A Hybrid Sequential Classifier for Hyperspectral Imagery using Deep CNN with Ant Colony Optimization

This paper proposes a novel hybrid approach for hyperspectral image classification using a processing sequence consisting of deep Convolutional Neural Network (CNN) followed by Ant Colony Optimization (ACO). Thus, this hybrid classifier takes into account the joint spectral-spatial information. The first processing step performs hyperspectral pixel classification with CNN by taking into account both the spectral and also the spatial information. The second classification stage based on ACO model uses mainly the spatial pixel correlation and it refines the results obtained in first classification step. The proposed hybrid sequential classifier is evaluated on Indian Pines and Pavia University hyperspectral datasets. The proposed hybrid model has led to a better accuracy by comparison to CNN and SVM classifiers.