Cross-domain adaptive SAR image classification method and device based on simulation data, and equipment

Abstract

The invention relates to a cross-domain adaptive SAR image classification method and device based on simulation data, and equipment. SAR imaging conditions are introduced into the training process of a model, domain confusion measurement between two domains is calculated by utilizing a source domain simulation image and a target domain simulation image according to the imaging conditions, and the model is trained according to the domain confusion measurement so as to reduce the domain difference between the source domain and the target domain; and the model is trained by using the source domain actual measurement image, so that the model has a classification and identification function, and the target classification model obtained by final training can better identify the target in the target domain SAR image different from the source domain imaging condition, and has higher accuracy when being classified.

Description

Technical field [0001] Technical Field The present application relates to SAR image classification, and more particularly to a cross-domain adaptive SAR image classification methods, apparatus, and equipment based on simulation data. Background technique [0002] SAR (Synthetic Aperture Radar), i.e. synthetic aperture radar is an active observation of the system, it can be installed on aircraft, satellites, spacecraft and other flying platform, during the day, the weather observation embodiment, and has a certain the ground penetrating ability. Therefore, SAR systems in disaster monitoring, environmental monitoring, marine monitoring, resource exploration, has a unique advantage in the application of crop yield estimation, mapping and other aspects of the military can play other remote sensing instruments difficult to play a role. [0003] SAR images are very sensitive to the image forming condition. However, deep learning model SAR image automatic target recognition using mostly...

AI Technical Summary

Problems solved by technology

However, most of the deep learning models used in SAR image automatic target recognition are based on the enhanced network structure for processing optical image classification tasks, which is obviously unreasonable, because sometimes there is a huge gap between the data of the training model and the data of the testing model. That is to say, the imaging conditions of the image data used for training are different from the imaging conditions of the image data for testing, which will cause the model to fail to recognize the target in the image well when it is actually used.

Images