Procter & Gamble (United Kingdom)
Open Access Institutional Repository at Robert Gordon UniversityNot a member yet
9563 research outputs found
Sort by
Hyperspectral image classification using a multi-scale CNN architecture with asymmetric convolutions from small to large kernels.
Get PDFDeep learning-based hyperspectral image (HSI) classification methods, such as Transformers and Mambas, have attracted considerable attention. However, several challenges persist, e.g., (1) Transformers suffer from quadratic computational complexity due to the self-attention mechanism; and (2) both the local and global feature extraction capabilities of large kernel convolutional neural networks (LKCNNs) need to be enhanced. To address these limitations, we introduce a multi-scale large kernel asymmetric CNN (MSLKACNN) with the large kernel sizes as large as 1×17 and 17×1 for HSI classification. MSLKACNN comprises a spectral feature extraction module (SFEM) and a multi-scale large kernel asymmetric convolution (MSLKAC). Specifically, the SFEM is first utilized to suppress noise, reduce spectral bands, and capture spectral features. Then, MSLKAC, with a large receptive field, joins two parallel multi-scale asymmetric convolution components to extract both local and global spatial features: (C1) a multi-scale large kernel asymmetric depthwise convolution (MLKADC) is designed to capture short-range, middle-range, and long-range spatial features; and (C2) a multi-scale asymmetric dilated depthwise convolution (MADDC) is proposed to aggregate the spatial features between pixels across diverse distances. Extensive experimental results on four widely used HSI datasets show that the proposed MSLKACNN significantly outperforms ten state-of-the-art methods, with overall accuracy (OA) gains ranging from 4.93% to 17.80% on Indian Pines, 2.09% to 15.86% on Botswana, 0.67% to 13.33% on Houston 2013, and 2.20% to 24.33% on LongKou. These results validate the effectiveness of the proposed MSLKACNN
Single-atom nanozymes for electrochemical sensing oxidative stress biomarkers.
No full textDeveloping reliable and low-cost sensors for oxidative stress biomarkers has gained significant interest for understanding and managing chronic diseases. Recently, single-atom nanozymes (SANs) with atomically dispersed metal sites and unique metal–nitrogen–carbon (M–N–C) structures have emerged as promising redox enzyme mimetics, offering superior catalytic efficiency and specificity, and with exceptional sensing capabilities for oxidative stress biomarkers such as hydrogen peroxide (H2O2), nitric oxide (NO), glutathione (GSH), and uric acid (UA). This review discusses key progress, challenges, and future directions in SANs for electrochemical determination of oxidative stress biomarkers, aiming to guide future research toward practical and impactful solutions in healthcare and disease management
Towards ordinal few-shot learning for automated essay grading.
No full textOrdinal essay grading, an essential task in educational domain and natural language processing (NLP), involves categorising essays based on quality, such as grading scale levels. This is crucial in automated assessment systems that evaluate student writing and provide feedback on aspects like coherence, argumentation, and language proficiency. However, challenges arise from limited data resources, such as when a new assessment is introduced and no data is available to train algorithms, as well as the complexity of essay structures in real-world grading scenarios. This research explores the use of few-shot learning, a technique that learns from a limited number of labeled examples, to address these challenges in ordinal essay grading. By leveraging few-shot learning's ability to identify class representations from minimal examples, we aim to mitigate data scarcity in essay grading. With the rise of Large Language Models (LLMs), we seek to improve few-shot prompting performance by introducing novel strategies for example selection, enhancing class representation in the demonstrations provided to prompts. Finally, we aim to apply prototypical methods to agents architectures for agent selection on the basis of similarity weighted by ordinal class knowledge
Attention-based framework for automated symbol recognition and wiring design in electrical diagrams.
Get PDFThe digitization of electrical diagrams plays a crucial role in modern construction industries, enabling efficient reuse, seamless distribution, and accurate archiving. Despite technological advances, many of these diagrams remain in undigitized formats, leading to labor-intensive manual analysis for tasks such as cost estimation and wiring design. These challenges are aggravated by the diversity of symbols, high inter-class similarities, and the inherent complexities of wiring layouts, which require advanced recognition and efficient wiring design. This paper presents a deep learning framework that integrates an attention mechanism for symbol recognition, followed by a graph-based algorithm for fully automated wiring design. Through comparative evaluation, Efficient Channel Attention emerged as the most effective attention module, improving the mean average precision by 3.2%. The wiring algorithm leverages an improved pathfinding approach that reduces bends and total wiring length by 43% while adhering to boundary constraints and electrical rules. Extensive experiments on proprietary and public datasets demonstrate that the proposed framework significantly improves the recognition of complex electrical symbols, outperforming the baseline model. This research sets a new benchmark for automating electrical diagram analysis, offering substantial cost savings while reducing the manual effort associated with large-scale construction projects
Participants or pretenders? Addressing the challenge of inauthentic participation encountered during three social research studies on experiences of food insecurity in the UK.
No full textIndividuals participate in research for numerous reasons; however, the global economic downturn may have driven some to participate solely for monetary recompense. While inauthentic participation is more widely recognised in quantitative survey studies, it is increasingly becoming an issue in qualitative research. Drawing on our experiences and supported by the wider literature, we highlight ways in which inauthentic participation can be identified and addressed. We argue it is pertinent that researchers are aware of the risks and potential impact of inauthentic participants and recommend researchers consider this phenomenon from study planning stages onwards. We identify institutions, including universities, as well-placed to provide training and to ensure, where necessary, mitigation plans are in place. We suggest ethics committees, funders and publishers request inauthentic participation be considered and reported. These recommendations would establish awareness of this phenomenon, prevent wasting valuable project resources, increase transparency of reporting and ensure data integrity is protected
Cross-calibration of multiple optical instruments for the measurement of particle size distributions in water.
Get PDFParticle size spectra, describing particle abundance as a function of size, are essential for understanding marine ecosystem structure and biogeochemical processes. The slopes of size spectra provide insights into ecosystem characteristics. However, capturing size spectrum across the wide range of particle sizes requires integrating multiple imaging systems, as no single instrument spans the entire size spectrum of diverse particle types. This study employed three imaging systems - Underwater Vision Profiler 5 (UVP5), the Continuous Plankton Imaging and Classification Sensor (CPICS) and LISST-HOLO2 - each with a distinct size resolution, to construct a continuous particle size spectrum spanning a broad size range. Calibration experiments were carried out using olive stone granules (0 -1000 µm) divided into five size classes to ensure size spectra obtained from each instrument are directly comparable. Four binarization methods were evaluated for particle edge detection to measure particle sizes, with results highlighting method-specific biases. Otsu thresholding underestimated sizes for low-contrast particles, while Canny thresholding overestimated sizes for interference-affected particles. Optimal methods were selected for each instrument, enabling alignment of size spectra across systems. The study demonstrates that integrating imaging systems and applying appropriate data processing methods can effectively generate size spectra over broad size ranges. This approach provides a robust framework for studying particle-driven processes and carbon cycling in marine ecosystems, emphasizing the importance of careful thresholding and visual validation
Deep learning-based direction of arrival estimation for underwater acoustic sources using BPSK signals and multiple array geometries.
Get PDFAccurate direction of arrival (DoA) estimation is critical for underwater acoustic applications such as target localization, communication, and environmental monitoring. The performance of DoA estimation algorithms strongly depends on the geometry of the sensor array used. This paper presents a comprehensive performance comparison of the commonly used array configurations linear and circular arrays for underwater DoA estimation. The study simulates five acoustic sources in a multipath environment and analyzes received signals at two different array geometries with eight hydrophone elements. To enhance source localization, study employs a Convolutional Neural Network (CNN) trained on received time-domain signals. The CNN model achieves an average DoA estimation accuracy of 2.5° RMSE with circular arrays and 4.2° RMSE with linear arrays across all sources. The model was tested under varying signal-tonoise ratios (SNRs) from 0 dB to 20 dB. At 10 dB SNR, the circular array maintained 90% accuracy (within ± 5°), while the linear array dropped to 76% accuracy, demonstrating higher robustness of the circular configuration in noisy environments. In terms of angular resolution, the circular array resolved sources separated by as little as 8°, compared to 12° for the linear array. However, the linear array required 25% less training time and approximately 30% lower computational cost during inference due to its simpler geometry and reduced spatial diversity
Using the medical treatment contract as an instrument for replacing the tort liability system in England.
No full textIn modern medical law, there is much criticism of the existing medical liability system in England, which is based on the duty of care and breach of that duty (negligent behaviour). The duty of care plays a crucial role as its violation is the foundation for establishing negligent behaviour. It is set through general principles, whose content cannot effectively encompass the diverse types of relationships between a patient and the healthcare provider. The main participants in the relationship built around medical treatment are not satisfied with the existing system. Patients do not feel that the system works for them. They are struggling with a lack of information and complicated requirements of proof in litigation, especially with demonstrating whether the doctor acted negligently. This further leads to long litigation and high costs. Moreover, courts do not interpret negligent behaviour as one of the conditions for medical liability, in the same way in similar situations, which causes different outcomes and lack of reliability. Additionally, doctors fight stigmatisation in the case of mistakes because the focus is on their potentially negligent actions during the medical treatment provision. They answer this challenge by practising defensive medicine. Despite defensive behaviour, the number of litigations constantly rises, which, together with defensive medicine, induces significant costs for the NHS and society. As one of the possible solutions for the problems described above, this work appraised a new balanced contractual model for England that should effectively regulate the relationship between a healthcare provider and a patient, including potential disputes between them. The new model was created through an analysis and comparative research of case law, discussions of legal and socio-legal concepts found in academic literature, and interpretation of legal rules. This resulted in a combination of the standard term framework agreement as a roof structure and particular contracts, both with a relational nature, centred in most situations around the obligation to achieve a particular result ('fit for purpose, obligation) and exemption clauses. Further, this research examined how the proposed model can effectively resolve the problems identified in English jurisdiction. Then, the solutions proposed in the new contractual model were compared with the legal framework in England to consider how its approach fits into English law and to consider if it can apply universally to all patients and healthcare providers. Finally, the developed model was compared with alternative solutions such as tort liability, no-fault liability and the classic contract. The conclusion is that a new contractual model created through this work offers a mixture of predictability, certainty, balance, efficiency, and adaptability to different situations. Based on that, this research proposed universal model, applicable to all patient-healthcare provider relationships, established on mutuality in obligations and rights, delivering quality care, and achieving the expected goals, which provides necessary efficiency in the case of establishing liability, to be adopted in English law as a replacement for the existing system
Nature-based algorithms for deep learning based systems and applications.
No full textDeep Learning Based Systems (DLBS), characterized by their layered processing, in-model feature transformation, and high complexity, have revolutionized problem-solving across numerous domains. However, the manual design of optimal DLBS architectures is prohibitively time-consuming and resource-intensive. Nature-Based Algorithms (NBA), inspired by natural and biological processes, present a promising solution for automating this optimisation due to their ability to handle non-differentiable, discontinuous, and multi-modal problems. This research systematically addresses key challenges in applying NBA to optimise DLBS across distinct problem types. First, addressing the optimisation of complex DLBS for tabular data classification, we developed the MUlti-Layer heterogeneous Ensemble System (MULES) and the COnnection framework for Multi-layer Ensemble (COME). MULES introduces a novel NBA approach using NSGA-II to simultaneously select optimal classifiers and features at each layer of a DLBS. COME pioneers an NBA-driven framework to discover optimal inter-layer connections between classifiers within DLBS, moving beyond fixed input structures for subsequent layers. Turning to the critical challenge of computational expense in NBA for DLBS, particularly with variable-length architectural encodings, we proposed a novel LSTM-based encoder-decoder surrogate model integrated within a Surrogate-Assisted Evolutionary Algorithm (SAEA) framework. This innovation effectively bridges the gap between the variable-length encodings essential for representing diverse CNN architectures and the fixed-length input requirements of standard surrogate models. Finally, focusing on the application of NBA-optimised DLBS for medical image segmentation, we made two major contributions. We developed a weighted ensemble framework where optimal weights for combining predictions from diverse deep segmentation models are determined efficiently using NBA, maximising the Dice coefficient. Furthermore, we introduced a surrogate-based Optimal Decision Template method (ODTwS) that leverages surrogate models to drastically reduce the cost of optimising decision templates. The algorithms and frameworks developed throughout this thesis will be made publicly available, which would provide robust tools to support both academia and industry
Design and fabrication of a simple device for folding towel.
Get PDFElectronic technology makes work more manageable, and various electronic gadgets have been invented to assist humans. One of the most time-consuming activities is household chores, such as laundry. Daily laundry tasks are easier to manage with the help of washing and drying machines. However, the folding task is still done by hand and is not automated. A towel is one of the main clothes in daily life. Commonly, people use bare hands to fold the towel. However, this task consumes much time and energy; consequently, boredom, tiredness, and fatigue occur. The existing device to fold towels usually has a large physical compartment and is mainly used for industrial purposes, such as at hotels or laundry shops. The towel-folding machine is in increasing demand in our daily lives. Therefore, this project was developed and designed to eliminate the tedious folding of towels. The main aim is to design and develop an effective mechanism for folding rectangular towels using electronic components. The other objective is to compare the timing of folding rectangular towels using a simple device and by hand. As a result, the prototype represented a semi-automation system incorporating mechanical and electronic designs. The prototype assembly has a folding board made from polypropylene plastic. The amount of time to fold one towel using a semiautomatic folding board remains the same throughout the process, while the amount of time required to fold towels by hand increases. In conclusion, a prototype was designed and developed successfully with various electronic components: HC-SR04 ultrasonic sensor, MG996R servomotor and Arduino. Besides, by comparing the timing of folding rectangular towels using a simple device and by hand, 94 seconds was reduced when folding 50 sheets with the aid of the device, compared to by hand. In other words, the effectiveness of the device is 20%
Data sources:
Access Repository Dashboard
Do you manage Open Access Institutional Repository at Robert Gordon University? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!