Funders: Global Challenges Research Fund (GCRF) & BBSRC
Total Funding: £21,000

Role: Africa Lead

Overview:
Developed a decision support system with modules for farmer education, market access, and crop yield prediction, aimed at improving farming decisions in rural areas.

Key Achievements:

• Led the development of the system into a user-friendly Android mobile app.
  
  

• Published research on the system’s impact on agricultural decision-making.

Outcome:
Successfully completed within the project timeline, providing farmers with tools to enhance yield predictions and make informed decisions.

Funder: Innovate UK (50877)
Total Funding: £90,189

Role: Principal Investigator & Technical Lead

Overview:
Led the design and development of a proof-of-concept model for detecting blackberry ripeness, addressing challenges faced by fruit growers in assessing maturity. The project involved creating a novel multi-input convolutional neural network ensemble classifier to detect subtle ripeness traits.

Key Achievements:

• Developed a functional ripeness detection model.
  
  

• Published research on the model’s effectiveness.
  
  

Outcome:
Successfully delivered a solution for more accurate blackberry ripeness detection, benefiting growers.

Funder: Collaborative Industry Project with National Composites Centre (NCC)
Total Funding: £188,642

Role: Co-Investigator & Technical Lead

Overview:
Designed and developed a robust machine learning framework, incorporating modified YOLO object detection algorithm and novel pre-processing pipeline for detecting and classifying out-of-plane wrinkle defects during the preform stage of wind turbine blade manufacturing.

Key Achievements:

• Designed efficient protocol and system for on-manufacturing carbon fibre data collection.

• Created a proof-of-concept automated system for carbon fibre defect detection and classification in the actual manufacturing process.

Outcome:
Successfully implemented an automated solution for quality control in wind turbine blade production.

Funder: Belron International Ltd
Total Funding: £42,421.67

Role: Co-Investigator & Technical Lead

Overview:
Developed a solution to reliably localize regions of uneven illumination and light saturation on a calibration board, addressing calibration system failures due to occlusion over calibration marks. The project applied computer vision techniques to improve the accuracy of light localization.

Key Achievements:

• Delivered software to accurately identify and correct illumination inconsistencies on the calibration board.
  
  

Outcome:
Successfully enhanced calibration process reliability through machine vision.

Funder: Belron International Ltd
Total Funding: £60,777.78

Role: Co-Investigator & Technical Lead

Overview:
The project aimed to develop a computer vision and machine learning framework for accurate ADAS camera pose estimation, focusing on yaw and row pattern changes. Stage 1 involved creating a functional pose estimation model, while Stage 2 extended the investigation to test real-world application at Belron Workshops, exploring constraints and identifying optimal conditions for model deployment.

Key Achievements:

• Developed a high-accuracy pose estimation model.
  
  

• Conducted real-world experiments, demonstrating impressive pitch and yaw estimations with minimal error.
  
  

• Delivered recommendations for Belron International Ltd based on findings.
  
  

Outcome:
Functional pose estimation model with clear application guidelines for ADAS camera replacement.

Funder: Innovate UK
Total Funding: £491,929 (£148,746 for UWE)

Role: Technical Lead

Overview:
This collaborative project focused on leveraging computer vision and machine learning techniques to create an AI framework for the early detection of digital dermatitis (DD) and white line disease in dairy cattle hooves. The project presented a challenge in real-time data collection, which led to the development of a novel data selection approach. This approach utilized a custom YOLO model, combined with active learning principles and dynamic programming with a memoization algorithm. 

Key Achievements:

• Designed and developed a novel selection algorithm that ensures high accuracy and minimal error in associating each cow with its ID, eliminating redundant frames, categorizing hooves, and preserving lifted hooves for both cow identification and hoof classification. 

• Created a hoof detection model for DD and white line identification.
  
  

Outcome:
Successfully developed a filtering algorithm and detection model for real-time hoof condition monitoring.

Funder: Cancer Research UK
Total Funding: £35,000

Role: Technical Lead

Overview:
The project focused on evaluating the ability of AI/ML algorithms to differentiate between normal oesophagus tissue and Barrett’s oesophagus, achieving an impressive NPV of 100%, sensitivity of 100%, and specificity of 90% using the PIVI criteria. While the model showed strong results, challenges arose when distinguishing between similar samples such as LGD and inflammation. The follow-on project will focus on developing algorithms to accurately differentiate LGD from regenerative inflammation and predict its progression to HGD.

Key Achievements:

• Developed a classification model for Barrett’s oesophagus with high diagnostic accuracy.
  
  

• Laid the groundwork for future AI/ML models to predict LGD progression to HGD.
  
  

Outcome:
Barrett’s oesophagus classification model demonstrating strong performance for clinical use.