Selected work

Representative Papers

HADS-Net:A Hybrid Attention-Augmented Dual-Stream Network with Physics-Informed Augmentation for Breast Ultrasound Image Classification

May 19, 2026

This study addresses the challenge of classifying benign, malignant, and normal tissues in breast ultrasound images, which is hindered by speckle noise, acoustic shadowing, and inter-class visual ambiguity. To tackle this, the authors propose a novel dual-stream network architecture that integrates a local stream specifically designed for lesion boundary refinement with a physics-informed global texture stream. Feature fusion and joint optimization are achieved through a cross-attention mechanism. Built upon an EfficientNet-B3 backbone, the model incorporates Sobel edge extraction, physics-aware data augmentation, and an adaptive class-weighted Focal Loss. Evaluated on the BUSI dataset, the method achieves 96.58% accuracy, a macro ROC-AUC of 0.9978, and a macro F1-score of 0.9654, with zero malignant lesions misclassified as normal—demonstrating substantially improved diagnostic reliability.

0 citationsRead paper

an interpretable vision transformer framework for automated brain tumor classification

Apr 23, 2026

This study addresses the challenges of time-consuming and subjective manual diagnosis of brain tumors in MRI by proposing an interpretable four-class classification method based on Vision Transformer (ViT-B/16) to automatically distinguish glioma, meningioma, pituitary tumor, and healthy tissue. The approach incorporates clinically oriented preprocessing with CLAHE enhancement, two-stage fine-tuning, hybrid MixUp/CutMix augmentation, exponential moving average (EMA), and test-time augmentation (TTA) to significantly enhance model performance. Clinical interpretability is achieved through Attention Rollout–generated saliency maps. Evaluated on a dataset of 7,023 MRI scans, the model achieves 99.29% accuracy and a 99.25% macro F1-score, with perfect recall (100%) for both healthy tissue and meningioma, consistently outperforming CNN-based baselines.

0 citationsRead paper

Evaluating the Performance of Nigerian Lecturers using Multilayer Perceptron

May 22, 2025

This study addresses the limitations of current lecturer performance evaluation in Nigerian universities—namely, its overreliance on subjective, single-dimensional metrics disconnected from actual teaching effectiveness. To overcome these challenges, we propose a multidimensional, data-driven web-based evaluation system. Methodologically, the system integrates novel indicators—including student evaluations of teaching, research output, years of teaching experience, administrative responsibilities, and real-time classroom engagement metrics—trained on a custom-built educational performance dataset using a Multilayer Perceptron (MLP). Object-Oriented Analysis and Design (OOAD) principles ensure interpretability, bias mitigation, and end-to-end maintainability. Experimental results demonstrate high predictive accuracy (91%–96%), with MSE = 256.99 and MAE = 13.76. The framework significantly enhances fairness, objectivity, and evidence-based decision-making in academic performance assessment, offering a transferable technical architecture and methodological paradigm for quality governance in higher education within developing countries.

0 citationsRead paper

An Explorative Analysis of SVM Classifier and ResNet50 Architecture on African Food Classification

May 20, 2025

To address the scarcity of research on African food image recognition, this paper introduces the first unified six-class African food dataset (1,658 images) and conducts the first fair comparative evaluation of two paradigms: fine-tuned ResNet50 via transfer learning and handcrafted HOG features combined with an RBF-SVM classifier. Experimental results demonstrate that ResNet50 achieves significantly higher overall accuracy (+12.3%), yet the SVM-based approach exhibits superior generalization on certain food categories. A comprehensive evaluation—including accuracy, precision, recall, F1-score, and confusion matrix analysis—reveals the respective strengths and limitations of deep learning versus traditional methods in low-data, high-diversity African dietary contexts. The study establishes methodological guidelines and empirically validated baselines for food recognition in resource-constrained settings, enabling reproducible development and deployment of practical solutions for African food identification.

0 citationsRead paper

Development of a WAZOBIA-Named Entity Recognition System

May 10, 2025

This work addresses the critical gap in named entity recognition (NER) for low-resource African languages. We introduce WAZOBIA-NER, the first unified NER system for Nigeria’s three major indigenous languages—Hausa, Yoruba, and Igbo. To overcome data scarcity, we propose the first cross-lingual, low-resource NER framework that jointly processes OCR-extracted text and raw images via a novel multimodal input pipeline. Our architecture innovatively integrates Conditional Random Fields (CRF), Bidirectional Long Short-Term Memory (BiLSTM), and BERT-enhanced RNNs in a synergistic ensemble. WAZOBIA-NER is the first end-to-end, multilingual NER solution for African languages with native OCR augmentation. Evaluated on standard test sets, it achieves an F1-score of 0.9564 (precision: 0.9511, recall: 0.9400, accuracy: 0.9301), demonstrating the feasibility and effectiveness of high-accuracy NER modeling for low-resource African languages.

0 citationsRead paper

Recent publications

Latest Papers

HADS-Net:A Hybrid Attention-Augmented Dual-Stream Network with Physics-Informed Augmentation for Breast Ultrasound Image Classification

May 19, 2026

This study addresses the challenge of classifying benign, malignant, and normal tissues in breast ultrasound images, which is hindered by speckle noise, acoustic shadowing, and inter-class visual ambiguity. To tackle this, the authors propose a novel dual-stream network architecture that integrates a local stream specifically designed for lesion boundary refinement with a physics-informed global texture stream. Feature fusion and joint optimization are achieved through a cross-attention mechanism. Built upon an EfficientNet-B3 backbone, the model incorporates Sobel edge extraction, physics-aware data augmentation, and an adaptive class-weighted Focal Loss. Evaluated on the BUSI dataset, the method achieves 96.58% accuracy, a macro ROC-AUC of 0.9978, and a macro F1-score of 0.9654, with zero malignant lesions misclassified as normal—demonstrating substantially improved diagnostic reliability.

0 citationsRead paper

an interpretable vision transformer framework for automated brain tumor classification

Apr 23, 2026

This study addresses the challenges of time-consuming and subjective manual diagnosis of brain tumors in MRI by proposing an interpretable four-class classification method based on Vision Transformer (ViT-B/16) to automatically distinguish glioma, meningioma, pituitary tumor, and healthy tissue. The approach incorporates clinically oriented preprocessing with CLAHE enhancement, two-stage fine-tuning, hybrid MixUp/CutMix augmentation, exponential moving average (EMA), and test-time augmentation (TTA) to significantly enhance model performance. Clinical interpretability is achieved through Attention Rollout–generated saliency maps. Evaluated on a dataset of 7,023 MRI scans, the model achieves 99.29% accuracy and a 99.25% macro F1-score, with perfect recall (100%) for both healthy tissue and meningioma, consistently outperforming CNN-based baselines.

0 citationsRead paper

Evaluating the Performance of Nigerian Lecturers using Multilayer Perceptron

May 22, 2025

This study addresses the limitations of current lecturer performance evaluation in Nigerian universities—namely, its overreliance on subjective, single-dimensional metrics disconnected from actual teaching effectiveness. To overcome these challenges, we propose a multidimensional, data-driven web-based evaluation system. Methodologically, the system integrates novel indicators—including student evaluations of teaching, research output, years of teaching experience, administrative responsibilities, and real-time classroom engagement metrics—trained on a custom-built educational performance dataset using a Multilayer Perceptron (MLP). Object-Oriented Analysis and Design (OOAD) principles ensure interpretability, bias mitigation, and end-to-end maintainability. Experimental results demonstrate high predictive accuracy (91%–96%), with MSE = 256.99 and MAE = 13.76. The framework significantly enhances fairness, objectivity, and evidence-based decision-making in academic performance assessment, offering a transferable technical architecture and methodological paradigm for quality governance in higher education within developing countries.

0 citationsRead paper

An Explorative Analysis of SVM Classifier and ResNet50 Architecture on African Food Classification

May 20, 2025

To address the scarcity of research on African food image recognition, this paper introduces the first unified six-class African food dataset (1,658 images) and conducts the first fair comparative evaluation of two paradigms: fine-tuned ResNet50 via transfer learning and handcrafted HOG features combined with an RBF-SVM classifier. Experimental results demonstrate that ResNet50 achieves significantly higher overall accuracy (+12.3%), yet the SVM-based approach exhibits superior generalization on certain food categories. A comprehensive evaluation—including accuracy, precision, recall, F1-score, and confusion matrix analysis—reveals the respective strengths and limitations of deep learning versus traditional methods in low-data, high-diversity African dietary contexts. The study establishes methodological guidelines and empirically validated baselines for food recognition in resource-constrained settings, enabling reproducible development and deployment of practical solutions for African food identification.

0 citationsRead paper

Development of a WAZOBIA-Named Entity Recognition System

May 10, 2025

This work addresses the critical gap in named entity recognition (NER) for low-resource African languages. We introduce WAZOBIA-NER, the first unified NER system for Nigeria’s three major indigenous languages—Hausa, Yoruba, and Igbo. To overcome data scarcity, we propose the first cross-lingual, low-resource NER framework that jointly processes OCR-extracted text and raw images via a novel multimodal input pipeline. Our architecture innovatively integrates Conditional Random Fields (CRF), Bidirectional Long Short-Term Memory (BiLSTM), and BERT-enhanced RNNs in a synergistic ensemble. WAZOBIA-NER is the first end-to-end, multilingual NER solution for African languages with native OCR augmentation. Evaluated on standard test sets, it achieves an F1-score of 0.9564 (precision: 0.9511, recall: 0.9400, accuracy: 0.9301), demonstrating the feasibility and effectiveness of high-accuracy NER modeling for low-resource African languages.

0 citationsRead paper