MCU-CNN Based High-resolution Image Building Change Detection

Using remote sensing images for change detection is one of the important means of earth observation.Especially with the continuous improvement of image resolution,the spatial details of features are more enriched,and the types of ground objects that can be interpreted are more abundant,which promotes the development of high-resolution image change detection application.Change detection is to detect whether ground objects have changed in the same geographical position.The data source is multi-temporal images.Compared with the application of ground object classification,processing data features is more complicated.In the quality of data processing,the traditional classification and change detection method can not fully mine the change characteristics.In the processing process,the error generated by classification will be transmitted to the change detection results,and the accuracy cannot meet the requirements of practical application.Convolutional neural network has mature applications in the field of computer vision.In this paper,combining the characteristics of high-resolution images and the development of existing technologies,the structure of convolutional neural network and its application in remote sensing are deeply studied,and a change detection technology framework based on convolutional neural network is proposed.The buildings that can clearly reflect the urban structure are selected as representative research objects to verify the effectiveness of the MCU-CNN based high resolution image building change detection model.Specifically,based on the object-oriented analysis technology,the minimum change unit is established for the nested segmentation of two images,and the convolutional neural network is used as the feature extractor to extract and predict the building change features and obtain the building change information.The MCU-CNN based on high-resolution image building change detection has the following features:(1)The minimum change unit is constructed,which not only effectively avoids the salt and pepper phenomenon easily generated by taking pixels as the basic processing unit,but also can describe the change primitives in a fine manner to improve the change detection accuracy.(2)Convolutional neural network is used to mine image features and fully extract the underlying features of changes between image pairs.(3)Creating samples with building change attribute,enrich the attribute information of the building change detection result,and reflect the change process of the feature attribute.(4)Direct change detection method is adopted to avoid error transmission in the detection method of first classification and then change.At the same time,the traditional classification followed by change detection model and multichannel convolutional neural network based change detection model were used as comparative experiments.Design transfer learning experiment to further test the model.These methods all proved the feasibility of the model proposed in this paper and the model had certain application potential.Through the comparison of experimental results and precision,it can be seen that the building change detection based on MCU-CNN high-resolution image proposed in this paper has significant advantages.In addition,the combination of deep learning technology and Object-Based Image Analysis is a new research direction in the current change detection field,which will promote the development of change detection in the future.