Multiclass Approaches for Support Vector Machine Based Land Cover Classification

TL;DR

This paper compares six multiclass SVM approaches for remote sensing land cover classification, highlighting the superior accuracy and efficiency of the One vs. One method.

Contribution

It provides a comparative analysis of multiple multiclass SVM strategies, emphasizing the effectiveness of the One vs. One approach in remote sensing applications.

Findings

One vs. One approach achieves higher accuracy.

One vs. One reduces computational cost.

Other methods are less effective in accuracy and efficiency.

Abstract

SVMs were initially developed to perform binary classification; though, applications of binary classification are very limited. Most of the practical applications involve multiclass classification, especially in remote sensing land cover classification. A number of methods have been proposed to implement SVMs to produce multiclass classification. A number of methods to generate multiclass SVMs from binary SVMs have been proposed by researchers and is still a continuing research topic. This paper compares the performance of six multi-class approaches to solve classification problem with remote sensing data in term of classification accuracy and computational cost. One vs. one, one vs. rest, Directed Acyclic Graph (DAG), and Error Corrected Output Coding (ECOC) based multiclass approaches creates many binary classifiers and combines their results to determine the class label of a test pixel. Another catogery of multi class approach modify the binary class objective function and allows simultaneous computation of multiclass classification by solving a single optimisation problem. Results from this study conclude the usefulness of One vs. One multi class approach in term of accuracy and computational cost over other multi class approaches.