Over the last couple of decades, sparse signal processing has been constantly the focus of research interests aiming to exploit all benefits of sparseness property. Particularly, the sparsity has been utilized in numerous applications such as compression and coding, spectral analysis, channel estimation, and recently for the development of compressive sensing (CS) approaches and signal reconstruction algorithms. Indeed, intensive development of the approaches for sparse signal recovery in the CS scenario has created a new sensing paradigm and opened up a new perspective of sensing technology design. Moreover, a variety of CS scenarios have been exploited in different real-world applications, not only for sensing purposes but for example in signal/image denoising problems. A number of approaches and algorithms have been designed to cope with under-sampled and random-sampled signal recovery, depending on the signal nature.

Although signal processing algorithms can be tested in software, there is a growing need for hardware realizations. Namely, software implementations of those approaches are generally not suitable enough for real-time applications. Therefore, the necessity for hardware implementations was raised due to the requirement of fast testing, which is hardly accomplished in software. There are various approaches starting from the less complex and much faster analog-circuits-based solutions to the more complex, slower but more precise digital approaches. Depending on the system requirements, analog or digital implementation is used. Generally, for fast and less complex circuits, analog implementation is better, while for slower circuits, a digital implementation is more suitable.

In this chapter, we will provide an overview of the hardware architectures for several approaches based on the gradient descent algorithm. The algorithm belongs to the group of convex optimizations and provides successful signal recovery from a relatively small number of acquired samples.

Cognitive sensors (CS) for healthcare are devices that utilize various sensor technologies to monitor a person's health condition and provide them with appropriate rehabilitation services. These sensors can be worn on the body, embedded in clothing, or integrated into other devices such as smartphones, smartwatches, and tablets. These sensors can monitor various indices of health and diseased conditions. Data related to human health are increasingly integrated with cognitive sensors, which facilitates decision-making and better disease treatment. Further, real-time communication and patient monitoring devices have become a central concept of smart healthcare services. The CS networks consist of a large number of sensors that can collect localized information automatically and intelligently. In addition to information fusion, intelligent sensors, sensing grids, communication protocols, network routing, and complex event processing architectures, CS intersects with other trends in computing. It has been shown that such CS can significantly improve the quality of life and support for the elderly and physically handicapped by designing interactive, adaptable, and intelligent multisensory surveillance systems, which can be equipped with panic buttons, current, bed, and water flow sensors. The use of CS technology, such as sensors, can reduce directional errors and assist in the neural health of patients. Many cognitive sensor systems have been developed for rehabilitation purposes to assist various groups of individuals in overcoming physical or mental health conditions. However, only a few attempts have been made to assist individuals with traumatic brain damage, schizophrenia, or dementia.

The most prominent issue in the acquisition of biomedical signals, apart from the problems associated with the sensors for capturing these signals, is the presence of artifacts. In this chapter, these artifacts are discussed mainly with respect to electroencephalographic (EEG) signals. The chapter introduces EEG and the different methods employed to capture the EEG signals. In particular, the advantages of non-invasive EEG and the challenges associated with the captured signals are highlighted. The artifacts present in EEG can be broadly classified as (a) physiological and (b) non-physiological. The physiological artifacts such as ocular cardiac artifacts and non-physiological artifacts such as power line interference are elaborated along with the reason for such artifacts. In this chapter, three physiological artifacts, namely, ocular, cardiac, and electromyogenic, are explained along with two non-physiological artifacts, namely, electrode movement and power line interference. To effectively utilise EEG, the signals need to undergo artifact removal. Multiple methodologies are employed to remove the artifacts from EEG, and the most widely used methods involve Independent Component Analysis (ICA) and Wavelets. The ocular and electromyogenic artifacts can be handled using a range of methods such as adaptive filtering, blind source separation, wavelet transform, Riemannian Geometry, and deep learning approaches. Cardiac artifacts are commonly eliminated using adaptive filtering as well as blind source separation and deep learning approaches. One commonly used method to tackle the electrode placement artifact is independent component analysis. Power line interference is mainly eliminated using notch filters, however, adaptive filtering and hybrid methods of wavelet and blind source separation are also explored in multiple works. These methods are explained and analysed in this chapter, highlighting each one's advantages and disadvantages. Additionally, the challenges in employing a commercial EEG headset compared to a medical EEG headset are also explored. The chapter concludes with an overview of various methods employed for artifact removal in EEG and associated active research areas.

The manual techniques for detecting Parkinson's disease (PD) include computer examination tests, interviews, and activity-based recognition. However, such methods are burdensome, error-prone, and time-consuming. Therefore, there is a need for an automatic decision-making support system for the timely detection of PD. This chapter explores the utility of electroencephalogram (EEG) signals for accurate PD detection. The EEG provides detailed information about distinct neural characteristics during PD. But, the nonstationary and nonlinear nature of EEG makes its analysis difficult, resulting in reduced system performance. Therefore, this chapter presents an empirical wavelet transform-based nonlinear analysis of EEG signals. Various hidden characteristics (statistical features) are extracted from the decomposed multi-component EEG and classified using nonlinear multiple machine learning techniques. The effectiveness of the proposed model is tested by evaluating various performance measures and comparing them with existing state-of-the-art methods. The proposed model has obtained the highest accuracy of 98.22% and 98.65% in classifying EEG events of healthy control (HC) subjects vs. PD patients with the absence of medication (PDAM) and HC vs. PD with medication (PDOM). In addition to this, five performance matrices are also evaluated to test system performance. The developed model automatically detects PD from different brain regions. The brain's frontal area provides insightful information in PD detection with the highest accuracy. In contrast, the temporal part of the brain has obtained the least accuracy of PD detection. The developed model is ready to be deployed for PD detection by neuro experts in real-time scenarios.

Cognitive Internet of Things (IoT) sensors are advanced sensing devices that incorporate artificial intelligence (AI) and machine learning (ML) algorithms to process and analyze real-time data from various sources. These sensors are widely used in smart industrial and biomedical applications for their ability to provide real-time information and improve operational efficiency. In industrial applications, cognitive IoT sensors can be used to monitor production processes and provide predictive maintenance. In biomedical applications, they can be used to monitor patient health and improve diagnosis and treatment. With the continued growth of the IoT and advancements in AI and ML, the use of cognitive IoT sensors is expected to increase in both industrial and biomedical applications in the coming years. This book chapter presents the innovative approach to decision making in healthcare and industrial settings, focusing on the handling of data transmission, trade, and record keeping through a cost-effective preservation method. The chapter serves as an overview of cognitive sensors and the current state of IoT in smart industries. It also delves into the impact of cognitive sensors on the operations of hospitals and industrial organizations, and explores the use of these sensors to understand the reasons behind resistance to smart products among both producers and consumers.

Industrial Internet of Things (IIoT) is focused on optimization of industrial working in several fields including assembly lines, manufacturing processes and production of any kind, including agriculture. As the equipment and system are data intensive and largely dependent on wireless transmission of data from the interface point of the primary sensing element to where it is going to be utilised, adding cognitive capabilities to the radio system involved can add benefits of optimal utilisation of the frequency spectrum and faster transmission of critical process data. IIoT also offers advanced functionalities such as preventive maintenance, asset tracking and better optimized resources in the operation and control of industrial processes for manufacturing, production, or processing. This chapter focuses on the evolution of the sensors and process control equipment from traditional versions to those being used today for IIoT.

Agricultural lands are generally vast, remote, and sensitive to weather, making data collection difficult. Despite these limitations, additional data are being captured as technologies evolve and prices reduce. Proximal, aerial, satellite, and ancillary data are obtained to study the crop for various objectives of smart agriculture. Geographically unequal areas lack vital data to address their challenges. Even in regions with adequate infrastructure and resources, data being obtained for agricultural environments produce knowledge gaps. Many aspects must be considered and measured for a region's whole description or examination of the targeted problem. Data from a single type of sensor is not very effective for algorithms, and results are imprecise. However, fusing sensor data is one of the key approaches for crops' quality assessment, damage, and smart planning for their management. The data fusion explores complementarities and synergies of extracting trustworthy and relevant information from data. Exploring the correspondence and interactions between various types of data is the core concept behind data fusion, which aims to derive more trustworthy and applicable information regarding the topics being investigated. Despite some success, several obstacles still limit a more general adoption of this type of strategy. This is especially true for the highly complicated ecosystems that can be found in rural and agricultural settings. In this part of book, we provide a comprehensive review on data fusion to agricultural problems and study of plant disease using different sensor data for smart agriculture. Conclusively, the study contributes on emphasizing the data acquisition scales, data types, and their fusion for decision making in smart agriculture.

Ghulam Mustafa1 and Qurban Ali contributed equally to this chapter.

Natural disasters such as floods, landslides, and forest fires have caused infrastructure destruction, damage to property, injuries, and casualties worldwide. With the advent of sensing technology and data science methodologies, the impact of a disaster can be lessened through in-depth retrospective analysis. Of these, a priori hazard susceptibility analysis, which includes spatial analysis using remotely sensed images of the terrain, is widely adopted. Susceptibility analysis is thus related to spatial aspects of hazard analysis using data captured using both active and passive sensing. In the case of certain disasters, temporal information plays the role of causality owing to which spatiotemporal aspects are relevant. Historical data of occurrences play a vital role in determining the likelihood of disaster occurrence in a new area, or recurrences in an affected site. Such hazard mapping also helps in predicting the degree of the aftermath of a hazardous process in those areas. Overall, state-of-the-art susceptibility analysis uses remote sensing images, land survey data, and other related geospatial and non-spatial data in performing data preprocessing and analysis to calculate the probability of forthcoming hazard events. Data preprocessing involves noise removal, the transformation of raw data to the desired data structure or computed values, dimensionality reduction, feature ranking, and so on. Data analysis includes statistical, machine learning, and deep learning approaches. In this chapter, the widely used preprocessing and data analysis techniques applied in disaster susceptibility analysis are discussed. Additionally, disaster susceptibility analysis-related case studies, e.g. flood and landslide, are presented to elaborate the data science workflows. The resultant maps of the disaster susceptibility analysis are intended to be used by land developers, urban planners, and related authorities in innovating the land use planning and smarter disaster management.

Detailed discussion of cognitive sensing technologies, their fundamentals, review and appropriate applications provide lot of insights as what would be future applications of smart and cognitive sensing. In years to come, the sensing technologies are going to dominate various industries such as driver-less cars, environment monitoring, smart transport, and smart city. In all such applications, unlimited automation, control, and monitoring shall be salient features that are expected to be realized in Metaverse, Web 3.0 and similar technologies using digital twins, digital avatars, etc. This chapter presents future direction of smart and cognitive sensing for various sectors and their sustainable growth through digital transformation and Industry 4.0.