Loading...
Dr. Mandeep Kaur is an Assistant Professor in the Department of Electronics and Communication Engineering at Punjabi University, Patiala, India. Her research interests include Computer Vision, Deep Learning, Pattern Recognition, Intelligent Transportation Systems, and Artificial Intelligence.
Traffic Sign Recognition (TSR) is an essential element of Intelligent Transportation Systems (ITS), Advanced Driver Assistance Systems (ADAS), and driverless vehicles. Precise identification of traffic signs facilitates safer driving, superior traffic management, and heightened awareness among road users. Conventional image-processing methods frequently encounter difficulties due to fluctuations in lighting, atmospheric conditions, obstructions, motion blur, and intricate backdrops. Deep Learning (DL) has transformed traffic sign identification through enhanced feature extraction, categorisation, and real-time detection capabilities. Advanced models, including Convolutional Neural Networks (CNNs), Residual Networks (ResNet), Vision Transformers (ViTs), and hybrid deep learning architectures, have exhibited remarkable efficacy in the recognition and classification of traffic signs under varied situations. This chapter examines the principles, methodology, datasets, architectures, applications, problems, and future trajectories of deep learning-based traffic sign recognition systems, highlighting their significance in intelligent transportation and autonomous driving technologies.
The incorporation of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) into the healthcare sector has profoundly altered illness detection, medical imaging analysis, clinical decision-making assistance, and patient risk evaluation. Nonetheless, singular AI systems might encounter constraints while processing diverse, noisy, high-dimensional, and multimodal healthcare information. Hybrid Artificial Intelligence models mitigate these constraints by integrating several algorithms, learning methodologies, or specialised knowledge into a cohesive diagnostic structure. These models can amalgamate convolutional neural networks, recurrent neural networks, transformers, ensemble learning, fuzzy logic, optimisation techniques, and clinical expertise to enhance diagnostic precision and resilience. This section offers an extensive examination of hybrid AI frameworks for smart healthcare diagnostics. It examines sources of healthcare data, diagnostic processes, hybrid AI frameworks, medical imaging analysis, electronic health record evaluation, disease forecasting, multimodal diagnostics, interpretable AI, uncertainty assessment, privacy-conscious learning, edge and cloud healthcare infrastructures, and clinical decision-making support.Ultimately, the discourse encompasses nascent research trajectories pertaining to multimodal foundation models, federated learning, digital twins, generative AI, and personalised medicine.
Agentic Artificial Intelligence (Agentic AI) is a novel framework in artificial intelligence that broadens traditional AI functionalities from specialised prediction and content creation to autonomous decision-making and action aligned with specific objectives. An autonomous AI system is capable of sensing its surroundings, analysing data, reasoning about goals, devising strategies, engaging with external instruments, performing actions, assessing results, and modifying its future conduct in response to feedback. Recent advancements in extensive language models, generative AI, reinforcement learning, retrieval-augmented generation, multimodal learning, and tool-utilizing frameworks have expedited the evolution of intelligent autonomous agents. This chapter delineates the principles of Agentic AI and explores its progression from rule-based frameworks, machine learning, and deep learning to generative and autonomous AI structures. The principles outlined in this chapter establish a conceptual framework for comprehending the evolution and practical implementation of intelligent autonomous AI systems.
This chapter delineates the principles, structure, techniques, uses, and prospective trajectories of AI- and ML-driven intelligent healthcare systems. It examines traditional machine learning methods, deep learning frameworks, ensemble techniques, and hybrid predictive models employed in disease forecasting. Prominent domains of application encompass cardiovascular ailments, diabetes, malignancies, neurological conditions, respiratory illnesses, and infectious infections. The section additionally explores healthcare Internet of Things (IoT) innovations, wearable devices, remote surveillance, and interconnected medical systems. Emphasis is placed on data integrity, model clarity, algorithmic prejudice, patient confidentiality, cybersecurity measures, clinical verification, and regulatory factors. A comprehensive approach for constructing illness prediction systems is introduced, encompassing data collection, preprocessing, feature extraction, model training, assessment, and clinical implementation. The chapter wraps up by pinpointing research prospects in multimodal learning, federated learning, foundational models, reliable AI, and tailored predictive healthcare. The accountable incorporation of these technologies might facilitate timely intervention and enhance healthcare efficiency when prediction results are accurately evaluated and interpreted by competent healthcare practitioners.
This section offers an in-depth analysis of deep learning methodologies for identifying plant diseases and pests. It examines the core tenets of convolutional neural networks, transfer learning, object detection, semantic segmentation, vision transformers, and multimodal learning for the purposes of diagnosing plant diseases, categorising symptoms, pinpointing insect pests, and assessing infestation intensity. The section further explores techniques for image capture, agricultural datasets, preprocessing methodologies, model training approaches, evaluation criteria, and deployment frameworks that incorporate mobile devices, edge computing, drones, and Internet of Things systems. Special focus is directed toward the difficulties arising from fluctuations in lighting, intricate backgrounds, overlapping symptoms, insufficient training data, class imbalance, and the disparity between laboratory datasets and actual agricultural settings. A pragmatic methodological structure is suggested for creating a dependable system for identifying plant diseases and pests. Ultimately, the chapter examines elucidable artificial intelligence, self-supervised learning, efficient neural networks, multimodal agricultural intelligence, and ethical implementation. The discourse emphasises how powerful deep learning algorithms may facilitate prompt diagnosis, precise pesticide administration, diminished crop losses, and sustainable farming methods.
Cerebral neoplasms constitute a significant category of neurological ailments necessitating prompt identification, precise categorisation, and suitable clinical evaluation. Medical imaging modalities, including magnetic resonance imaging (MRI), computed tomography (CT), and some positron emission tomography (PET) applications, furnish critical insights regarding tumour positioning, dimensions, shape, and tissue properties. Nonetheless, the manual analysis of medical pictures might prove difficult due to discrepancies in tumour morphology, image resolution, anatomical features, and disease manifestation. Deep learning has surfaced as a viable method for automating specific phases of brain tumour evaluation by acquiring distinguishing representations straight from medical imagery. Convolutional neural networks (CNNs), transfer learning, vision transformers, attention mechanisms, and hybrid architectures have been explored for the purposes of tumour detection, segmentation, and classification. It additionally presents a suggested cohesive framework for the analysis of brain tumour imagery and addresses explainable artificial intelligence, model resilience, data leakage, class imbalance, privacy concerns, and clinical validation. Significant attention is directed toward differentiating research-dataset categorisation from clinically confirmed diagnosis. The chapter wraps up by delineating prospective research avenues encompassing multimodal imaging, self-supervised learning, transformer architectures, federated learning, and elucidative clinical decision support.