High Resolution Remote Sensing Change Detection Of New Urban Construction Land Based On Deep Learning

Newly-added urban construction land is an important indicator of urban construction and development,and its accurate extraction can assist urban planners and managers in dynamic monitoring and further planning,so as to grasp urban construction and development trends.With the rapid development of satellite remote sensing technology,high-resolution remote sensing images are widely used in various industries of national economic construction.At present,there are many methods for change detection using remote sensing images,but there are disadvantages such as high image preprocessing requirements,low automation,and difficulty in feature selection and feature combination.Deep learning algorithms have strong learning capabilities.They can fit complex mapping relationships through multi-layer convolution,solve complex tasks,and provide a new method for remote sensing image processing.Therefore,this paper uses high-resolution remote sensing images as data sources to carry out deep learning-based research on high-scoring remote sensing change detection of urban new construction land.The paper uses U-Net as the backbone network,and improves and optimizes the U-Net network structure and optimization through experiments.Loss function,designed a set of remote sensing change detection process based on deep learning for high-scoring urban newly-added construction land.The main research content and results of the article are as follows:(1)A high-resolution image data set of newly-added construction land in the city has been established.Currently,there is no open new construction land data set at home and abroad.This article first defines the types of new construction land in the city and the status quo of changes.Taking the tea garden in Chongqing as an example,a high-resolution image data set of new construction land in the city is established.The main process of data set establishment includes: making a raster map of newly added construction land in the city,manual labeling,data cutting and division,data merging,data augmentation,image value normalization,etc.(2)Optimizing the best loss function through experimental comparison.In order to reduce the impact of data set category imbalance on the accuracy of change detection,this article has carried out research on the loss function of the model,and selected three loss functions for category imbalance.Based on the U-Net model,it is guaranteed that other parameters are not Under the condition of change,train separately to obtain the optimal loss function.The experimental results show that the weighted cross-entropy loss function has the highest training accuracy among the three functions,and the graphics extraction effect of the newly added region is the best.(3)Designed and implemented a new method for detecting changes in urban newly-added construction land.In order to improve the fineness of the segmentation results,and to avoid the model degradation problem caused by the deepening of the network layer,the residual Res Ne Xt structure is introduced to replace the coding layer in the U-Net model,and the U-Net model is structurally improved.Improve.Designed and constructed an improved U-Net change detection model based on residuals,and designed a comparative experiment.The experimental results show that the improved model has higher extraction accuracy and the extracted image contour is closer to the real label map.The article carried out research on the change detection problem of urban newly-added construction land,realized the improvement and optimization of the U-Net-based change detection method,and designed and implemented a complete set of urban newly-added construction land change detection technical process,which can be It is used to quickly and extensively extract the area of newly-added construction land in the city,facilitate the real-time monitoring of land use,effectively combat various illegal land use activities,and realize the rational management of urban construction land and the effective use of land resources.