Disease Detection and Forecasting
This project involved the preparation of a highly-accurate, multi-modal dataset to train predictive models for Ganoderma disease in Ghanaian oil palm plantations. My role was to design and execute a comprehensive data labeling strategy that went beyond standard image annotation. The process included: Image Annotation: I used bounding boxes and polygon segmentation to precisely outline infected oil palm trees in satellite and aerial drone imagery. This involved identifying subtle visual cues, such as discolored fronds and fruiting bodies, which required deep subject matter expertise. Time-Series Annotation: I annotated time-series data from environmental sensors (e.g., humidity, temperature, soil moisture) to correlate specific environmental conditions with disease outbreaks and progression over a ten-year period. Text and Field Data Classification: I categorized and labeled historical field reports and soil analysis documents.