCNN algorithm,which effectively optimises the detection of irregular buildings in the remote sensing field.(2)Aiming at the difficulty of detecting small buildings and the inability to meet the real-time requirements in the detection process,an improved YOLO V3 target detection algorithm was proposed by modifying the resolution of feature map in Darknet-53 network and adjusting the dimension of prior box.The experimental results show that the proposed algorithm improves the detection accuracy by 5.35% and 2.34%,respectively,compared with the original YOLO V3 algorithm,which effectively solves the problem that small buildings in remote sensing field are difficult to detect.