Coconut Detection and Maturity Estimation

The problem

Assessing coconut maturity by eye is inconsistent and does not scale. The thesis asked whether machine-learning models could detect coconuts in photographs and classify their maturity stage reliably enough to support that assessment.

What the project involved

• Frame the detection and maturity-classification problem and design the experimental approach.
• Build the machine-learning workflow end to end, from image data through trained models to evaluation against held-out test data.
• Write the thesis documentation, including an honest account of the models' limitations and failure cases.

Approach

• Framing maturity as distinct, well-defined stages came before any modeling — the labels drive everything downstream.
• Detection and classification were treated as separate concerns, each evaluated on its own terms.
• Model quality was judged against held-out data rather than the images the models had already seen.
• The write-up reports limitations and failure cases alongside results, which is what makes the evidence reusable.

Capabilities demonstrated

• Problem framing for applied machine learning
• Image data and detection pipelines
• Model training and evaluation
• Technical writing under academic review

Technology categories

• Python
• Machine learning
• Computer vision
• Model evaluation

Evidence

This project is public; the repository contains the source and the full write-up.

View the thesis repository on GitHub