Analysis of vegetation classification on hyper spectral images using swarm optimization techniques

Abstract

Hyperspectral Image classification and segmentation is unique technique to examine diversified land cover images. The main resource controlling productivity for terrestrial ecosystem in land cover classification and the challenges are subjected to curse of dimensionality, which is termed as Hughes phenomenon. Nowadays, it is vital to distinguish how land cover has distorted over time. Over the globe, land area is confined only to 29% of the surface while the majority share is covered by enormous spread of water. Human has to depend upon less than one- third of the surface, which is around 148,300,000 sq.km.area, as a habitable part. It is estimated that more than two-third of the land cover area is not suitable for habitation. Approximately 27% of land area is characterised by cold environment. Dry and extreme weather conditions also accounts for 21% of the land surface. Nearly 4% of the total land surface accounts for uneven geographic conditions. Hence, it exemplify that land is inimitable and ubiquitous resource. This research is focussed on provision for land use vegetation classification using AVIRISHyperspectral images. The contiguous narrow bandwidth of hyperspectral data facilitates detailed classification of land cover classes. The issues of utilizing hyperspectral data are that they are normally redundant, strongly correlated and subject to Hughes phenomenon. Thus, various techniques have been proposed to overcome the Hughes phenomenon in classification. Recently, image processing approaches are utilized in different applications. This procedure relies upon satellite image, MRI image, worldview image, and so forth the machine learning approaches are utilized for classification and segmentation of those images. Based on the application sets, the machine learning procedures are performed with advancement strategies for improving the exactness of the images. The proposed methodology utilizes the vegetation classification, worldview segmentation identification and mapping of natural vegetations, medical image for disease detection, and optimization techniques. Normal vegetation and its developments are distinguished utilizing a AVIRIS spectral imagery. The classification of the AVIRIS image and subordinate topical information was performed by utilizing an Improved Relevance Vector Machine with Mosquito Flying behaviour based swarm intelligence Optimization (IRVM-MFO) technique. Additionally, texture features, for example, Fourier spectrum and GLCM features are extracted to improve the characterization precision. In IRVM, the MFO approach is used to optimize the kernel functions of parameters to improve the preparation procedure. The proposed strategy brings about a high precision rate when compared with the existing SVM and RVM classifiers newline