Small sample SAR image ship classification method based on transfer learning

Abstract

The invention relates to a small sample SAR image ship classification method based on transfer learning, and belongs to the technical field of computer vision. The method comprises the following stepsof 1) preprocessing the SAR image ship slices so as to enable the ship slices to meet the requirements of the transfer learning for the input pictures, performing image enhancement on the ship slices, and synthesizing the similar ship slice images through a DCGAN network so as to meet the requirements of a CNN classification network for the data quantity; 2) extracting the image features of the images through a denoising auto-encoder, reducing the noise added when the DCGAN generates the images, and reducing the influence of different sea condition backgrounds on the classification results; and 3) carrying out transfer learning by adopting a ResNet network, and further improving the classification accuracy by adopting a fine-tune method. The small-sample SAR image ship classification method ensures that the small-sample SAR image ship classification can reach a certain accuracy.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and relates to a small-sample SAR image ship classification method based on transfer learning. Background technique [0002] Spaceborne SAR (Synthetic Aperture Radar) is an active microwave sensor that uses the principle of synthetic aperture. Through spaceborne SAR, the functions of monitoring, detection, and classification of ships can be realized. With the development of spaceborne SAR, SAR images have richer details, and the effects of ship detection and classification by naked eye observation or traditional feature detection cannot meet the requirements. [0003] Based on classification requirements, the most commonly used classification method is to use deep learning models to classify SAR images, but this classification method requires a large number of samples with true value labels. For example, the patent application "Image classification method based on cross-domain transfer l...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the object of the present invention is to provide a small-sample SAR image ship classification method based on transfer learning, which solves the problem of classifying SAR image ships when it is difficult to obtain SAR images and there are few marked SAR images; Applied to the classification of ship targets in SAR images, so as to meet the needs of marine surveillance such as coastal ship surveillance and maritime rescue

Images