OpenPlant: A Large-Scale Benchmark Dataset for Agricultural Plant Classification Using CNNs, ViTs, and VLMs

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Deep learning models are crucial for precision agriculture tasks like crop monitoring and weed control.
• Existing plant datasets often lack scale, environmental diversity, and data integration capabilities.
• These limitations hinder the development and accuracy of deep learning models in agriculture.

Purpose of the Study:

• Introduce OpenPlant, a novel, large-scale, open-access dataset for agricultural plant classification.
• Establish a benchmark for evaluating deep learning models in plant identification.
• Address the limitations of existing datasets in scale, diversity, and data integration.

Main Methods:

• Developed OpenPlant dataset with 635,176 RGB images across 1167 plant species.
• Included diverse plant growth stages, structures, and environmental conditions.
• Benchmarked 10 Convolutional Neural Networks (CNNs), 6 Vision Transformers (ViTs), and 12 Vision-Language Models (VLMs).

Main Results:

• OpenPlant provides a comprehensive benchmark for agricultural plant classification.
• Evaluated the performance of various deep learning architectures on the dataset.
• Identified insights into the strengths and weaknesses of different models for plant recognition.

Conclusions:

• The OpenPlant dataset serves as a valuable resource for advancing deep learning in agriculture.
• The benchmark results offer guidance for future research and model development.
• Enables more accurate and robust plant classification for smart farming and precision agriculture.