Preprocessed Brinjal Leaf Image Dataset for Healthy and Diseased Leaf Detection (Phomopsis Blight and Little Leaf) Using Deep Learning

This dataset contains preprocessed images of Solanum melongena (Brinjal) leaves collected under real agricultural field conditions for automated plant disease detection using computer vision and deep learning techniques.

The dataset is divided into two primary categories:

1) Healthy_Leaves
2) Unhealthy_Leaves

The Unhealthy_Leaves category is further subdivided into two disease classes:

• Phomopsis Blight (Phomopsis vexans)
• Little Leaf (Brinjal Leaf Curl)

All images were manually cropped to retain only the leaf region of interest (ROI) to minimize background noise and enhance disease feature visibility. The cropped images were resized to a uniform spatial resolution of 224×224 pixels to ensure dimensional consistency across the dataset.

All images are stored in JPG (.jpg) format to maintain uniformity. Any PNG files present during dataset restructuring were converted to JPG to ensure consistent file format across all classes.

Each image was processed using 16 preprocessing and augmentation techniques to enhance structural, color, and texture-based disease patterns. These include contrast enhancement, filtering, morphological transformations, geometric augmentation, and texture-based filtering.

The images were captured under natural field lighting conditions using standard smartphone cameras during routine agricultural observation. Leaves exhibiting visible symptoms of disease and healthy samples were manually collected and photographed directly on plants to preserve natural texture and color variations present in real cultivation environments.

This dataset was created to support research in automated plant disease diagnosis using deep learning and computer vision techniques. The preprocessing variants were generated to highlight different visual characteristics of leaf images, including color distribution, venation structures, lesion boundaries, and texture irregularities.

Each original image is accompanied by 16 preprocessing variants:

V1 Original (Resized)
V2 Grayscale
V3 CLAHE
V4 Gamma Correction
V5 HSV Adjustment
V6 Brightness Enhancement
V7 Contrast Enhancement
V8 Sharpening
V9 Gaussian Blur
V10 Median Blur
V11 Bilateral Filtering
V12 Morphological Top-Hat
V13 Morphological Black-Hat
V14 Rotation (+25°)
V15 Horizontal Flip
V16 Unsharp Masking

Each processed image follows a structured naming convention:

ClassName_SerialNumber_PreprocessingType.jpg

This naming format ensures reproducibility, traceability, and structured experimentation for machine learning and deep learning workflows.

The dataset is suitable for:

• Binary classification (Healthy vs Diseased)
• Multi-class disease classification
• CNN benchmarking and transfer learning
• Precision agriculture research
• AI-based crop health monitoring systems

Researchers may use the dataset for training and evaluating machine learning models, benchmarking preprocessing pipelines, or developing automated crop monitoring systems for precision agriculture applications.