A Multidimensional Pixel-wise Convolutional Neural Network for Hyperspectral Image Classification

This paper presents a novel multidimensional pixel-wise convolutional neural network (MPCNN) to extract spatial and spectral-spatial information from the hyperspectral image (HSI). A hyperspectral image consists of narrow spatial and spectral band information based on the nature of visible materials and infrared regions of the electromagnetic spectrum. The release electromagnetic energy from visible material makes the specific wavelength which is used to classify the objects. The classification of hyperspectral image is one of the challenging task due to its narrow band energy formation. In this paper, we propose a MPCNN algorithm for classification of HSI based on two and three dimensional pixel-wise information. The term pixel defines the spectral vectors of proposed MPCNN that represents the ground material's energy radiation to the entire detection bands. This is done by using the convolutional neural network (CNN) to obtain spectral-spatial semantic feature information of hyperspectral image. The effectiveness of the proposed MPCNN is measured by classifying the objects in spatial and spectral-spatial domain and compared with different traditional CNN methods. The comparison result shows that the proposed MPCNN algorithm is capable to classify the hyperspectral image with 99.09% accuracy, while the MS-CLBP method achieves 91.51% accuracy.