Research On Potential Region Extraction Method Of Multi-scale Target In SAR Image

Because the target is closely related to the environmental scene,quickly obtaining the potential area of the target in the SAR image based on the prior information of the SAR image scene is the first step to improve the efficiency of target detection,reduce false alarms,and achieve accurate target detection.High-resolution SAR image features have obvious fluctuations and rich texture features.The calculation granularity of the pixel-level processing method is too fine,and the spatial organization relationship between pixels is not fully considered,so it is difficult to effectively obtain the multiscale target potential area of the SAR image.This article focuses on SAR images,mainly to solve the problems of low accuracy of multi-scale object classification and poor extraction of potential target regions.In this paper,the classification of SAR images is carried out based on the prior spatial distribution characteristics of ground objects and multi-scale segmentation.The pros and cons of SAR image data sources,the division of the hierarchical structure of the ground features and the selection of scales all have an important impact on the classification results.Aiming at multi-scale targets in SAR images,this paper mainly studies the following three aspects:1)In the same image scale,the method of data fusion is used to improve the classification accuracy.Since the information obtained by a single polarization channel is limited,consider using an appropriate data fusion method to effectively use more ground feature information.This paper proposes an adaptive network fusion method to make decision fusion on the data of VV channel and VH channel,which improves the classification accuracy.2)For multi-scale targets,perform hierarchical classification based on image blocks to achieve rapid extraction of target potential regions.According to the principle of regional consistency,the adaptive image block is divided,and the optimal segmentation scale of the feature is obtained through experimental analysis,and combined with the feature of the image feature and prior knowledge,a hierarchical structure of classification is established.The multi-scale hierarchical classification method proposed in this paper is compared with the unsupervised K-means clustering segmentation method and superpixel segmentation method.The method in this paper is more effective.3)On the basis of the proposed hierarchical classification,obtain more refined target extraction results.Because the shape of the river is quite different from other features,and there are other features around the river,directly classifying them together with other features will reduce the overall classification effect.Based on the results of land classification,this paper uses an improved region growing algorithm to extract rivers.In addition,the results of hierarchical classification can also be used for tasks such as target detection.In this paper,based on the extraction of the ocean area,the ship target detection is carried out.Compared with the direct ship detection,the false alarm is reduced.extraction results.Because the shape of the river is quite different from other features,and there are other features around the river,directly classifying them together with other features will reduce the overall classification effect.Based on the results of land classification,this paper uses an improved region growing algorithm to extract rivers.In addition,the results of hierarchical classification can also be used for tasks such as target detection.In this paper,based on the extraction of the ocean area,the ship target detection is performed to avoid detecting false alarm targets on the land.