Explore chapters and articles related to this topic
Fundamentals
Published in N.C. Basantia, Leo M.L. Nollet, Mohammed Kamruzzaman, Hyperspectral Imaging Analysis and Applications for Food Quality, 2018
The food industry has widely applied both imaging and spectroscopic technologies for quality and safety evaluation. Spectral imaging integrates the two technologies by combining their main features to acquire spatial and spectral information simultaneously. Conventional imaging technology is one of the mostly widely used alternatives to manual inspection and has become an integral part of the food industry’s move toward automation (Sun, 2010). Imaging systems usually include computer vision by a camera utilizing either monochromatic (black and white) or polychromatic (color-based) light, complemented with image processing and analysis, involving mathematics, computer programming, and software programming. These systems acquire either two- or three-dimensional spatial information and can often assess several objects per second, leading to very high online throughput (ElMasry & Sun, 2010).
Device characterization
Published in Sharma Gaurav, Digital Color Imaging Handbook, 2017
A major concern in spectral imaging is the substantial increase in the amount of data to be handled (i.e., from 3 to 30 or more channels). This necessitates an efficient encoding scheme for spectral data. Most encoding techniques are based on the well-known fact that spectra found in nature are generally smooth and can be well approximated by a small number (i.e., between five and eight) of basis functions.85 The latter can be derived via principal-component analysis (PCA) described in an Chapter 1. PCA yields a compact encoding for spectral data and can serve as the device-independent color space in a multispectral imaging framework. An important consideration in selecting the PCA encoding is to ensure compatibility with current colorimetric models for color management.86
Application of Image Processing and Data Science in Advancing Agricultural Design
Published in Ankur Dumka, Alaknanda Ashok, Parag Verma, Poonam Verma, Advanced Digital Image Processing and Its Applications in Big Data, 2020
Ankur Dumka, Alaknanda Ashok, Parag Verma, Poonam Verma
Hyperspectral imaging is a spectral imaging technique which employs each pixel of image for acquiring set of images with certain spectral bands. This technique combines the advantages of optical spectroscopy as an analytical tool with two-dimensional (2-D) object visualization obtained by optical imaging. Thus, the sensor samples the hyperspectral cube in four different ways as spatial scanning, snapshot imaging, spectral scanning, and spatio-spectral scanning.
Accuracy of Hyperspectral Imaging Systems for Color and Lighting Research
Published in LEUKOS, 2023
Aiman Raza, Dominique Dumortier, Sophie Jost-Boissard, Coralie Cauwerts, Marie Dubail
Spectral imaging combines conventional imaging with spectroscopy and provides image data containing spatial and spectral information. For each pixel of the image, the spectral power distribution (SPD) is measured (or retrieved) to generate datasets with three dimensions (also called data-cubes). With spectral imaging, radiometric measurements are no more restricted to a limited number of points but can be applied to the visible surface of the object or to an entire visual scene. There is no standardized limit defined in spectral imaging to differentiate between multispectral and hyperspectral imaging (Foster and Amano 2019; Westland et al. 2012). While it seems to be discipline-dependent, the difference is always related to the number of spectral bands. In general, a system is called multispectral if it has strictly more than 3 spectral bands (to differentiate it with a conventional RGB camera) and hyperspectral if it has more than 20 bands (Vasefi et al. 2016). Even if some studies suggest that 10 bands are sufficient to recover the spectral information (Imai et al. 2003), for high spectral accuracy, a higher number of bands are required (Vilaseca et al. 2015). Particularly if one wants to measure the effect of narrow spectral peaks like those under LED sources. For lighting and color research, CIE (Commission Internationale de l’Eclairage) recommends the calculation of chromatic coordinates from spectral data with a spectral resolution of 5 nm or less (CIE 15, 2018). This corresponds to a minimum of 80 bands in the range of 380 to 780 nm.
SHBO-based U-Net for image segmentation and FSHBO-enabled DBN for classification using hyperspectral image
Published in The Imaging Science Journal, 2023
Tatireddy Subba Reddy, V. V. Krishna Reddy, R. Vijaya Kumar Reddy, Chandra Sekhar Kolli, V. Sitharamulu, Majjaru Chandrababu
Spectral imaging is an imaging technique that contains various bands over the electromagnetic range. The extension of multi-spectral sensors is regarded as hyper-spectral sensors and the brief information delivered by this sensor improves the power of accurately discriminating materials of interest with high classification accuracy. This research proposes an innovative approach for HSI classification using the proposed FSHBO-DBN. HSI is considered as an input for the whole process acquired from a specific dataset represented in Ref. [27] and this input image is pre-processed using a Gaussian filter to eradicate the redundant noises. The next step is the segmentation, which is efficiently performed by exploiting U-Net and this classifier is finely tuned using SHBO. After commencing the segmentation process, features like spectral spatial features and vegetation indices are extracted by considering the pre-processed image as an input in which spectral spatial features include Local Binary Pattern (LBP) [28], Local Gabor XOR Pattern (LGXP), Multi-Texton, whereas the Vegetation index features include Ratio Vegetation index (RVI), Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Wide Dynamic Range Vegetation Index (WDRVI). After that, the segmented result and feature vector is assumed as input for HSI categorization and this process is effectively accomplished through DBN. The weights of DBN are optimized for wielding FSHO. The proposed FSHO is obtained by consolidating the FC [29] concept into SO and HBO. The pictorial illustration of HSI classification using FSHBO-DBN is illustrated in Figure 1.