Histogram Binning and Morphology based Image Classification

A hyperspectral image is characterized by a large dimensionality data, recorded at very fine spectral resolution in hundreds of narrow frequency bands. These bands provide a wealth of spatial and spectral information of the scene, imaged by the sensors. Histogram binning and morphological operator based Extreme Learning Machine classifier is proposed. Histogram binning and morphological operator is used as a preprocessing step which reduces the computational complexity, which involves in the hyperspectral data processing. Each object have distinct reflectance value based on which the objects are classified by choosing appropriate features and the classifier. Histogram binning represents the image by reduced gray level. Morphological operations envisage the finer inner details of the image. Now from this new version of the processed image, statistical features such as mean, median, standard deviation, mode and variance have been extracted. The extracted features are used for classification. The Extreme Learning Machine is proposed to classify the image. The classification is done with different types of kernels. The performance of each type of kernel is evaluated. The experiment is conducted on the AVIRIS hyperspectral dataset taken over the North-western Indiana s Indian Pine Site.