Dimensionality reduction for hyperspectral imagery plays a major role in different scientific and technical applications. It enables the identification of multiple urban-related features on the surface of the earth, such as building, highway (road), and other natural and man-made structures. Since manual road detection and satellite imagery extraction is time-consuming and costly, data time and cost-effective solution with limited user interaction will emerge with road and building extraction techniques. Therefore, the need to focus on a deep survey for improving ML techniques for dimensionality reduction (DR) and automated building and road extraction using hyperspectral imagery. The main purpose of this chapter is to identify the state-of-the-art and trends of hyperspectral imaging theories, methodologies, techniques, and applications for dimensional reduction. A different type of ML technique is included such as SVM, ANN, etc. These algorithms can handle high dimensionality and classification data.
TopIntroduction
Remote hyperspectral detection also referred as to spectroscopy, reflects a full use of field-vegetation, resources, and land use/terror monitoring by scientists and researchers. While this information has been available since 1983 in various fields of engineering and science, it is primarily used by many complicated factors. For many years’ scientists, in particular, physicists have used spectroscopy for the identification of material composition. In the field of analytical chemistry, many techniques used for reflectance spectrum analysis have been developed. Identify the characteristics of the individual absorption by using solid/liquid chemical gassing bonds. Technological progress made it possible to extend image spectroscopy to satellite applications outside laboratory conditions to concentrate their applicants globally (Volchok, B. A., and Chernyak, M. M 1968). Figure 1 displays the schemes used in our research of the hyperspectral imaging system. The following components are typically included in a standard hyperspectral imaging device: a light source (illumination), a wavelength distributor (spectrometer), and a zone detector (camera), a transportation system (Qin, J., et.al., 2013). The light source can usually be divided into two groups for spectral imaging application: illumination and excitation source. Broadband lights are typically used as sources of illumination for reflection and transmitting images, whereas narrowband lights are widely used as sources of excitation. Illumination is therefore a key aspect of the hyperspectral imaging system. Compared to naked eyes, vision systems are influenced by the intensity and consistency of lighting. Illumination devices produce light that illuminates the image features evaluated; thus, the output of the illumination system can significantly affect the quality of the images and play an important role in the overall efficiency and accuracy of the system (Liu, D.et.al., 2015). Effective lighting can help enhance image processing and analysis performance by minimizing noise, shadow, reflection, and improving contrast image (Zhang, B., et.al., 2014). Furthermore, the locations, types of lamps, and color quality of the lighting are all considered when selecting the most effective lighting. The widely used sources of light are incandescent lamps, fluorescent lamps, lasers, and infrared lamps (Kodagali, J. A., and Balaji, S 2012).
Figure 1. A hyperspectral imagery schematics system
Hyperspectral was used synonymously with the spectrometer for imagery in some books. Not all the spectral bands can be used in the electromagnetic spectrum for remote sensing purposes. The consuming bands appear to be isolated where remote sensing is possible by atmospheric windows or areas. In these atmospheric windows, hyperspectral images are measurements. The remote sensing technology combines the imaging and spectroscopy of the hyperspectral in one system, which produces large groups of data requiring complex handling procedures. In general, data set hyperspectral consist of around 100 to 200 spectral bands which, in contrast to the multi-spectral data sets, possess only five to ten bands with relatively large bandwidths that are relatively small. The Hymap or hyperspectral mapper used for airborne and visible / IRS imagery (AVIRIS) is an example of a hyperspectral device. NASA was first used in the early 1980s.
Hyperspectral remote sensor system records hundreds of relatively small bandwidth spectral bands (5 to 10nm) together with these details; it is greatly improved that unique tends have been detected and identified on the ground and atmosphere. It makes it possible to analyze land cover by far more specific. The emissivity levels of each band can be combined to form a spectral reflectance curve. Hyperspectral data may produce higher accuracy of the classification and a more detailed taxonomy. But it is also a unique challenge to classify hyperspectral data (Wilheit, T. T., et.al., 1977).
Figure 2. Concept of an imaging spectrometer