This work is part of a research project carried out during the COVID-19 pandemic, involving the design and realization of an autonomous mobile hospital robot. Many real-world robotic tasks suffer from the critical characteristics: Noisy sensing, imperfect control, and environment changes. The Markov decision process MDP and its variants provide a mathematically based framework for modeling and solving robot decision and control tasks under uncertainty. This paper presents a review of Markov Decision Processes (MDPs) and their variants in the Control of Robotic systems. We begin by introducing the basic concepts of MDPs and their algorithms for solving completely observable decision problems, including value iteration and policy iteration. We then discuss the challenges associated with partially observable decision problems POMDPs. We review various approaches for solving POMDPs, including belief-state planning and Monte Carlo tree search. Finally, we discuss the concept of POMDP augmentation, which involves incorporating additional information into the decision-making process to improve performance. We present several examples of POMDP augmentation techniques, including the use of deep neural networks and transfer learning.
The research introduces a data sharing and retrieval method within the blockchain framework tailored for smart healthcare systems. This method addresses key challenges in data management, security, and accessibility. Two main components are presented: a distributed, parallel data-sharing scheme and an innovative data retrieval algorithm using the Merkle-Patricia Trie (MPT) and Bloom filters. The proposed healthcare blockchain system showcased improved performance metrics with notable efficiency in upload and download times, latency, delay, and response times. This adaptable approach promises enhanced patient care, medical research, and data analytics, emphasising effective, secure, and patient-centric healthcare data management.
Deep learning object recognition models, which are widely used in computer vision, may provide an opportunity to accurately recognize fruit trees. This is essential for fast data collection, selection and reducing human operational errors. This paper proposes a YOLOv5-based detection model for fruit tree detection in the farm plantation using UAV-collected data. This proposed model detects individual fruit trees from the agriculture field and also provide counts how many trees are detected. An image dataset was created from the publicly available UAV captured data which contains total 36 images. Among them 27 were used for training and 9 for testing the proposed model. Four different YOLOv5 scales for object recognition (YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5x) were selected for training, validation, and testing on image datasets.
The recent surge in monkeypox (mpox) cases across various nations has escalated into a significant public health challenge, underscoring the imperative for timely detection and diagnosis. In light of this, our research focused on identifying the most efficient deep learning model tailored for mpox detection. Clinically, the progression of monkeypox is characterized by four distinct stages: macular, papular, vesicular, and pustular. In this study, we explore the viability of leveraging a Deep Convolutional Generative Adversarial Network (DCGAN) in conjunction with a conditional vector to strengthen the diagnostic precision of monkeypox. By utilising the remarkable capabilities of GANs, we generate synthetic images mirroring monkeypox skin lesions, thereby amplifying our limited dataset. This approach of utilizing a conditional DCGAN, when paired with stage-specific conditions, results in a significant boost in classification accuracy, elevating it from 0.7532 to 0.8734, complemented by precision 0.91, recall 0.878 F1-score 0.8936. These promising results underscore the advantages of deploying GANs for data augmentation in biomedical image classification tasks, when compared with well known classification models. These models assures the potential of deep learning-assisted diagnostics in dermatology. Especially in therapeutic contexts, where real data may be sparse, the addition of synthetic images can act as a valuable resource, paving the way for enhanced diagnostic instruments.
This paper addresses the labor-intensive and wasteful nature of the traditional sugarcane bud cutting method. To overcome these challenges, the proposed approach leverages YOLOv5 technology for sugarcane bud identification. The machine-learning model is trained with diverse data samples, enabling it to accurately distinguish between sugarcane buds and other elements in the images. The implementation yields the best of 79
Vehicle-to-Vehicle communication (V2V-C) is a cutting-edge technology in the field of 6G networks that improves spectrum utilization and energy efficiency (EE). Despite the potential benefits, there are some considerable difficulties with V2V-C, such as cross-channel interference, co-channel interference and the demand for huge connectivity. To address these challenges, researchers have turned to simultaneous transmitting and reflecting reconfigurable intelligent surfaces (STAR-IRSs) as auxiliary devices to improve wireless network performance. These surfaces allow users on opposite sides to be served at the same time by sending and reflecting signals. However, the existing solution has been limited to either continuous or discrete spaces, limiting optimisation parameters to either continuous or discrete nature. To address these limitations, the proposed scheme use a hybrid space to optimise the EE of the network for the downlink STAR-IRS aided communication system in the presence of vehicle-to-vehicle pairs (V2VPs), allowing one parameter to be continuous and the other to be discrete. In this research work, the proposed scheme uses the parameterized deep Q-network (P-DQN) framework for estimating the beamforming vector and phase shift for EE optimisation. The results from the simulation demonstrate the efficacy of the system proposed by maximising the spectrum usage and energy efficiency.
In this innovative study, multi-task transfer study and Natural Language Processing or NLP join forces to fight the ever-growing challenge of identifying fake news. By simultaneously training a model on an array of related tasks sentiment analysis, language modeling, and fake news detection, it unlocks the potential to deeply comprehend natural language structures and patterns. This not only bolsters accuracy and effectiveness but also unlocks a treasure trove of advantages, including enhanced precision, efficiency, and adaptability to new data. Fearlessly pitted against fake-news detection tactics, the proposed method emerges victorious, showcasing superior accuracy and agility in computation times. Tested on a battlefield of news articles, it demonstrates the power of multi-task transfer learning and NLP in the fight against misinformation. In the grand scheme of things, this study offers a novel and creative approach, revolutionizing the study of identifying fake-news, and paving the way for diverse applications, from social media monitoring to news filtering and beyond. In the vast, interconnected world of NLP, this groundbreaking method emerges as a beacon of hope, guiding us towards a future where truth prevails, and fake news is vanquished.
This research offers an in-depth comparative analysis of various pre-trained Convolutional Neural Network (CNN) models such as VGG16, ResNet50, InceptionV3, MobileNetV2, and Xception to predict stock market trends. Our approach involves the conversion of time-series financial data into 2D image-like structures through the application of two distinct techniques: the Gramian Angular Field (GAF) and the Markov Transition Field (MTF). By applying this transformation, we leverage the power of CNNs. We utilize the ideas of transfer learning and try to evaluate the performance of each model using several measures including predictive accuracy, precision, recall, F1-score, and computational efficiency. The analysis highlights the unique advantages and limitations of each model, thereby offering valuable insights into their suitability for stock market prediction tasks. This study is a significant contribution to the current body of literature on financial time series forecasting, providing a novel perspective on using pre-trained CNN models in the Indian Financial Sector. It carries important implications for future work and practitioners in the finance and investment sectors, offering a tool for more e-market predictions.
Due to the rapid growth and evolution of smart devices like smartphones and cameras, a huge amount of digital data is generated in the form of digital images. Digital images are foundational pillars of data because they are a reliable source of information because of their visual appeal and information. Modern software and technologies have opened the doors for new and creative ways to forge or tamper images. Digital image forgery means manipulating the digital image to suppress some meaningful and factual information inside the image or misguide any concerned organization. The detection of forged images is inspired by the requirement for authenticity and integrity maintenance. Researchers have used Deep Learning (DL) techniques for the detection of image tampering and forgeries. This paper proposes an image forgery detection method based on Error Level Analysis (ELA) and Convolutional Neural Network (CNN). ELA is an image processing technique used for detecting inconsistencies and potential manipulation using compression artefacts of images with lossy compression. CNN are a class of neural networks specialized for their superior performance with images. The proposed method uses ELA, which is pipelined to a CNN model. The analysis is performed on the standard CASIAv2 dataset which consists of 7491 authentic images and 5123 forged images. The proposed method attains a superior accuracy of 94
The ability to recognise and interpret emotional expressions is crucial since emotions play a significant role in our daily lives. Emotions are multifaceted phenomena that affect our behavior, perception, and cognition. As a result, numerous machine-learning and deep-learning algorithms for emotion analysis have been studied in previous works. Finding emotion in an obscured face, such as one covered by a scarf or hidden in shadow, is considerably harder than in a complete face, though. This study explores the effectiveness of deep learning models in occluded facial emotion analysis through a transfer learning approach. The performance of two individual pre-trained models, MobileNetV2 and EfficientNetB3, is compared alongside a hybrid model that combines both approaches. This comparison is conducted using the FER-2013 dataset. The dataset consists of 35,887 images and categorizes emotions into seven emotional categories. The results indicate that the hybrid model attained the highest accuracy, with a score of 93.04
Machine learning has revolutionized research by extracting complicated patterns from complex data, particularly in healthcare and medical imaging, where accurate diagnosis is critical. The concept of federated learning has gained popularity in the field of machine learning as a viable technique for addressing privacy issues in distributed settings. This research explores federated learning in healthcare, demonstrating its capability to achieve results comparable to centralized data while enhancing the accuracy of deep learning models for clinical data interpretation. To ensure reliable model performance during federated learning rounds, this study introduces a proactive mechanism for coordinating server updates with equitable client modifications. The equitable model, designed to reduce accuracy fluctuations, consistently improves accuracy across multiple training rounds on a non-IID dataset. We achieved smooth accuracy improvement by implementing the novel Equitable model, resulting in robust model development. As healthcare AI continues to advance, federated learning emerges as a critical tool for developing precise prediction models while preserving patient data privacy and aligning with increasingly strict data standards worldwide, such as GDPR regulations. This strategic approach not only promotes ethical, efficient, and secure progress in medical research and practice, but it also emphasizes the importance of protecting patient data privacy while utilizing machine learning’s potential.
Accurate classification of diseases in potato crops is vital for optimizing yield and ensuring crop health. We propose a generalized framework using Federated Learning (FL) for accurate classification of potato crop diseases. The dataset from Plant Village includes diverse potato leaf images with imbalanced class distributions. By incorporating FL, which enables collaborative model training without sharing raw data, we leverage the collective intelligence of distributed datasets while ensuring privacy. CNN as base model, achieves a 92
The viability of biomass to be used as a consumer product relies heavily on the cost of a Biomass supply network (BSN) that links biomass producers with biorefineries and, finally, with end customers. The current study aims to establish a cost optimization model to minimize the financial burden of BSN. A MILP model has been established and implemented to reduce the costs of a BSN. A comparatively lesser-used stochastic technique, Ant Colony Optimization (ACO), has been used in the present paper to minimize the cost of BSN. Although the ACO technique has succeeded in other settings, it is seldom tested in the context of BSN. The results from the ACO approach have been compared with another popular stochastic optimization technique called the Non-sorting Genetic Algorithm (NSGA-II). According to empirical research, the ACO approach is the most cost-effective optimization technique to lower BSN-related costs. The management may use the blueprint of the optimization model and techniques to develop cost-cutting measures for BSN.
In the context of federated learning, the concept of federated unlearning has emerged, aiming to realize the "right to be forgotten". The current research primarily focuses on designing unlearning techniques for clients "right to be forgotten", it has often bypassed to consider the server's authority to discard the client contribution without taking any consent from participating clients, we named it "right to forget". These client contributions may contain adverse effects that could significantly impact global aggregation. In this research paper, we conduct a comprehensive review of previous studies related to federated unlearning and explore the server "right to forget" client's contributions. We also introduce new taxonomies to classify and summarize the latest advancements in federated unlearning algorithms. Moreover, we take the first step to present the server right to forget (SRF), a novel unlearning methodology that enables the server to remove unreliable client contributions to improve global model accuracy. Experiments on two different kinds of datasets and models demonstrate the effectiveness of our method. We envision our effort as a first step toward the server's right to forget the client's contribution in the context of federated unlearning toward adherence to legal and ethical standards in a just and transparent manner.
This study primarily focuses on a novel approach to Covid-19 prediction utilizing X-ray images. The images are used for the initial stage of training of the CNN Convolution neural network model. For improved classification and prediction accuracy, images are trained and tested using a hybrid GANs (Generative Adversarial Networks based Convolution neural network) - CNN (Convulational Neural Network) model. The noise cancellation technique of image processing has been used to minimize the noise in images and used for the GANs-CNN hybrid model. Each method of the proposed model has resulted in better accuracy, in which the validation accuracy on every 15 epochs is 79.2
Blockchain is a shared database that makes use of a shared ledger distributed amongst the different nodes present in a network. It provides the foundation for storing the data in such a way that it can neither be deleted nor altered. Blockchain has been widely used in various application areas ranging from banking sectors to finance, health care to supply chains and many more, thus providing better accuracy, transparency, and cost reduction. Despite exhibiting features like decentralization, trust, and immutability, blockchain is susceptible to many security attacks. This paper focuses on the concept of blockchain, providing insights regarding its features and application areas. Most importantly this paper covers the analysis of various security attacks to which blockchain is prone and hence many sectors are reluctant to use this technology. Some future directions and key takeaways are covered in the conclusion section that will form the basis for future work.
Alzheimer's disease (AD) is a debilitating, irreversible neurological condition that leads to a severe decline in patients' health, often resulting in fatality. Detecting AD and its early stages, such as mild cognitive impairment (MCI), which can manifest as stable (sMCI) or progressing (pMCI), is crucial for effective treatment planning and personalized therapy. Recent advancements in noninvasive retinal imaging technologies, including Optical Coherence Tomography (OCT), OCT angiography, and digital retinal photography, have enabled the examination of the neuronal and vascular structure of the retina in AD patients. Furthermore, the development of computer algorithms tailored to these imaging techniques has significantly enhanced AD research. This paper presents a comprehensive study on early AD identification that leverages state-of-the-art deep learning techniques and medical images or scans. It also explains the potential benefits of using emerging retinal scans for enhanced detection. It also explains various deep learning techniques that harness both local and global features to enhance accuracy by utilizing extensive scan data.
Deep learning has become an effective approach over the past few years to addressing intricate computer vision problems, and Convolutional Neural Networks (CNNs) have been the primary driving force behind this progress. Developing CNNs, however, comes with the obstacle of requiring huge, labeled datasets. Gathering and annotating a large dataset for any specific job is costly and time-consuming. To overcome this challenge, researchers can employ transfer learning, a technique that involves using pre-trained deep learning models on extensive datasets. This study primarily aims to investigate the application of various transfer learning methods in conjunction with Deep Convolutional Neural Networks (CNNs) for image classification. The research utilizes the Visual Object Classes Challenge 2012 (VOC2012) dataset as the foundation for its analysis. To classify a diverse range of object images, the study applies well-established transfer learning techniques, specifically fine-tuning pre-trained CNN models. Model performance is assessed through metrics like FPS (frames per second) and mAP
To assure the safety and efficiency of components used in aviation industry, the area of aerospace business process validation relies on stringent protocols. The qualification process is multifaceted as it entails multiple phases and application of Image processing and pattern recognition techniques in analysing and assessing component quality and integrity according to industry standards. This paper introduces an approach that integrates user interface and UX design principles to optimize the processes for enhanced traceability and progression tracking in qualification lifecycle and the outcomes of image processing and pattern recognition applications. The goal is to transform algorithmic outcomes into intelligible visual representations, promoting informed decision-making by combining agile approaches with user-centred design concepts for AddQual Qualification process. AddQual is a UK based company that specializes in providing advanced quality solutions for the aerospace industry, offering expertise in material testing and component qualification. This project illustrates the successful implementation of through a series of case studies. It results in an average 86
Aims: In the modern era, substance abuse is a global problem. CDC’s (Centre for Disease Control) National Center for Health Statistics reported in July 2021 that, more than 93000 drug overdose deaths occurred alone in the United States. Design: A cohort study. Setting: The drug consumption data is collected online in United Kingdom. Total 2000 records are present, out of which, 1885 records are considered for the work after initial data processing. The dataset has the drug consumption output outcome of 18 drugs, heroin is one of them. Measurements: In this work, an intelligent approach has been proposed for detecting abuse of one of the most illicit substances - heroin. A random forest-based machine learning model is proposed which can predict heroin abused individuals with very high accuracy. For the abuse prediction, various supervised machine learning methods are applied, and their performance is compared. These algorithms are applied to different sets of features. Among the applied five algorithms, feature importance score is calculated for logistic regression, ensemble learning (gradient boosting) and random forest. The feature importance score for each feature is calculated for all applied algorithms k-nearest neighbour and naïve bayes. The ranking of all features is done based on the obtained score. Findings: It is found in the study that the heroin-abuse dataset collected from UCI, two-classification based on random forest (RF) and gradient boosting (GB) achieved more than 90
