Multiobjective evolutionary algorithm assisted stacked autoencoder for PolSAR image classification

Polarimetric synthetic aperture radar (PolSAR) image classification is a vital application in remote sensing image processing. In recent years, deep learning models like stacked autoencoder and its variants have been utilized to handle this problem and perform well. But their performances highly depend on proper hyper-parameter configuration. In this paper, we propose a multiobjective evolutionary algorithm assisted stacked autoencoder (SAE_MOEA/D) for PolSAR image classification, which could adaptively optimize its parameters and hyperparameters such as weights, activation functions and the balance factor in the loss function of stacked autoencoder, and decide how many layers of the network should be used according to datasets. Its performance has been tested on five PolSAR images. Compared with commonly used methods, our method obtains competitive results and could save manpower.

Automated summary

This study introduces a new approach using a multiobjective evolutionary algorithm assisted stacked autoencoder for PolSAR image classification, which can adaptively optimize parameters and hyperparameters to achieve competitive results.