Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
130 tokens/sec
GPT-4o
10 tokens/sec
Gemini 2.5 Pro Pro
46 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
3 tokens/sec
DeepSeek R1 via Azure Pro
55 tokens/sec
2000 character limit reached

iNatAg: Multi-Class Classification Models Enabled by a Large-Scale Benchmark Dataset with 4.7M Images of 2,959 Crop and Weed Species (2503.20068v1)

Published 25 Mar 2025 in cs.CV

Abstract: Accurate identification of crop and weed species is critical for precision agriculture and sustainable farming. However, it remains a challenging task due to a variety of factors -- a high degree of visual similarity among species, environmental variability, and a continued lack of large, agriculture-specific image data. We introduce iNatAg, a large-scale image dataset which contains over 4.7 million images of 2,959 distinct crop and weed species, with precise annotations along the taxonomic hierarchy from binary crop/weed labels to specific species labels. Curated from the broader iNaturalist database, iNatAg contains data from every continent and accurately reflects the variability of natural image captures and environments. Enabled by this data, we train benchmark models built upon the Swin Transformer architecture and evaluate the impact of various modifications such as the incorporation of geospatial data and LoRA finetuning. Our best models achieve state-of-the-art performance across all taxonomic classification tasks, achieving 92.38\% on crop and weed classification. Furthermore, the scale of our dataset enables us to explore incorrect misclassifications and unlock new analytic possiblities for plant species. By combining large-scale species coverage, multi-task labels, and geographic diversity, iNatAg provides a new foundation for building robust, geolocation-aware agricultural classification systems. We release the iNatAg dataset publicly through AgML (https://github.com/Project-AgML/AgML), enabling direct access and integration into agricultural machine learning workflows.

Summary

Insights into Fine-Grained Agricultural Classification with iNatAg

The paper introduces iNatAg, a comprehensive dataset aimed at advancing research in fine-grained classification of crop and weed species, with potential implications for precision agriculture and sustainable farming practices. The dataset, curated from iNaturalist, contains a vast image repository of over 4.7 million images representing 2,959 distinct crop and weed species, annotated with precise taxonomic labels ranging from broad binary classifications (crop versus weed) to specific species identification. This dataset serves as a pivotal resource for training models tailored for agricultural applications, with capabilities to accommodate real-world variability in captured images, thus addressing the practical limitations of existing agricultural datasets.

The primary contribution lies in the scale and diversity of iNatAg, facilitating robust model training and evaluation across diverse taxonomic levels. The Swin Transformer architecture, with its ability to handle multi-class and hierarchical classification tasks, forms the backbone for benchmarking these models. Experiments detail nine configurations of Swin Transformer models with varied image inputs and geospatial data and incorporate LoRA fine-tuning adaptations. The benchmarks demonstrate substantial performance achievements, notably a 92.38% accuracy in crop and weed classification and significant efficacy in narrower taxonomic tasks like species-level categorization (79.40%). This multi-layered classification approach signifies a leap towards accurate agricultural AI models equipped to handle fine-grained distinctions that are crucial under field conditions.

Numerical results emphasize the impact of various model architectures and design choices, such as the inclusion of geospatial metadata and model size variations. Larger Swin Transformer models outperformed smaller variants by leveraging high-capacity architectural features and finer image details, suggesting an inherent advantage in complex leaf and plant morphology differentiation critical to species-level classification. Conversely, LoRA fine-tuning offered noticeable improvement for large model configurations, indicating that adaptation layers are optimally utilized in resource-rich architectures rather than smaller ones where it can impair generalization.

Analysis of model performance across taxonomic classifications identifies that misclassifications are predominantly constrained within correct genera and families, supporting taxonomy-aware modeling approaches. This finding aligns with past research advocating for taxonomic hierarchy considerations in evaluation metrics, offering a foundation for designing more informed and holistic agricultural modeling strategies. Moreover, incorporating geospatial inputs consistently enhanced classification outcomes, capitalizing on region-specific patterns that are particularly potent in agricultural and ecological contexts.

The iNatAg dataset, made publicly available through AgML, invites future explorations in species-specific agricultural applications and provides a scaffold for developing geospatially informed agricultural AI systems. By introducing such a detailed, real-world agricultural image dataset, iNatAg holds promise for future enhancements in precision agriculture that demand high specificity and adaptability. Broad practical implications include improved crop health assessments, precision farm management, and invasive species control, representing a vital step towards operationalizing AI technologies in agriculture.

In conclusion, iNatAg establishes a robust benchmark dataset coupled with an intensive evaluation strategy, effectively augmenting the capacity for nuanced agricultural classification. With results paving the path for actionable AI models in agriculture, the dataset can serve as a catalyst for expanding research horizons into precision agriculture workflows and taxonomy-driven AI methodologies — fostering progress in achieving sustainable farming practices globally.

Dice Question Streamline Icon: https://streamlinehq.com

Follow-up Questions

We haven't generated follow-up questions for this paper yet.