A MULTIMODAL FUSION ARCHITECTURE COMBINING DTICF-3D- CNN–BI-LSTM AND RGB-ENHANCED 2D-CNN FOR EARTH SURFACE HSI CLASSIFICATION

Abstract

Hyperspectral imaging technology is one of the most efficient and growing technologies
derived from Image Sensing technology. HSI classification contains rich spectral
information which is used to identify and classify the information of the earth surface. Due
to large dimensionality of spectral information, HSI classification still faces several
challenges and also creates the problem of detecting the target of the classes accurately. To
overcome this problem, we proposed a Spectral-Spatial Classification of HSI using fusion
of 3-Dimensional Convolutional Neural Network with Bi Directional Long Short Term
Memory Network (3D-CNN-BiLSTM)and 2-Dimensional-CNN (2D-CNN) approach.
Finally, Markov Random Field (MRF) is utilized to encourage the neighboring pixel which
helps to enhance the classification results. To prove the better performance of the proposed
techniques, some of the existing Machine learning and deep learning techniques are
compared. From this analysis, it is evaluated that, the proposed techniques provide the best
results, when compared to other techniques.