ACHENY: A standard Chenopodiaceae image dataset for deep learning models

Area of Science:

• Botany and Plant Science
• Computer Science and Artificial Intelligence
• Biodiversity and Conservation

Background:

• The Chenopodiaceae family, comprising approximately 1500 species, holds significant ecological importance globally.
• Biodiversity conservation of these species is threatened by human activities, necessitating effective identification and surveillance methods.
• Deep learning offers a promising approach for automated identification, but requires relevant, high-quality datasets.

Purpose of the Study:

• To introduce the ACHENY dataset, a novel collection of images for training deep learning models.
• To facilitate the development of efficient deep learning models for categorizing Chenopodiaceae species in their natural habitats.
• To support biodiversity conservation efforts through improved plant identification technologies.

Main Methods:

• The ACHENY dataset was collected from natural desert and semi-desert habitats in Iran.
• It comprises 27,030 RGB images of 30 Chenopodiaceae species, with varying numbers of images per species (300-1461).
• Images were captured under diverse real-world conditions (varying distances, viewpoints, angles, sunlight) and resized to 224x224 and 64x64 dimensions. Data was split into 72% training, 18% validation, and 10% testing sets.

Main Results:

• The dataset is imbalanced, reflecting natural species distribution.
• Images were collected without pre-processing, except for resizing, maintaining real-world variability.
• The dataset is structured for direct use with various deep learning architectures.

Conclusions:

• The ACHENY dataset provides a valuable resource for advancing deep learning applications in plant identification.
• It enables the creation of efficient models for Chenopodiaceae species categorization, crucial for conservation.
• This dataset supports research into automated biodiversity monitoring in challenging environments.