Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Naitik Jain, Amogh Joshi, Mason Earles

TL;DR

The paper introduces iNatAg, a large-scale agricultural image dataset with 4.72 million images spanning 2,959 crop and weed species, annotated across species to crop/weed levels and enriched with geolocation metadata. It benchmarks Swin Transformer models of varying sizes, evaluates LoRA fine-tuning, and demonstrates that incorporating geospatial features improves taxonomic classification, achieving up to 92.38% crop/weed accuracy and 79.40% species accuracy. The dataset is designed to support multi-task learning and taxonomy-aware evaluation, addressing real-world variability and enabling robust, geolocation-aware agricultural AI. This work provides a foundation for scalable, region-aware plant classification systems with practical impact on precision agriculture and sustainable farming.

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.

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

TL;DR

The paper introduces iNatAg, a large-scale agricultural image dataset with 4.72 million images spanning 2,959 crop and weed species, annotated across species to crop/weed levels and enriched with geolocation metadata. It benchmarks Swin Transformer models of varying sizes, evaluates LoRA fine-tuning, and demonstrates that incorporating geospatial features improves taxonomic classification, achieving up to 92.38% crop/weed accuracy and 79.40% species accuracy. The dataset is designed to support multi-task learning and taxonomy-aware evaluation, addressing real-world variability and enabling robust, geolocation-aware agricultural AI. This work provides a foundation for scalable, region-aware plant classification systems with practical impact on precision agriculture and sustainable farming.

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.

Paper Structure

This paper contains 23 sections, 1 equation, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. The iNatAg Dataset
  3. Dataset Development
  4. Dataset Diversity and Expansiveness
  5. Accessing iNatAg via AgML
  6. Experiments
  7. Model Architecture & Training Strategy
  8. LoRA Fine-Tuning.
  9. Geospatial Features.
  10. Data Preprocessing & Augmentation
  11. Training Objective
  12. Implementation Details
  13. Model Evaluation & Inference
  14. Results and Discussion
  15. Design Analysis
  16. ...and 8 more sections

Figures (5)

  • Figure 1: Examples of 40 crop species from the iNatAg dataset. This visual snapshot highlights variation in leaf shape, color, and background due to real-world, user-generated data.
  • Figure 2: Examples of 40 weed species from the iNatAg dataset. This snapshot demonstrates the visual complexity introduced by growth stage variation, occlusion, and environmental background.
  • Figure 3: Visual diversity within species and real-world crop–weed associations. In sequential order, rows illustrate variations in Phaseolus vulgaris (crop), Amaranthus retroflexus (weed), Sorghum bicolor (crop), and Cyperus rotundus (weed).
  • Figure 4: Global density map of the iNatAg dataset. Colors represent a continuous density scale, with dark red indicating more images and blue indicating fewer images. The iNatAg dataset spans multiple continents, reflecting broad ecological and geographic diversity.
  • Figure 5: Comparison of confusion matrices for crops and weeds.