Cassava Bacterial Blight (CBB)

Project Background
Cassava is a vital food crop in Africa, Asia, and South America, serving as a staple 
food for over 800 million people worldwide. However, cassava leaves are susceptible 
to various diseases, such as Cassava Bacterial Blight (CBB), Cassava Brown 
Streak Disease (CBSD), Cassava Green Mottle (CGM), and Cassava Mosaic 
Disease (CMD). These diseases can lead to significant yield losses or even crop 
failure, posing a serious threat to food security.
Traditional disease detection methods rely on visual inspection by agricultural 
experts, which is time-consuming, labor-intensive, and prone to subjective errors. 
With the advancement of computer vision and deep learning technologies, automated 
disease detection methods based on Convolutional Neural Networks (CNNs) have 
become a research hotspot. By training deep learning models, it is possible to quickly 
and accurately identify the health status and disease types of cassava leaves, helping 
farmers take timely preventive measures to reduce crop losses.
Project Task
The task of this course project is to train a Convolutional Neural Network (CNN) 
model for Cassava Leaf Disease Classification. You are required to design and 
implement a CNN model, train it using the provided training and validation data, and 
finally evaluate the model's accuracy on an unseen test dataset.
Project Requirements
Model Design:
You are free to design the CNN model using convolutional layers, pooling layers, 
fully connected layers, etc., as introduced in the course.
Pre-trained models (e.g., ResNet, EfficientNet) can be used for transfer learning.
Dataset:
Training and validation data are provided on Moodle.
The test dataset is not publicly available, and the final model will be evaluated on this 
unseen test set.
Submission Files:
Source Code: All Python files related to the project.
Dependency Packages: A *.yml file listing the required packages.
Checkpoint File: The checkpoint file of your well-trained CNN model.
Hardware Limitations:
The training process is allowed to use GPUs with performance not 
exceeding NVIDIA RTX 4090 24GB.
Evaluation Criteria:
Model Accuracy: The classification accuracy of the model on the unseen test dataset.
Code Readability: The source code should be well-structured, commented, and easy 
to understand.
Dataset Description
The dataset contains images of cassava leaves, with each image labeled according to 
the health status or disease type of the leaf. The specific categories are:
• 0: Cassava Bacterial Blight (CBB)
• 1: Cassava Brown Streak Disease (CBSD)
• 2: Cassava Green Mottle (CGM)
• 3: Cassava Mosaic Disease (CMD)
• 4: Healthy