Article Preview
TopIntroduction
In many application areas where visual representation in form of images is involved, the use of spectral information is necessary to perform various tasks such as image enhancement, segmentation, and classification. Spatial and Spectral information is used to analyze the presence of chemical patterns in the high-resolution image to provide a better evaluation of features. The use of spectral correlation among various features forms a basis for the need to select relevant bands from multispectral images having a pool of hundreds of bands. In this paper, the study of high-resolution images such as multispectral and hyperspectral images is illustrated with an utter need to select bands to reduce their dimensionality. The band Selection process is then evaluated with classification technique and is explained in detail in further sections.
Taking into consideration of simple image, represented by a 2-d matrix of any object in a scene; Images can be analog or digital (J Kuruvilla et. al, 2016). All aerial photographs are the type of analog images whereas latter images are acquired by electronic sensors of different wavelengths and are stored on the computer system in digital form. Other classifications for the type of image are binary images, colored images, multispectral images, hyperspectral images, and super spectral images. Binary images have the intensity values as 0 or 1, having only 1 band whereas colored or RGB images have intensity values based on the number of bits each pixel holds and has only 3 bands. High-Resolution Images such as multispectral, hyperspectral, and super spectral images are the ones having more than three bands. These images are used to capture minute details of a particular object in a scene across various spectral bands. Here, remote sensing image as a digital image is used in the research work. A Remote Sensing image is an image that represents part of the Earth’s surface. Each pixel in an image shows some area on Earth and has an average intensity value that measures solar radiance in a wavelength band reflected from the ground. These images are generally multi-layered images that are constructed by stacking the images taken for the same area by the same or different sensors. Multispectral image (Sotoca, J. M. et. al, 2007) is a special case of a multi-layered image consisting of a few image layers of a particular scene. Each image layer is taken at a particular wavelength depending upon capturing sensor. The image used in this paper is AVIRIS image consisting of several bands in blue, green, red, near-IR, SWIR & thermal bands. Another set of images having hundreds of bands, known as hyperspectral images, is also available which has deep spectral knowledge about an object in the scene, enabling better identification and classification. Currently, this paper is focused on the analysis of high resolution images and their classification (De Backer, S. et. al, 2005).
In figure 1, the concept of spectral imaginary is shown. An airborne sensor simultaneously samples multiple spectral wavebands over a large area in a ground scene. After appropriate processing, the resultant image contains a sample of spectral reflectance measurement which is interpreted to identify the object present. It presents the spectral variation in reflectance for soil, water, and air. Multispectral images are used to measure light in the electromagnetic spectrum, it is different from hyperspectral bands in the way that latter images consist of several bands which increase the complexity of bands. These types of images are used for space imaging and likewise for reporting in painting and for investigation purposes.