This paper investigates the efficiency of spectral screening as a tool for speedup in hyperspectral image processing. Spectral screening is a technique for reducing the hyperspectral data to a representative subset of spectra. The subset is formed such that any two spectra in it are dissimilar and, for any spectrum in the original image cube, there is a similar spectrum in the subset. The similarity can be described through various spectral distances and can be controlled by a threshold value. The spectral screening is improved by associating with each spectrum in the subset a weighing factor proportional to the number of spectra in the original image that are similar to it. Following its generation, the subset is used in further computations instead of the full data. The resulting processing mappings are then applied to the data. The investigation focused on the comparison between distance measures such as spectral angle and spectral correlation angle, in terms of efficiency of the results and speedup obtained when tested with Principal Component Analysis (PCA) and Independent Component Analysis (ICA), two processing techniques used when dealing with hyperspectral data. We also investigated the advantage of weighting versus non‐weighting the spectral subset, and the optimum performance of the screening algorithm. The experiments were performed on HYDICE, Hyperion and AVIRIS data and validate the usefulness of spectral screening for data reduction. Preprocessing through spectral screening provides significant speedup to PCA and ICA without reduction in data accuracy.

Recent advances in AI include a Law firm hiring a robot lawyer and companies developing autonomous vehicles with robot drivers. Findings from our study have gauged the current cognitive capacity of such systems, indicating areas for improvement. We focus on autonomous vehicles, i.e., those that conduct automated driving and need to make autonomous, i.e., independent decisions. We propose an approach enabled with commonsense knowledge (CSK) from worldwide repositories to simulate intuitive humanlike decision-making in autonomous vehicles. We consider the repository WebChild with a multitude of CSK concepts, properties and relations. We investigate this and related domain-specific knowledge bases (domain KBs) to harness them within our proposed approach. Accordingly we build a transportation domain KB incorporating CSK and the needs of autonomous vehicles. This would be useful in guiding automated driving and making the systems get closer to the thresholds of human cognition. This work thereby makes contributions to smart mobility in smart cities. The paper presents our vision with design, implementation, experiments, recommendations and a future roadmap. As a broader impact, it propels more joint work between AI, Law and related areas.

In this paper we introduce a novel feature extraction method based on Nonnegative Matrix Factorization (NMF) for hyperspectral image processing. Given the large size of the hyperspectral imagery, feature extraction plays an important role in producing fast and accurate results. Traditional approaches such as Principal Component Analysis and Independent Component Analysis generate the features as a linear combination of the hyperspectral bands emphasizing on the decorrelation or independence of the features. Compared to this, NMF offers a decomposition solution that is less restrictive requiring only the positivity of the features and the associated linear transform. Such scenario has a natural meaning in hyperspectral imagery where each pixel observation is thought to be formed as a linear positive mixture of reflectance values of the materials in the scene (endmembers) covered by the pixel. With hyperspectral imagery spatial resolution ranging from millimeters to kilometers, it likely that the data observed are formed as a mixture. In this case, the linear transform used to generate the features would be associated to the endmembers and the resulting features would be associated to the abundance of each endmember in the pixels. We present our results on using NMF for feature extraction by performing experiments with hyperspectral digital imagery collection experiment (HYDICE) data as well as in-house imagery collected with a SOC 700 hyperspectral camera. The experiments suggest that NMF outperforms PCA in feature and endmember extraction.

Feature extraction refers to the groups of techniques that, when applied to large dimensional and redundant data result in significant dimensionality reduction while preserving or even enhancing the information content. Among various techniques investigated for feature extraction, of new interest is nonnegative matrix factorization (NMF). In NMF, it is assumed that the data is formed as a linear nonnegative combination of positive sources and the NMF solution recovers the original sources and the mixing matrix. In this paper, we first look at ways NMF can be applied for feature extraction in hyperspectral imagery a data known for large sizes and redundancy. While some of the associations are natural to linear mixing model (LMM - that assumes that hyperspectral images are formed as a linear mixture of endmember information), we also show NMF to be a slow method. To counter this, we investigate alternative solutions such as projected NMF approaches and provide an insight to how parallel implementations would contribute to speedup. Experimental results on various data show projected NMF outperforming regular NMF with parallel implementations providing a promising speedup advantage.

We investigate the use of sparse principal components for representing hyperspectral imagery when performing feature selection. For conventional multispectral data with low dimensionality, dimension reduction can be achieved by using traditional feature selection techniques for producing a subset of features that provide the highest class separability, or by feature extraction techniques via linear transformation. When dealing with hyperspectral data, feature selection is a time consuming task, often requiring exhaustive search of all the feature subset combinations. Instead, feature extraction technique such as PCA is commonly used. Unfortunately, PCA usually involves non-zero linear combinations or 'loadings' of all of the data. Sparse principal components are the sets of sparse vectors spanning a low-dimensional space that explain most of the variance present in the data. Our experiments show that sparse principal components having low-dimensionality still characterize the variance in the data. Sparse data representations are generally desirable for hyperspectral images because sparse representations help in human understanding and in classification.