Abstract: Due to its capability of weather-independent imaging and sensitivity to target scattering and geometric properties, Polarimetric Synthetic Aperture Radar (PolSAR) offers crucial support for the built-up areas (BA) information analysis. However, labelled BA samples with unique orientations are hard to come by, and PolSAR BA with wide orientation angles is frequently mistaken for vegetation. Additionally, the trained models and labelled BA samples hardly ever perform well in the cross-domain BA analysis of PolSAR data. This article describes a PolSAR BA extraction method that combines subspace alignment with eigenvalue statistical components (ESC) and PU-Learning (PUL) to achieve cross-domain BA extraction (SA). Building orientation effects and the roll invariance of the eigenvalues of the coherency matrix are examined first. Regional statistical information is then obtained by using the Wishart-Eigen value unsupervised classification.[1-2] In our previous paper titled “Machine Learning Approach for Predicting Earthquakes in a Geographic Location” Machine Learning Approach was used with lower accuracy. In this paper key methods of pre-processing were used to increase the accuracy of SVM.
Keywords: Machine learning, eigenvalue, Polarimetric Synthetic Aperture Radar (PolSAR), PU-Learning (PUL), and eigenvalue statistical components (ESC).
|
DOI:
10.17148/IJARCCE.2022.111140
