Semi-supervised polarimetric SAR ground object classification method based on characteristic mixup

Abstract

The invention discloses a semi-supervised polarimetric SAR terrain classification method based on characteristic mixup. The method comprises the following steps: preparing a data set; making a data set; preprocessing the data; designing a semi-supervised network structure; extracting double-branch features; carrying out feature mixup fusion; predicting a classification result by multiple branches;carrying out network training; predicting a ground object category; and evaluating the network performance. According to the method, the problems caused by mixup and single-branch prediction on the input in the prior art are effectively solved, the performance of the network is greatly improved, and the robustness of the network is enhanced. The method can be used for land cover type discrimination, and can also be used as an intermediate link of target detection, geological exploration and vegetation type discrimination.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a feature mixup-based semi-supervised polarimetric SAR feature classification method. Background technique [0002] As an important research content of polarimetric SAR image understanding and interpretation, polarization SAR image classification has attracted more and more researchers' attention in recent years. The classification of polarimetric SAR images is to classify the same or similar pixels in the image into the same category, and divide the pixels with different characteristics into different categories according to the polarization characteristics and other information. In this paper we propose a semi-supervised network model structure for polarimetric SAR object classification, which achieves state-of-the-art performance on publicly available datasets. [0003] With the development of deep learning, the application of deep learning is becoming mo...

AI Technical Summary

Problems solved by technology

[0004]The training time of the existing method is too long, because the training process of such a network model is to train layer by layer, and the efficiency is relatively low. If there are many network layers, the time is complicated higher degree

Images