Skip to main content

← All projects

Rice Leaf Classifier

Classifying rice-leaf disease with a CNN.

Preprocessing
Research
2024

The problem

Early disease identification in crops is valuable but needs accurate, automatable classification.

What it does

Trains a ResNet18 classifier on a multi-class Kaggle rice-leaf-disease dataset, with preprocessing, a training loop and prediction visualization.

Stack

Python
PyTorch
ResNet18
torchvision

My role

Preprocessing
Model selection
Training loop
Visualization

Highlights

Transfer learning

Fine-tunes a pretrained ResNet18 instead of training from scratch, getting strong accuracy on a small set.

Visualized predictions

Outputs predicted disease classes alongside the input leaves for quick sanity checks.

Screens

How Rice Leaf Classifier works: Kaggle dataset → torchvision preprocessing → ResNet18 train/val loop → loss curves + predictions — How it works

How it works

Sample rice-leaf images across the disease classes — The disease classes

The disease classes

Train and validation loss curves — Train / validation loss

Train / validation loss

Grid of predictions vs ground truth on test leaves — Predictions vs. ground truth

Predictions vs. ground truth

Single prediction on a rice-leaf image — A single prediction

A single prediction

View on GitHub All projects