Sartak Fig Leaf Dataset (SFLD)

Description

The Sartak Fig Leaf Dataset (SFLD) comprises 10,185 high-resolution fig leaf images (512×512 PNG) collected from real-world orchards in Sartak Valley, Kurdistan Region of Iraq. The dataset includes 2,037 original photographs captured using a Nikon D7100 DSLR (6000×4000 pixels) across three consecutive days (8:00 AM - 6:00 PM) to capture diverse daylight conditions. The dataset spans four classes: Mosaic (2,755 images), Anthracnose (2,240 images), Cercospora_leaf_spot (2,625 images), and Healthy (2,565 images). All labels were verified by a certified plant pathologist. Preprocessing included center-cropping to 1:1 aspect ratio and resizing to 512×512 using LANCZOS interpolation. Four augmentation techniques (horizontal flip, rotation ±30°, brightness ±20%, and contrast ±20%) were applied independently, generating 8,148 augmented images. The dataset is organized into Original/ and Augmented/ folders, accompanied by metadata.csv, dataset_manifest.json, README.md, and LICENSE.txt files detailing object IDs, source files, augmentation parameters, environmental conditions, and usage guidelines. Quality control removed 374 images (15.5%) due to blur, overexposure, or underexposure. Designed for AI-driven fig leaf disease classification, object detection, segmentation, plant pathology, and precision agriculture applications.

Steps to reproduce

DATA COLLECTION - Location: Sartak Valley, Kurdistan Region of Iraq (near Mount Bamo, ~80 km from Sulaymaniyah) - Duration: 3 consecutive days, 8:00 AM – 6:00 PM - Camera: Nikon D7100 DSLR (6000×4000 pixels, 24-bit RGB) - Settings: ISO 100, f/5, 1/125s, 30–75mm focal length, no flash - Raw file size: 5–7 MB per image LABELING - All images labeled by a certified plant pathologist based on disease symptomatology - Four classes: Mosaic, Anthracnose, Cercospora_leaf_spot, Healthy QUALITY CONTROL - 2,411 raw images captured; 374 (15.5%) excluded due to blur, overexposure, underexposure, or camera issues PREPROCESSING - Center-cropped to 1:1 aspect ratio - Resized to 512×512 pixels using LANCZOS interpolation - Saved as lossless PNG AUGMENTATION - 8,148 augmented images generated using 4 independent techniques (never stacked): * Horizontal flip (probability = 1.0) * Rotation (±30°) * Brightness (factor 0.8–1.2) * Contrast (factor 0.8–1.2) FINAL DATASET - 2,037 original + 8,148 augmented = 10,185 total images - Organized into Original/ and Augmented/ folders - Includes metadata.csv and dataset_manifest.json SOFTWARE - Python (Pillow/OpenCV) for preprocessing and augmentation - pandas for metadata generation REPRODUCIBILITY - All parameters documented in metadata.csv and dataset_manifest.json

Institutions

• Sulaimani Polytechnic University

  Sulaymaniyah, Sulaymaniyah
• University of Halabja

  Sulaymaniyah, Halabja
• University of Sulaimani

  Sulaymaniyah, Sulaymaniyah

Categories

Licence