Multimodal robust ML models for food safety target detection

Overview

This innovative solution leverages an ensemble of advanced signal processing, computer vision, and machine learning methods to develop predictive models for food safety diagnostics. The models are designed to accurately infer results from diagnostic test kit images and device outputs, achieving gold standard accuracy. By utilizing these technologies, the solution aims to enhance the precision and reliability of food safety assessments.

Technical specifications

Key features:

• Ensemble of machine learning models combining signal processing and computer vision
• High specificity and sensitivity for diagnostic accuracy
• Application of pretrained models and transfer learning to optimize development time
• Utilization of Python, Git, CVAT, and MLFlow for development and MLOps

The approach includes defining data requirements, leveraging existing labeled datasets, and using potential synthetic data to enhance model robustness. Feature extraction and validation are integral to the process, employing computer vision and geometric measures to ensure accuracy before predictions are made.

Technology readiness level

The current development stage is at Technology Readiness Level 2, indicating that the basic principles have been observed and reported. The research is in the early stages of defining input data types, outputs, accuracy requirements, and the data pipeline necessary for further development and testing.