Deep-level feature learning and watershed-based synthetic aperture radar (SAR) image classification method

Abstract

The invention discloses a deep-level feature learning and watershed-based synthetic aperture radar (SAR) image classification method and belongs to the image processing technical field. The main objective of the invention is to solve the problems of high possibility of wrong classification of fields, poor region consistency and boundary burrs when middle-and-lower-layer features are applied to SAR image classification. The classification method includes the following steps that: watershed over-segmentation class label L calculation is performed on an inputted SAR image; Gabor features F1 of the inputted SAR image are calculated; after sampling is performed on the F1, sampled F1 are inputted into a K-mean singular value decomposition (KSVD) algorithm, so that a training dictionary D can be obtained; convolution and maximum value pooling are performed on the F1 and the D, so that convolution features F2 can be obtained; the F2 are inputted into a sparse auto-encoder, so that deep-level features F3 can be obtained; the F3 are inputted into a SVM so as to be classified, and classification results R1 can be obtained; and vote statistics is performed on the R1 at each sub block of watershed segmentation results, so that final classification results can be obtained. The deep-level feature learning and watershed-based SAR image classification method of the invention has the advantages of high computation speed, accurate edge classification and high region consistency, and can be applied to SAR target recognition.

Description

technical field [0001] The invention belongs to the field of image processing, in particular to a method related to SAR image classification, which can be applied to target recognition. Background technique [0002] Synthetic Aperture Radar (SAR) is a high-resolution microwave imaging radar for all-weather and all-weather detection and reconnaissance of the earth. SAR can effectively identify camouflage and penetrate cover, so it has been widely used in military and civilian fields such as remote sensing mapping, military reconnaissance, and earthquake relief. SAR images have the characteristics of rich texture information and strong multiplicative coherent speckle noise. In the imaging process of SAR images, different ground features have different back reflection and scattering characteristics, thus presenting different textures, so SAR images often contain rich texture information. On the other hand, since SAR adopts a coherent imaging system, the imaged SAR image is in...

AI Technical Summary

Problems solved by technology

[0008] The present invention aims to solve the defects of the existing SAR image classification technology for mining middle-level features, and proposes a SAR image classification method based on deep learning and watershed

Images