Julius Mboli

The transition from paper-based to electronic trade documents (eTDs) represents a critical step in modernising global supply chains; however, the absence of reliable mechanisms to verify the physical movement of goods vis-à-vis the eTD workflow remains a significant challenge. This study proposes the PROGRESS framework, an integrated digital trade architecture that combines AI-driven document verification with real-time physical tracking to ensure end-to-end assurance of consignments. The approach leverages advanced optical character recognition (OCR) techniques, including transformer-based attention models, convolutional recurrent neural networks (CRNNs) with connectionist temporal classification (CTC), and vision transformers (ViTs), enhanced through context-aware validation to ensure semantic consistency across key trade documents, such as commercial invoices, packing lists, air waybills, and certificates of origin. A pilot scenario modelling ~5100 consignments was evaluated and benchmarked against real-world operations for the UK customs report and Teesside International Airport report. Results indicate that the proposed system reduces per-transaction document processing time from 19 min in traditional paper-based systems to approximately 1.8 min, achieving a document verification accuracy of 98.61% (F1 = 98.38%) across a 503-document evaluation corpus, with a demonstrated processing time of 1.78 min per transaction—projecting a theoretical throughput of 33.6 transactions per hour under sustained operation. Furthermore, integration with RFID-based tracking, geofencing, and private 5G connectivity enables continuous monitoring of consignments, increasing the platform throughput from 3.2 to 33.6 consignments per hour. The findings demonstrate that combining digital documentation with verifiable physical assurance significantly enhances transparency, efficiency, and regulatory compliance. This research contributes to the advancement of digital trade by addressing key limitations in current eTD implementations and supporting the effective adoption of the United Kingdom’s Electronic Trade Document Act.

Platelets undergo significant morphological transformations during activation influenced by chemical compounds such as zinc and milrinone. Manual classification methods are subjective, time-consuming, and prone to error. This research aimed to apply and evaluate the efficacy of different pre-trained deep learning Convolution Neural Networks, used for automated binary (treated vs untreated) and multiclass classification (control vs zinc vs milrinone vs combination) of platelet morphological changes induced by these solvents, addressing the limitations of manual classification methods. A dataset of 176 microscopic platelet images from Hull York Medical School was processed, comprising control (n = 43), zinc-treated (n = 45), milrinone-treated (n = 40), and combination treatment (n = 48) samples. Ten pre-trained CNN architectures were evaluated across 166 individual training iterations for both binary and multiclass classification tasks. Progressive optimisation strategies included contrast-limited adaptive histogram equalisation (CLAHE) enhancement, selective layer unfreezing, cross-validation, and hyperparameter tuning. For binary classification, Xception achieved optimal balanced performance (81% accuracy, 76% precision, 74% recall) with proper inter-class discrimination, despite higher numerical metrics (91% accuracy) showing classification bias. For multiclass classification, InceptionV3 and Xception demonstrated superior effectiveness with 72% accuracy. This study demonstrated that Xception and InceptionV3 architectures effectively capture subtle morphological features distinguishing treated platelets, offering reproducible alternatives to subjective manual methods.

Platelets undergo significant morphological transformations during activation influenced by chemical compounds such as zinc and milrinone. Manual classification methods are subjective, time-consuming, and prone to error. This research aimed to apply and evaluate the efficacy of different pre-trained deep learning Convolution Neural Networks, used for automated binary (treated vs. untreated) and multiclass classification (control vs. zinc vs. milrinone vs. combination) of platelet morphological changes induced by these solvents, addressing the limitations of manual classification methods. A dataset of 176 microscopic platelet images from Hull York Medical School was processed, comprising control (n=43), zinc-treated (n=45), milrinone-treated (n=40), and combination treatment (n=48) samples. Ten pre-trained CNN architectures were evaluated across 166 individual training iterations for both binary and multiclass classification tasks. Progressive optimisation strategies included contrast-limited adaptive histogram equalisation (CLAHE) enhancement, selective layer unfreezing, cross-validation, and hyperparameter tuning. For binary classification, Xception achieved optimal balanced performance (81% accuracy, 76% precision, 74% recall) with proper inter-class discrimination, despite higher numerical metrics (91% accuracy) showing classification bias. For multiclass classification, InceptionV3 and Xception demonstrated superior effectiveness with 72% accuracy. This study demonstrated that Xception and InceptionV3 architectures effectively capture subtle morphological features distinguishing treated platelets, offering reproducible alternatives to subjective manual methods.

The transition from paper-based to electronic trade documents (eTDs) represents a critical step in modernising global supply chains; however, the absence of reliable mechanisms to verify the physical movement of goods vis-a-vis the eTD workflow remains a significant challenge. This study proposes the PROGRESS framework, an integrated digital trade architecture that combines AI-driven document verification with real-time physical tracking to ensure end-to-end assurance of consignments. The approach leverages advanced Optical Character Recognition (OCR) techniques, including Transformer-based attention models, Convolutional Recurrent Neural Networks (CRNN) with Connectionist Temporal Classification (CTC), and Vision Transformers (ViT), enhanced through context-aware validation to ensure semantic consistency across key trade documents such as commercial invoices, packing lists, airway bills, and certificates of origin. A simulation involving 5,100 consignments was conducted and benchmarked against real-world operations at Teesside International Airport. Results indicate that the proposed system reduces document processing time from 24–72 hours in traditional paper-based systems to approximately 1–2 minutes, achieving a validation accuracy of 98%. Furthermore, integration with RFID-based tracking, geofencing, and private 5G connectivity enables continuous monitoring of consignments, reducing customs dwell times from 72 to 44 hours and increasing throughput by up to 65%. The findings demonstrate that combining digital documentation with verifiable physical assurance significantly enhances transparency, efficiency, and regulatory compliance. This research contributes to the advancement of digital trade by addressing key limitations in current eTD implementations and supporting the effective adoption of the United Kingdom’s Electronic Trade Document Act.

The grading of trading cards and collectibles remains a labour-intensive and subjective task, traditionally dependent on expert visual inspection. This paper investigates the integration of Artificial Intelligence (AI) into automated grading systems, with a focus on balancing accuracy, scalability and interpretability. We compare conventional image-processing methods based on OpenCV with modern Convolutional Neural Network (CNN) models and propose a hybrid framework that incorporates YOLOv8 with Oriented Bounding Box (OBB) support for localised defect detection. The system operates through a six-stage pipeline encompassing centring, corner wear, edge integrity, surface quality and holistic scoring. To support reproducibility and consistency, we developed a robotic imaging platform capable of capturing standardised high-resolution images (6000 × 4000 pixels) under controlled lighting conditions, enhanced by Flat Field Correction (FFC) for illumination normalisation. Using this platform, we constructed a large-scale dataset comprising over 43,000 training samples after augmentation, covering a wide range of real-world card conditions. Experimental evaluation using localisation accuracy, region-level grading agreement and professional grading consistency metrics demonstrates that CNN-based classification approaches significantly outperform traditional OpenCV pipelines in predictive accuracy but lack spatial interpretability. In contrast, the proposed hybrid framework achieves improved defect localisation and grading consistency relative to CNN-only baselines, while providing transparent visual explanations through YOLO-OBB detections. The integration of deterministic image analysis with deep learning enables both quantitative subgrades and interpretable defect localisation, offering an explainable and scalable solution for automated collectible card grading.

Efficient waste sorting is crucial for enabling circular-economy practices and resource recovery in smart cities. This paper evaluates both traditional machine-learning (Random Forest, SVM, AdaBoost) and deep-learning techniques including custom CNNs, VGG16, ResNet50, and three transfer-learning models (DenseNet121, EfficientNetB0, InceptionV3) for binary classification of 25 077 waste images (80/20 train/test split, augmented and resized to 150x150 px). The paper assesses the impact of Principal Component Analysis for dimensionality reduction on traditional models. DenseNet121 achieved the highest accuracy (91 %) and ROC-AUC (0.98), outperforming the best traditional classifier by 20 pp. Principal Component Analysis (PCA) showed negligible benefit for classical methods, whereas transfer learning substantially improved performance under limited-data conditions. Finally, we outline how these models integrate into a real-time Data-Driven Decision Support System for automated waste sorting, highlighting potential reductions in landfill use and lifecycle environmental impacts.)

Efficient waste sorting is crucial for enabling circular-economy practices and resource recovery in smart cities. This paper evaluates both traditional machine-learning (Random Forest, SVM, AdaBoost) and deep-learning techniques including custom CNNs, VGG16, ResNet50, and three transfer-learning models (DenseNet121, EfficientNetB0, InceptionV3) for binary classification of 25077 waste images ($80 / 20$ train/test split, augmented and resized to $150 \times 150 \text{px}$). The paper assesses the impact of Principal Component Analysis for dimensionality reduction on traditional models. DenseNet121 achieved the highest accuracy (91 %) and ROC-AUC (0.98), outperforming the best traditional classifier by 20 pp. Principal Component Analysis (PCA) showed negligible benefit for classical methods, whereas transfer learning substantially improved performance under limited-data conditions. Finally, we outline how these models integrate into a real-time Data-Driven Decision Support System for automated waste sorting, highlighting potential reductions in landfill use and lifecycle environmental impacts.)

The traditional linear economy using a take‐make‐dispose model is resource intensive and has adverse environmental impacts. Circular economy (CE) which is regenerative and restorative by design is recommended as the business model for resource efficiency. While there is a need for businesses and organisations to switch from linear to CE, there are several challenges that needs addressing such as business models and the criticism of CE projects often being small scale. Technology can be an enabler toward scaling up CE; however, the prime challenge is to identify technologies that can allow predicting, tracking and proactively monitoring product's residual value to motivate businesses to pursue circularity decisions. In this paper, we propose an IoT‐enabled decision support system (DSS) for CE business model that effectively allows tracking, monitoring, and analysing products in real time with the focus on residual value. The business model is implemented using an ontological model. This model is complemented by a semantic decision support system. The semantic ontological model, first of its kind, is evaluated for technical compliance. We applied DSS and the ontological model in a real‐world use case and demonstrate viability and applicability of our approach.