基于深度学习的监控建筑变化影像识别

Recognition of Monitoring Building Change Image Based on Deep Learning

传统方法对长焦距摄像头影像进行变化区域提取时,由于光照、摄像头抖动等影响,导致像素点不能精确配准,变化检测不能准确识别建筑物变化的问题,本文提出基于深度学习的监控建筑影像变化检测算法。首先利用图像相似性进行筛选,粗略提取变化区域图像;再利用Faster R-CNN网络对变化区域图像进行建筑物识别与提取。通过桂林西站图像采集试验,结果表明本文方法相比差值法提取变化区域进行变化检测,虚检率降低0.126,漏检率降低0.518,正确率提高0.124,完整率提高0.519,质量提高0.12,在城乡结合部由于建筑物与背景区别更大,具有更好的检测结果和泛化能力。

In order to solve the problem that the pixels can not be accurately registered and the change detection can’t accurately identify the building changes due to the environmental effects,such as illumination and camera jitter when the tradition method extracts the change area of the long focal length camera image,a monitoring building image change detection algorithm based on deep learning is proposed.Firstly,the image similarity is used to filter and roughly extract the change area image.Then,Faster R-CNN network is used to recognize and extract buildings from the changed area image.The experimental results using the images collected from Guilin West Station show that compared with the difference methods to extract the change area for change detection,the proposed method reduces the false detection rate reduce by 0.126,and the omission rate cut down 0.518,and the accuracy rate increase of 0.124,and the integrity rate improved by 0.519 and the quality by 0.12.The proposed method has better detection results and generalization ability in complex urban and rural environments.