Rice Leaf Disease Classification

The goal of this project is to create a classifier that can accurately identify different diseases affecting rice leaves. The dataset used for training is sourced from Kaggle and contains multiple classes of rice leaf diseases.

Visit github

Preprocessing
Model Selection
Training Loop
Prediction and Visualization

Sample rice leaf image showing visible disease symptoms

Preprocessing

Import all needable libraries: torch, matplotlib, pandas and etc.

The dataset is split into training (80%), validation (10%), and test sets (10%) using random_split.

Data loaders are then created for each subset, with a batch size of 14 for training and validation, and 1 for testing.

Rice leaf dataset preview from the training set

Model Selection

The ResNet18 model is loaded using timm, with timm.create_model().

The model is initialized with pretrained weights, and the number of output classes is set to 8 (indicating there are 8 types of leaf conditions).

ResNet18 model architecture used for classification

Loss Function and Optimizer

The loss function used is CrossEntropyLoss, appropriate for multi-class classification.

The optimizer is Adam, with a learning rate of 3e-4.

Training Loop

The training loop runs for 10 epochs.

Each epoch involves both training and validation phases.

The model parameters are updated using backpropagation during training.

Validation is performed without gradient tracking, and the validation loss and accuracy are recorded.

Training and validation loss curves over 10 epochs

Prediction and Visualization

The code iterates through a batch of images and predicts their classes.

Predictions are compared to actual labels.

The result and prediction for random samples are visualized using matplotlib.

Training and validation accuracy scores are plotted.

Predicted vs actual labels visualized for sample test images