Dataset for: Nutrispace: A Novel Color Space to Enhance Deep Learning Based Early Detection of Cucurbits Nutritional Deficiency
Description
This dataset is an extension of our research titled "Nutrispace: A Novel Color Space to Enhance Deep Learning Based Early Detection of Cucurbits Nutritional Deficiency," which is currently under review. For utmost reproducibility, we've included a ZIP file and a Python file: The ZIP file, "dataset_images," comprises 2700 segmented RGB images spanning 9 classes. The dataset is balanced and divided into train, validation, and test sets with a ratio of 0.70:0.15:0.15. Class distribution of the dataset: Crop | Class | Number of samples ---------------|------------------|------------------- Ash gourd | Healthy | 300 Ash gourd | N deficiency | 300 Ash gourd | K deficiency | 300 ---------------|------------------|------------------- Bitter gourd | Healthy | 300 Bitter gourd | N deficiency | 300 Bitter gourd | K deficiency | 300 ---------------|------------------|------------------- Snake gourd | Healthy | 300 Snake gourd | N deficiency | 300 Snake gourd | K deficiency | 300 ---------------|------------------|------------------- | | Total = 2700 The converter.py file is a Python class file for transforming the RGB images to HSV, CIELAB, and our proposed color space - Nutrispace. The .py file contains a short documentation within it for the ease of use.
Files
Steps to reproduce
We captured the leaf samples with an iPhone 13 Pro Max from July 8 to August 16, 2022, using the primary 12 MP wide camera with an f/1.5 aperture and an exposure of -1. The pictures were taken from a directly overhead perspective with a 1:1 aspect ratio, and the flash was disabled. After that, we utilized a k-means clustering strategy to segregate leaves from the background in order to reduce the complexity brought on by the intricate surroundings. Finally, we balanced the dataset through spatial augmentation, incorporating various transformations such as horizontal flip, rotation, shear, skew, and zoom with Python's Augmentor library.