Cassava Disease Image Classification

Problem Statement

This project involves approximating a function that classifies images ( X ) of cassava leaves into the correct disease categories ( Y ).

Dataset

The dataset consists of images labeled into five classes:

• "0": Cassava Bacterial Blight (CBB)
• "1": Cassava Brown Streak Disease (CBSD)
• "2": Cassava Green Mottle (CGM)
• "3": Cassava Mosaic Disease (CMD)
• "4": Healthy

The dataset includes the following number of samples per class:

• CBB: 921

• CBSD: 1831

• CGM: 1993

• CMD: 11027

• Healthy: 2166

• Input Dimensions: ( X ) -> [17938 x 3 x 800 x 600]

• Output Labels: ( Y ) -> [17938 x 5]

Preprocessing

• Images are resized to 256x192 to reduce computational cost.
• The .csv file containing image-label pairs was refactored for better usability.

Models

State-of-the-Art Architectures

1. ResNet50

   • 50-layer deep convolutional neural network (CNN).
   • Pretrained on ImageNet.
   • Uses residual connections to prevent overfitting and improve training.

2. RexNet150

   • 150-layer deep CNN.
   • Pretrained on ImageNet.
   • Lightweight architecture using 1x1 bottleneck kernels with expanded iterations, which helps reduce model complexity and training time.

Custom Models

Locally implemented custom models using different combinations of convolutional blocks and fully connected layers.

Training Parameters

The following hyperparameters were used for training the models:

• Number of Filters: 64
• Kernel Size: 3
• Pooling Size: 2
• Dense Layer Size: 512
• Stride: 1
• Learning Rate: 1e-5
• Weight Decay: 1e-5
• Optimizer: Adam
• Loss Weights: [ [19.5588, 9.8417, 9.0510, 1.6351, 8.3200] ]

Hyperparameter Search - Bayesian Optimization

Bayesian optimization was used to find the optimal hyperparameters. The best parameters were:

• Learning Rate: 0.00024526126311336793
• Optimizer: Adam
• Number of Epochs: 19
• Weight Decay: 0.00021830968390524624
• Optimal F1 Score: 0.6623861743301297