Integrating Deep Learning for Optimized Accuracy of Hyperspectral Long Distance Imagery Classification
Abstract
Hyperspectral imagery is a complex form of imaging technique usually opted by researchers for long distance image processing. These images are usually captured by two different acquisition devices, namely color acquisition and shape acquisition device, and then fused together in order to get the aggregated information about the full image. The fused image is then used for various kinds of applications based processes like clustering, classification. In this paper authors propose a novel classification technique which uses deep learning and adaptive selection of feature so as to optimize the classification accuracy. The proposed system obtains more than 95% accuracy when tested for crop identification, which is nearly a 5% improvement than the standard neural network based classification systems. In this work, brovey and Principal Component Analysis (PCA) are hybridized for fusion, and color maps & shape maps are used for feature extraction
Keywords: Hyperspectral, deep learning, color map, shape map, classification, fusion.