Performance evaluation of spectral indices and classification algorithms for built-up area extraction using PRISMA hyperspectral images

Abstract

This article aims to evaluate and compare NDBI, NBI, PB1BI and HIBI spectral indices and SVM, ANN and MLC classification algorithms in order to identify and extract urban constructions using Prisma hyperspectral images. took The findings of this research indicate that the classification algorithms in both Tehran and Urmia have higher accuracy than the spectral indices; So, in Tehran city, PB1BI and HIBI indices have higher accuracy than NDBI and NBI indices with overall accuracy of 85% and 86% and kappa coefficient of 70% and 72% respectively from left to right. On the other hand, in Urmia city, NBI indices with 88% overall accuracy and 77% kappa coefficient and NDBI with 87% overall accuracy and 75% kappa coefficient showed better performance than PB1BI and HIBI indices. Also, in Urmia city, the overall accuracy and Kappa coefficient of SVM and ANN classification algorithms were more accurate than MLC with over 90%. Also, in the city of Tehran, SVM and ANN algorithms with overall accuracy and high Kappa coefficient of 90% and 83% performed better than the MLC algorithm. In general, according to the effectiveness of various factors including the scope of the study, the spectral range used, the type of roof of the buildings, the types of uses, etc., the combined and comparative use of indices and spectral algorithms improves the results.