MLD-BD: A Comprehensive Image Dataset for Mango Leaf Disease Detection in Bangladesh

The "MLD-BD" dataset is a primary and augmented image dataset specifically curated for mango leaf disease identification and classification, collected from major mango-producing regions in Bangladesh, namely Rajshahi, Chapainawabganj, and Mymensingh. Mango cultivation plays a vital role in the agricultural economy of Bangladesh; however, diseases affecting mango leaves significantly impact production yield and quality. Thus, timely and precise detection is crucial for effective disease management and control measures.

Dataset Overview:

This dataset comprises a total of 9,000 images, systematically organized into two subsets:

Primary (Original) Dataset: Consisting of 3,000 images, with each class containing 600 original images captured directly from the fields.

Augmented Dataset: This subset comprises 6,000 images, with each class having 1,200 images, generated through various data augmentation techniques, including rotation, flipping, scaling, translation, brightness adjustments, and noise addition. This augmentation ensures a diverse representation of each disease, helping improve the robustness and generalization capabilities of machine learning and deep learning models.

Classes:

The dataset covers 5 distinct classes of mango leaf conditions, crucial for practical agricultural applications:

Anthracnose: A fungal disease causing dark, sunken lesions on leaves, severely impacting mango production.

Bacterial Canker: A bacterial infection characterized by water-soaked lesions leading to leaf necrosis and branch dieback.

Die Back: Fungal-induced disease causing progressive death of branches, noticeable leaf browning, and drying.

Gall Midge: Insect infestation leading to abnormal leaf growth, curled leaf structures, and deformation.

Healthy: Images representing disease-free mango leaves for comparative and control purposes.

Data Collection Methodology:

Images were captured using high-resolution smartphone cameras under natural environmental conditions to ensure real-world variability, including different lighting conditions, angles, and backgrounds. Data collection spanned multiple mango orchards across the selected geographical locations to ensure extensive diversity and representativeness.

Regions Covered: Rajshahi, Chapainawabganj, and Mymensingh, Bangladesh.

Image Resolution: Standardized to a resolution suitable for training machine learning models 512×512 pixels.

Format: JPG format, optimized for computational efficiency and easy integration with popular deep learning frameworks.

Applications:

Automated mango leaf disease identification.

Comparative analysis of machine learning and deep learning models.

Development and evaluation of image-processing techniques in precision agriculture.

Usage and Accessibility:

This dataset will be publicly available on Mendeley Data, ensuring easy accessibility and utilization by researchers, data scientists, agriculturists, and practitioners in agricultural informatics.