Pol-Sar Image Classification Using Machine and Deep Learning –A Review

Abstract

In image processing for remote sensing applications, image classification using polarimetric synthetic aperture radar (Pol-SAR) is becoming more relevant. The process of image classification is Data acquisition, Data preprocessing, Feature extraction, Feature selection, and classification. During the process of feature extraction, the shape and orientation of the scatterers have a significant impact on the backscattering of the single-polarized Pol-SAR picture. Thus, the intricate spatial structural patterns and the labyrinthine performed in which a high-resolution Pol-SAR picture is characterized by high dimensions and nonlinearity .The uneven distribution of data is a critical issue as  the sample point's proximal is altered which results in significant reconstruction error. Thus, identifying intrinsic features for target recognition is difficult in feature selection After the feature selection, the classification process can be done for classifying the images, in which the lack of discriminative features and the appearance of speckles occurred due to complex orientation of structural patterns which makes a pixel-level Pol-SAR image classification difficult for obtaining more precise and coherent interpretation consequences. Considering the complexity of the situation, a novel technique is essential for the effectual classification of Pol-SAR images.