Semi-supervised deep learning method based on semi-supervised sparse filtering

Abstract

The invention discloses a semi-supervised deep learning method based on semi-supervised sparse filtering. The method solves a technical problem that a conventional deep learning method is complex in parameter regulation and low in classification precision when having low label data, and comprises steps of: inputting polarized SAR image data to be classified; extracting a training sample and a testing sample; solving the Wishart neighbor sample of the training sample; initializing the parameter of a deep sparse filtering network; pre-training the deep sparse filtering network; fine regulating the deep sparse filtering network; predicting the classification of the testing sample; and outputting the classification image and the classification precision of the polarized SAR image to be classified. The method, by constructing the novel deep sparse filtering network and adding a semi-supervised regular term to a pre-training process, reduces parameter regulation complexity of the deep learning network, increases polarized SAR image terrain classification, and can be used in the technical fields of environment monitoring, earth resource exploration, military systems and the like.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a polarization SAR image ground object classification method, in particular to a polarization SAR classification method based on a semi-supervised deep sparse filter network. It can be used in environmental monitoring, earth resource surveying and military systems, etc. Background technique [0002] Machine learning (Machine Learning, ML) is a multi-field interdisciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. Specializes in the study of how computers simulate or implement human learning behaviors to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their performance. In the field of polarimetric SAR image classification, there have been many breakthroughs in machine learning, such as Wishart maximum likel...

AI Technical Summary

Problems solved by technology

For example, learning rates, momentum, sparsity penalties, etc., and the final determination of these parameters needs to be obtained through cross-validation, which requires a lot of time and effort

With the continuous development of the field of remote sensing, applications such as environmental monitoring, earth resource surveying, and military systems have increased demand for polarimetric SAR image processing. It is desired to achieve ideal results for the classification of polarimetric SAR images, although deep learning is There are obvious advantages in machine learning methods, but the traditional deep learning network requires a large number of parameter adjustments, which will consume a lot of time. Improper parameter selection will directly affect the performance of the deep learning network and the final classification results, which restricts Application of deep learning methods in the field of remote sensing

Images