Automatic Change Detection Method Of Island Based On Deep Learning

China has a vast sea area.As an important part of the ocean,islands are of great significance to China's marine ecology,marine economy,and homeland security However,at present,China lacks the exploration of island change detection methods and technical routes.The island change detection work always has a long working period and a large delay,and the existing methods have a large workload and low accuracy.Therefore,it is imperative to carry out research on automatic detection of island changesThis paper aims at the detection of changes at the semantic level within the island,and builds an end-to-end remote sensing image change detection network structure.Combining the environmental characteristics of the island,the surface coverage characteristics of the island,and the current status of island change detection,the investigation can meet the actual needs of the current island monitoring and monitoring work.The automatic detection method of island changes in the situation finally completes the comparison experiment with other change detection methods,and performs change detection and analysis on typical islands in the East China Sea area.The main research contents of this article are as follows(1)Build an end-to-end UNet++change detection network,combine the deep supervision of multi-scale feature fusion,build a mixed loss function combining binary cross entropy and Dice coefficient,and improve the convolution block based on the residual structure and batch normalization operation structure.Design the activation function and parameter initialization method,and use the hot restart cosine annealing method to dynamically adjust the learning rate to optimize the model training process(2)Explore the automatic detection methods and key technologies of island changes.Analyze the current research background of island change detection,combine with the monitoring and monitoring of the four basic elements of islands,explore the key technologies in each link,and form an automatic detection method of island changes.Combined with multi-scale super-resolution models based on meta-learning,SIFT operator registration and other methods,a set of automatic detection methods for island changes that can meet high-resolution remote sensing data based on different sensors,different resolutions and different climatic conditions is constructed(3)Take the East China Sea island as an example to verify and analyze the change detection experiment.Through comparative experiments with IR-MAD,FC-EF and other change detection methods,qualitative and quantitative analysis is carried out to verify the effectiveness and accuracy of the change detection model proposed in this paper.Finally,the typical islands in the East China Sea were tested for changes and the results were analyzedThe end-to-end UNet++change detection model proposed in this paper improves the change detection accuracy and detection effect,reduces the impact of radiation differences on the change detection results,realize the change detection of ground objects of different sizes,and effectively covers the islands Feature categories at different scales in the scene;the change detection method for the island scene solves the difficulties in the current island monitoring and monitoring tasks,and can combine the remote sensing change detection model with the current data foundation and business status.At the same time,the end-to-end method enables the detection task to be expanded from image pairs to image sequences,improving the accuracy and time efficiency of the island change detection task,significantly reducing the time,workload and cycle of the change detection work,which can be monitored with the current island The results of the monitoring work are mutually converted.