Selected work

Representative Papers

A Comparative Evaluation of Deep Learning Object Detection Models on a Real-World Multi-Plant Dataset from Africa

Aug 11, 2026

This study addresses the limited real-world applicability of existing agricultural vision models, which are predominantly trained on idealized datasets and struggle in the complex, dynamic field conditions prevalent in under-resourced regions such as Africa. To bridge this gap, the authors present a systematic evaluation of six state-of-the-art object detection models—YOLOv5, YOLOv8, YOLO11, YOLO26, Faster R-CNN, and RT-DETR—on AgriAISeg, the first multi-crop, multi-challenge dataset collected directly from African farmlands. Experimental results demonstrate that RT-DETR achieves the highest performance with an mAP@0.5:0.95 of 0.624, while YOLO-family models exhibit consistently strong accuracy and training efficiency. In contrast, Faster R-CNN suffers significant performance degradation in complex scenarios. This work provides the first empirical evidence of the varying suitability of modern detectors in authentic African agricultural settings, offering critical guidance for deploying AI solutions in resource-constrained environments.

0 citationsRead paper

Sesame Plant Segmentation Dataset: A YOLO Formatted Annotated Dataset

Jan 12, 2026

This study addresses the absence of high-quality, pixel-level segmentation datasets for sesame plants—a critical gap that has hindered the application of artificial intelligence in precision agriculture. To bridge this gap, we present the first publicly available, high-resolution image segmentation dataset specifically curated for sesame crops, comprising 292 images captured in Nigerian fields using smartphone cameras. Pixel-level annotations were generated using Segment Anything Model v2 under the supervision of local farmers and released in a YOLO-compatible format. Evaluation using YOLOv8 demonstrates the dataset’s efficacy, achieving mAP@0.5 of 84% for detection and mAP@0.5:0.95 of 58% and 52% for detection and segmentation tasks, respectively. This resource fills a significant void in visual data for sesame farming in Africa and supports future AI-driven agricultural innovations.

0 citationsRead paper

Taxonomy and Trends in Reinforcement Learning for Robotics and Control Systems: A Structured Review

Oct 11, 2025

This paper addresses the persistent gap between theoretical advances in reinforcement learning (RL) and their practical deployment in robotics and control systems. To bridge this divide, we propose a structured taxonomy tailored to real-world robotic applications, grounded in the Markov decision process (MDP) framework and systematically incorporating mainstream deep RL algorithms—including DDPG, TD3, PPO, and SAC—across canonical domains such as motion control, dexterous manipulation, and multi-agent coordination. The taxonomy explicitly integrates training paradigms and deployment maturity metrics. Crucially, we identify recurring design patterns and evolutionary trends in high-dimensional continuous control tasks, thereby unifying theoretical insights with engineering constraints. Our framework advances reproducibility, transferability, and robustness in RL deployment on physical robots, offering both a methodological foundation and actionable guidelines for practitioners. (149 words)

0 citationsRead paper

Multi-Agent Reinforcement Learning in Intelligent Transportation Systems: A Comprehensive Survey

Aug 27, 2025

To address the challenges of dynamics, non-stationarity, and scalability in multi-agent collaborative decision-making within intelligent transportation systems (ITS), this paper proposes a unified classification framework for multi-agent reinforcement learning (MARL) tailored to ITS. The framework systematically categorizes MARL approaches into four paradigms: value-based, policy-gradient-based, actor-critic-based, and communication-enhanced methods. It further maps these to key ITS applications—including traffic signal control, cooperative autonomous driving, logistics dispatching, and on-demand mobility. Empirical evaluation is conducted across mainstream simulation platforms (SUMO, CARLA, CityFlow), identifying critical bottlenecks such as sim-to-real transfer, credit assignment, and environmental non-stationarity. This work establishes the first structured taxonomy that jointly considers algorithmic principles and traffic-domain semantics, providing a comprehensive survey, standardized benchmarks, and actionable research directions for advancing both the theoretical foundations and real-world deployment of MARL in ITS.

0 citationsRead paper

When Simpler Wins: Facebooks Prophet vs LSTM for Air Pollution Forecasting in Data-Constrained Northern Nigeria

Aug 22, 2025

Air pollution forecasting in Northern Nigeria faces significant challenges due to sparse, irregular monitoring data and limited infrastructure. Method: This study comparatively evaluates Facebook Prophet and Long Short-Term Memory (LSTM) networks for monthly forecasting of CO, SO₂, and SO₄ concentrations, using observational data from 19 states spanning 2018–2023. Contribution/Results: Prophet achieves accuracy comparable to or exceeding that of LSTM across most pollutants—particularly under conditions of strong seasonality, long-term trends, and limited sample size. These findings challenge the prevailing assumption that higher model complexity inherently yields superior performance, underscoring instead the critical importance of aligning model selection with intrinsic data characteristics (e.g., sparsity, periodic structure). The study proposes Prophet as a lightweight, interpretable, and cost-effective alternative for environmental time-series forecasting in resource-constrained settings. It advances a methodology prioritizing model appropriateness over complexity, offering practical guidance for sustainable air quality monitoring in data-scarce regions.

0 citationsRead paper

Recent publications

Latest Papers

A Comparative Evaluation of Deep Learning Object Detection Models on a Real-World Multi-Plant Dataset from Africa

Aug 11, 2026

This study addresses the limited real-world applicability of existing agricultural vision models, which are predominantly trained on idealized datasets and struggle in the complex, dynamic field conditions prevalent in under-resourced regions such as Africa. To bridge this gap, the authors present a systematic evaluation of six state-of-the-art object detection models—YOLOv5, YOLOv8, YOLO11, YOLO26, Faster R-CNN, and RT-DETR—on AgriAISeg, the first multi-crop, multi-challenge dataset collected directly from African farmlands. Experimental results demonstrate that RT-DETR achieves the highest performance with an mAP@0.5:0.95 of 0.624, while YOLO-family models exhibit consistently strong accuracy and training efficiency. In contrast, Faster R-CNN suffers significant performance degradation in complex scenarios. This work provides the first empirical evidence of the varying suitability of modern detectors in authentic African agricultural settings, offering critical guidance for deploying AI solutions in resource-constrained environments.

0 citationsRead paper

Sesame Plant Segmentation Dataset: A YOLO Formatted Annotated Dataset

Jan 12, 2026

This study addresses the absence of high-quality, pixel-level segmentation datasets for sesame plants—a critical gap that has hindered the application of artificial intelligence in precision agriculture. To bridge this gap, we present the first publicly available, high-resolution image segmentation dataset specifically curated for sesame crops, comprising 292 images captured in Nigerian fields using smartphone cameras. Pixel-level annotations were generated using Segment Anything Model v2 under the supervision of local farmers and released in a YOLO-compatible format. Evaluation using YOLOv8 demonstrates the dataset’s efficacy, achieving mAP@0.5 of 84% for detection and mAP@0.5:0.95 of 58% and 52% for detection and segmentation tasks, respectively. This resource fills a significant void in visual data for sesame farming in Africa and supports future AI-driven agricultural innovations.

0 citationsRead paper

Taxonomy and Trends in Reinforcement Learning for Robotics and Control Systems: A Structured Review

Oct 11, 2025

This paper addresses the persistent gap between theoretical advances in reinforcement learning (RL) and their practical deployment in robotics and control systems. To bridge this divide, we propose a structured taxonomy tailored to real-world robotic applications, grounded in the Markov decision process (MDP) framework and systematically incorporating mainstream deep RL algorithms—including DDPG, TD3, PPO, and SAC—across canonical domains such as motion control, dexterous manipulation, and multi-agent coordination. The taxonomy explicitly integrates training paradigms and deployment maturity metrics. Crucially, we identify recurring design patterns and evolutionary trends in high-dimensional continuous control tasks, thereby unifying theoretical insights with engineering constraints. Our framework advances reproducibility, transferability, and robustness in RL deployment on physical robots, offering both a methodological foundation and actionable guidelines for practitioners. (149 words)

0 citationsRead paper

Multi-Agent Reinforcement Learning in Intelligent Transportation Systems: A Comprehensive Survey

Aug 27, 2025

To address the challenges of dynamics, non-stationarity, and scalability in multi-agent collaborative decision-making within intelligent transportation systems (ITS), this paper proposes a unified classification framework for multi-agent reinforcement learning (MARL) tailored to ITS. The framework systematically categorizes MARL approaches into four paradigms: value-based, policy-gradient-based, actor-critic-based, and communication-enhanced methods. It further maps these to key ITS applications—including traffic signal control, cooperative autonomous driving, logistics dispatching, and on-demand mobility. Empirical evaluation is conducted across mainstream simulation platforms (SUMO, CARLA, CityFlow), identifying critical bottlenecks such as sim-to-real transfer, credit assignment, and environmental non-stationarity. This work establishes the first structured taxonomy that jointly considers algorithmic principles and traffic-domain semantics, providing a comprehensive survey, standardized benchmarks, and actionable research directions for advancing both the theoretical foundations and real-world deployment of MARL in ITS.

0 citationsRead paper

When Simpler Wins: Facebooks Prophet vs LSTM for Air Pollution Forecasting in Data-Constrained Northern Nigeria

Aug 22, 2025

Air pollution forecasting in Northern Nigeria faces significant challenges due to sparse, irregular monitoring data and limited infrastructure. Method: This study comparatively evaluates Facebook Prophet and Long Short-Term Memory (LSTM) networks for monthly forecasting of CO, SO₂, and SO₄ concentrations, using observational data from 19 states spanning 2018–2023. Contribution/Results: Prophet achieves accuracy comparable to or exceeding that of LSTM across most pollutants—particularly under conditions of strong seasonality, long-term trends, and limited sample size. These findings challenge the prevailing assumption that higher model complexity inherently yields superior performance, underscoring instead the critical importance of aligning model selection with intrinsic data characteristics (e.g., sparsity, periodic structure). The study proposes Prophet as a lightweight, interpretable, and cost-effective alternative for environmental time-series forecasting in resource-constrained settings. It advances a methodology prioritizing model appropriateness over complexity, offering practical guidance for sustainable air quality monitoring in data-scarce regions.

0 citationsRead paper