| High spatial resolution remote sensing images contain rich feature details and cover a wide range of areas.It has become an important data source for geographical national conditions monitoring,military reconnaissance,crop growth,urban construction and other fields.Image segmentation,as the premise and basis for highresolution remote sensing image information interpretation and Application,is a key step and part of the transition from data to information.However,due to the increase of spatial resolution,the difficulty of extracting feature information in images has also increased.The existing image segmentation method is difficult to extract a complete region with semantics and cannot meet the application requirements of the current highresolution remote sensing image.This paper focuses on the shortcomings of the existing research work and deeply studies the high-resolution remote sensing image segmentation method.This paper presents a remote sensing image segmentation method based on regional adjacency relationship.Firstly,the watershed segmentation algorithm based on morphology gradient is used to segment the image with smaller gradient thresholds,and over segmentation results are obtained.Then these over-segmentation results are established according to the spatial distribution relationship of neighboring regions,and a new regional adjacency topology is established.Based on the principle of maximum similarity,the regions are merged in the data network structure and iterative until convergence.Aiming at the problem of low efficiency of the traditional regional adjacency graph method,this paper designs adjacency area data structure to maintain the adjacency relationship between the areas.This method maximizes the computational efficiency while ensuring high segmentation accuracy.By comparing this method with the Mean-Shift segmentation method,the advantage of the method in segmentation accuracy is proved.Remote sensing image segmentation method based on regional adjacency relationship utilizes the spectral characteristics and spatial distribution characteristics of the bottom layer of the object area.It can represent the attribute information of the object area to some extent,but due to the lack of constraints of high-level semantic information,the segmentation accuracy is still difficult to meet the actual application requirements.Therefore,on the basis of remote sensing image segmentation based on regional adjacency relationship,this paper proposes a high-resolution remote sensing image segmentation method that combines regional adjacency relations and entity object feature models.This method establishes a multi-class entity object feature model to semantically describe features.According to the merge rule of the model,the regions are merged in the network structure of the neighboring regions,and finally a complete region with actual semantic information is obtained.This paper compares the method with the Mean-Shift segmentation method and the region adjacency method that does not include the entity object feature model.Using visual interpretation and segmentation accuracy evaluation indicators to evaluate the experimental results,the superiority of the method in segmentation accuracy was verified. |
