Change Detection Of Building In Large-Span Area Along High-Speed Railway Based On Convolutional Neural Network

Nowadays,high-speed railways have become closely related to the lives of the Chinese people.They not only provide more convenient means of transportation for the masses,but also provide strong support and guarantee in the process of promoting my country’s economic construction and promoting comprehensive modernization reforms.Buildings are the main carrier of human activities,and regular monitoring of their changes can effectively maintain the safety of high-speed rail operations and understand the impact of the high-speed rail economic belt on surrounding areas.The traditional method of building change detection mainly adopts manual field investigation,which is time-consuming and labor-intensive,and has a small coverage,so it is difficult to meet the application requirements in large-span areas.With the development of high-resolution remote sensing technology,it provides a good data foundation for the extraction of building change information along the high-speed railway.At the same time,the development of remote sensing image change detection technology and the emergence of convolutional neural networks have provided corresponding technical means for the building changes detection along high-speed railways.Based on the above background,this thesis analyzes the deficiencies of existing research,and conducts an in-depth study on the building changes detection along high-speed railways based on convolutional neural networks from the perspectives of whether buildings have changed and the specific types of changes.The content is as follows:(1)Firstly,the dilemma of spatial and temporal heterogeneity brought about by the long span regional characteristics of high-speed railways is analyzed in detail,which is embodied in two aspects: differences in building distribution caused by spatial span and differences in image expression caused by time span.Most of the object change detection datasets are simple in distribution and single in type,which cannot be adapted to the complex scenes along high-speed railways.On this basis,this thesis takes the Zheng-Xi high-speed railway as an example to establish a dataset for building change detection along the high-speed railway.The dataset has rich sample diversity and high complexity of change types,which can effectively support subsequent research.(2)A high-precision building change area extraction method that integrates image-level and semantic-level context information is proposed.According to the data characteristics along the high-speed railway,the semantic-level context and image-level context information aggregation module are respectively designed around the differences of building objects and image expressions.These modules can significantly enhance the global and category features of images,reduce the interference of complex background information,and improve the accuracy of model detection.The experimental results show that the method in this thesis can extract the building change area more accurately,and shows better performance compared with other mainstream models.(3)A method of building change direction detection is proposed.This method makes full use of the structural characteristics of the siamese network and the data characteristics of the class activation mapping.Based on the results of the building change area detection,by calculating the weight distribution of the feature map of the bitemporal remote sensing image,combined with the auxiliary discrimination of the remote sensing feature index,finally,the change direction of the building and the specific change category are obtained.The experimental results show that the detection network based on siamese class activation mapping can effectively identify the spatiotemporal process of building changes,so that the information contained in the building change detection result is more abundant and useful.