Research Projects

StomaGAN represents a sophisticated deep learning framework that leverages Generative Adversarial Networks to address the critical challenge of limited training data in stomatal analysis. This pioneering research demonstrates how a modified Deep Convolutional GAN architecture, enhanced with adaptive top-k selection mechanisms, spectral normalisation techniques, and strategic noisy label integration, can generate photorealistic synthetic stomatal images from minimal datasets. The framework successfully addresses training instability issues inherent in traditional GANs while producing high-fidelity artificial data that significantly augments existing stomatal image collections. The synthetic data generated through this methodology has been rigorously validated through its successful application in training deep convolutional neural networks for automated stomatal detection, achieving remarkable accuracy rates on previously unseen real-world imagery. This open-source contribution provides the research community with comprehensive code repositories, curated datasets, and analytical tools, establishing a new paradigm for data augmentation in plant phenotyping applications.

StomataHub constitutes a revolutionary collaborative ecosystem designed to democratise access to cutting-edge stomatal analysis methodologies through the integration of artificial intelligence and open science principles. This comprehensive platform, architected and maintained by Dr. Jonathon Gibbs, serves as a centralised repository for meticulously annotated stomatal datasets, peer-reviewed research publications, and state-of-the-art machine learning applications. The platform features an advanced web-based stomatal detection system currently under active development, incorporating the latest advances in computer vision and deep learning architectures. By fostering interdisciplinary collaboration between plant biologists, computer scientists, and agricultural researchers, StomataHub accelerates the pace of discovery in plant phenotyping research. The platform's expanding collection of publicly accessible datasets and sophisticated analytical tools empowers researchers worldwide to advance our understanding of stomatal traits, ultimately contributing to enhanced crop breeding programs and improved agricultural sustainability in the face of climate change.

This groundbreaking research introduces a novel deep learning methodology that revolutionises automated stomatal detection and morphometric analysis in high-resolution microscopy imagery. The investigation presents a custom-designed convolutional neural network architecture that incorporates advanced feature extraction mechanisms, achieving unprecedented accuracy rates exceeding 98% across diverse microscopic imaging modalities and plant species. The framework enables comprehensive high-throughput quantitative analysis of critical stomatal parameters including density distributions, morphological dimensions, and complex spatial patterning characteristics. These capabilities provide invaluable insights into fundamental plant physiological processes, particularly water-use efficiency mechanisms and environmental adaptation strategies. The research contribution includes the development of an extensively annotated dataset comprising thousands of stomatal images, establishing a new benchmark for computational plant biology research. This work bridges the gap between traditional botanical analysis and modern artificial intelligence, offering researchers unprecedented tools for understanding plant responses to environmental stressors and climate variability.

This comprehensive research program advances the field of autonomous plant phenotyping through the development of sophisticated robotic vision systems that seamlessly integrate precision mechanical engineering, advanced computer vision algorithms, and machine learning methodologies. The investigation presents novel path-planning algorithms specifically optimised for agricultural applications, incorporating multi-objective optimisation strategies that maximise information acquisition while minimising mechanical disturbance to delicate plant structures. The system architecture demonstrates remarkable innovation through its hybrid approach, combining volumetric reconstruction techniques with high-resolution point cloud processing methodologies to achieve sub-millimetre accuracy in complex plant architecture modelling. This technological advancement enables unprecedented precision in phenotypic measurements, facilitating detailed analysis of growth patterns, structural biomechanics, and morphological adaptations. The research incorporates GPU-accelerated processing pipelines utilising CUDA frameworks, ensuring real-time performance suitable for high-throughput agricultural research applications. These contributions establish new standards for non-invasive plant monitoring and provide essential tools for advancing crop improvement programs and agricultural sustainability research.

This sophisticated software suite represents a comprehensive solution for automated plant phenotyping, seamlessly integrating cutting-edge machine learning algorithms with established computer vision methodologies to enable large-scale trait extraction from digital imagery. The toolkit incorporates advanced image processing pipelines capable of automated extraction of complex morphological features including precise leaf area measurements, stomatal density quantification, and detailed morphometric characterisation across diverse plant species and imaging conditions. The system architecture employs ensemble learning approaches, combining multiple algorithmic strategies to ensure robust performance across varying environmental conditions and imaging protocols. Built upon industry-standard frameworks including OpenCV for computer vision operations and Scikit-learn for machine learning implementations, the tools provide researchers with reliable, scalable solutions for phenotypic analysis. The software suite includes comprehensive documentation, validation datasets, and performance benchmarks, enabling seamless integration into existing research workflows while maintaining the flexibility required for custom applications in plant biology and agricultural research.