Land Type Classification Of SAR Images Based On Complex Domain Neural Networ

Synthetic Aperture Radar(SAR)is an active remote sensing imaging system using microwave sensing.It is widely used in resource exploration,disaster detection,urban coverage survey,military target identification and other fields because of its all-day,all-weather working and high-resolution imaging capability.With the development of SAR system,a large amount of SAR data is generated,which makes the research of SAR image interpretation technology become particularly important.Among them,SAR image classification has always been one of the research hotspots in the field of SAR image interpretation.In recent years,with the rapid development of deep learning technology,many deep learning methods have been widely used in the field of SAR image classification and achieved good results.In this paper,the main work is to apply the deep learning encoder-decoder architecture network model to SAR image ground classification,and put forward some improvement methods based on SAR image characteristics.The main contents of this paper are as follows:1.Analyzed the amplitude and phase statistical characteristics and time-frequency statistical characteristics of SAR images of typical ground objects such as mountains,water,urban buildings and farmland.The results show that the amplitude distribution curves of different ground objects are obviously different.In addition,by analyzing the difference of time-frequency characteristics extracted from different ground objects based on the amplitude and phase information,the important role of SAR image amplitude and phase information for ground objects classification is proved from another Angle.This provides a theoretical basis for us to use complex domain neural network to classify SAR images.2.In order to effectively use the amplitude and phase information of SAR images,we propose a classification algorithm of SAR images based on complex domain neural network.Compared with the corresponding real domain neural network,the results indicate that the complex domain neural network can significantly improve the classification effect of SAR images.3.We design a dual-channel multi-scale complex domain neural network model based on attention mechanism for the classification of ground objects in dual-polarized SAR data.The experimental results show that the network model can achieve a high accuracy classification effect of SAR images by introducing dual-channel encoder,multi-scale feature fusion module,complex domain attention mechanism and other structures.