Leaf Classification

There's no shortage of leaf datasets for the world's most common plants. For leaves native to Bangladesh specifically, there was basically nothing — so the dataset had to be built first.

The Brief

Automatically identify plant species from leaf imagery, starting from the absence of any ready-made dataset for regional flora.

The Build

Working with a teammate, the pair assembled an original photo collection covering several leaf species native to Bangladesh, later published as an open dataset. The modeling work centered on designing and training classification models with TensorFlow.

• Custom-collected dataset of native Bangladeshi leaf species
• TensorFlow-based classification models
• Dataset and code released publicly

The Outcome

A model that exceeded 90% accuracy on the custom dataset, showing that even a small, locally-sourced dataset can yield strong computer vision results.