Conditional Optimal Filter Selection for Multispectral Object Classification

TL;DR

This paper proposes a conditional filter selection method for multispectral object classification that reduces spectral redundancy and misclassification, enabling low-cost hardware setups without sacrificing important spectral information.

Contribution

It introduces a new filter selection approach incorporating preset conditions, improving over existing methods by reducing misclassification and hardware costs.

Findings

Reduced misclassified objects from 318 to 124 on SMM dataset

Achieved lower spectral redundancy with minimal filter sets

Enhanced classification accuracy with cost-effective hardware

Abstract

Capturing images using multispectral camera arrays has gained importance in medical, agricultural and environmental processes. However, using all available spectral bands is infeasible and produces much data, while only a fraction is needed for a given task. Nearby bands may contain similar information, therefore redundant spectral bands should not be considered in the evaluation process to keep complexity and the data load low. In current methods, a restricted and pre-determined number of spectral bands is selected. Our approach improves this procedure by including preset conditions such as noise or the bandwidth of available filters, minimizing spectral redundancy. Furthermore, a minimal filter selection can be conducted, keeping the hardware setup at low costs, while still obtaining all important spectral information. In comparison to the fast binary search filter band selection method, we managed to reduce the amount of misclassified objects of the SMM dataset from 318 to 124 using a random forest classifier.