A Comparative Evaluation of Deep Learning Architectures for Weed Classification, with an Exploratory Out-of-Distribution Analysis of Albanian Field Images

Summary

Kumar et al. (2026) compared six deep learning configurations for nine-class weed classification on a stratified subset of the DeepWeeds benchmark, using a leakage-checked train/validation/test split and repeated training across multiple random seeds. They found that fine-tuned DINOv2 and EfficientNet-B0 performed best overall, but that differences among several high-performing models were not consistently statistically significant across seeds. An exploratory out-of-distribution probe on unlabelled Albanian field images highlighted geographic domain shift and the need for locally collected, labelled datasets before reliable regional deployment.

Practical implication. Those planning site-specific weed management in new regions such as the Western Balkans need locally collected and labelled data before deploying models trained on a public benchmark.

The Problem

Weed-classification studies usually rank deep learning architectures from single training runs, which can overstate differences between models and says nothing about transfer to new farming regions.

The Methodology

Comparative deep learning experiment on the DeepWeeds benchmark (nine classes) with leakage-checked splits, repeated multi-seed training of six architectures, statistical model comparison, and an exploratory out-of-distribution probe on unlabelled Albanian field images.

The Core Finding

Fine-tuned DINOv2 and EfficientNet-B0 performed best on nine-class DeepWeeds data, but differences among several high-performing models were not consistently statistically significant across random seeds. An exploratory test on Albanian field images revealed a clear geographic domain shift, so locally collected and labelled data are needed before deployment.

The Citation

Kumar, R., Vajjhala, N. R., Ramollari, E., & Joca, E. (2026). A Comparative Evaluation of Deep Learning Architectures for Weed Classification, with an Exploratory Out-of-Distribution Analysis of Albanian Field Images. Information, 17(8), Article 751. https://doi.org/10.3390/info17080751 Copy APA

The summary, key findings, methods, implications and relevance notes on this page are this website’s own description of the paper, written from the published abstract and text. The official abstract and citation details are given further down.

The study in detail

What question does this paper answer?

Which modern deep learning architectures classify weed species most reliably, how stable are their rankings across random seeds, and do models trained on a public benchmark transfer to field images from Albania?

What did the study find, in detail?

Across a multi-seed evaluation of six configurations on nine-class DeepWeeds data, fine-tuned DINOv2 and EfficientNet-B0 performed best overall — but the differences among several high-performing models were not consistently statistically significant across seeds. An exploratory test on unlabelled Albanian field images showed a clear geographic domain-shift problem.

Why does it matter?

Single-run leaderboard comparisons can overstate differences between models; this study shows why multi-seed evaluation, leakage-checked splits, and statistical testing matter for agricultural computer vision. For site-specific weed management in new regions such as the Western Balkans, locally collected and labelled data are needed before deployment.

Methods and data

Standards, datasets and methods named in this study

• DeepWeeds Benchmark dataset
  The benchmark from which the nine-class evaluation subset was drawn.
  Olsen, A., Konovalov, D. A., Philippa, B., et al. (2019). DeepWeeds: A multiclass weed species image dataset for deep learning. Scientific Reports, 9, 2058. https://doi.org/10.1038/s41598-018-38343-3
• DINOv2 Model
  One of the architectures compared, with linear probing and with full fine-tuning.
  Oquab, M., Darcet, T., Moutakanni, T., et al. (2023). DINOv2: Learning robust visual features without supervision. arXiv:2304.07193.
• EfficientNet Model
  EfficientNet-B0 is one of the architectures compared.
  Tan, M., & Le, Q. V. (2019). EfficientNet: Rethinking model scaling for convolutional neural networks. Proceedings of the 36th International Conference on Machine Learning (PMLR 97).

Definitions and key terms

Out-of-distribution (OOD) data

Data that differ systematically from a model’s training data — here, field images from a different country, soil, and lighting than the benchmark. — as used in Kumar et al. (2026)

Multi-seed evaluation

Training and testing each model several times with different random initializations to check whether performance differences are stable rather than due to chance. — as used in Kumar et al. (2026)

Linear probing vs. fine-tuning

Linear probing trains only a new output layer on top of a frozen pretrained model; full fine-tuning updates all of the model’s weights. — as used in Kumar et al. (2026)

All key terms across the publications →

Abstract

Automated weed classification supports site-specific weed management by enabling targeted interventions and reducing environmental and operational costs. This study presents a multi-seed evaluation of six deep learning configurations—YOLO26n-cls, Vision Transformer (ViT-B/16), DINOv2 (ViT-S/14) with linear probing and full fine-tuning, ResNet-50, and EfficientNet-B0—for nine-class weed classification using a stratified subset of the DeepWeeds benchmark. The models were evaluated using a leakage-checked train/validation/test split and repeated training across multiple random seeds to assess performance stability and statistical reliability. Among the evaluated approaches, DINOv2-FT and EfficientNet-B0 demonstrated the strongest overall performance, while statistical analysis showed that differences among several high-performing models were not consistently significant across seeds. The study also identifies the importance of appropriate transfer-learning strategies and hyperparameter selection, demonstrating that model performance can be strongly affected by optimisation choices. A reproducibility issue affecting ViT-B/16 was identified and transparently reported, leading to its exclusion from multi-seed statistical comparisons. An exploratory out-of-distribution probe using unlabelled Albanian field images further highlights the challenges of geographic domain shift and the need for locally collected, labelled datasets before reliable regional deployment. Overall, this work provides a systematic comparison of modern deep learning architectures for weed classification and emphasises reproducibility, statistical validation, and careful interpretation of model rankings.

Abstract as published in Information under a Creative Commons licence.

Limitations

• The Albanian field images were unlabelled, so the out-of-distribution analysis is exploratory.
• ViT-B/16 was excluded from multi-seed statistical comparisons because of a reproducibility issue.

Practical implications

• Researchers working on weed detection and precision agriculture in new regions should collect and label local data before deployment, because an exploratory probe on unlabelled Albanian field images highlighted geographic domain shift.
• For researchers in agricultural computer vision, the study shows why multi-seed evaluation, leakage-checked splits and statistical testing matter: single-run leaderboard comparisons can overstate differences between models.
• For researchers choosing among high-performing weed classification models, fine-tuned DINOv2 and EfficientNet-B0 were strongest overall, but differences among several high-performing models were not consistently significant across seeds.
• For researchers training models for weed classification, the study identifies the importance of appropriate transfer-learning strategies and hyperparameter selection: model performance can be strongly affected by optimisation choices.

Drawn from Kumar et al. (2026), Information, 17(8), Article 751.

When this research may be relevant

This article may be relevant to researchers comparing convolutional networks, Vision Transformers and self-supervised backbones such as DINOv2 for agricultural image classification, to work on weed detection and precision agriculture, and to studies of reproducibility, seed variance and domain shift in applied computer vision. It also speaks to the need for locally collected datasets in South-Eastern Europe.

Research topics addressed: weed classification; deep learning; computer vision; precision agriculture; DeepWeeds dataset; DINOv2; EfficientNet; Vision Transformer; YOLO; ResNet-50; transfer learning; reproducibility; out-of-distribution generalization; domain shift; Albania

How to cite

APA 7

Kumar, R., Vajjhala, N. R., Ramollari, E., & Joca, E. (2026). A Comparative Evaluation of Deep Learning Architectures for Weed Classification, with an Exploratory Out-of-Distribution Analysis of Albanian Field Images. Information, 17(8), Article 751. https://doi.org/10.3390/info17080751Copy APA citation

MLA 9

Kumar, Rajesh, et al. “A Comparative Evaluation of Deep Learning Architectures for Weed Classification, with an Exploratory Out-of-Distribution Analysis of Albanian Field Images.” Information, vol. 17, no. 8, 2026, article 751. https://doi.org/10.3390/info17080751.Copy MLA citation

IEEE

R. Kumar, N. R. Vajjhala, E. Ramollari, and E. Joca, “A Comparative Evaluation of Deep Learning Architectures for Weed Classification, with an Exploratory Out-of-Distribution Analysis of Albanian Field Images,” Information, vol. 17, no. 8, Art. no. 751, Aug. 2026, doi: 10.3390/info17080751.Copy IEEE citation

@article{kumar2026deep,
  title = {A Comparative Evaluation of Deep Learning Architectures for Weed Classification, with an Exploratory Out-of-Distribution Analysis of Albanian Field Images},
  author = {Kumar, Rajesh and Vajjhala, Narasimha Rao and Ramollari, Ervin and Joca, Earta},
  journal = {Information},
  volume = {17},
  number = {8},
  pages = {751},
  year = {2026},
  publisher = {MDPI},
  doi = {10.3390/info17080751},
  url = {https://doi.org/10.3390/info17080751}
}

TY  - JOUR
TI  - A Comparative Evaluation of Deep Learning Architectures for Weed Classification, with an Exploratory Out-of-Distribution Analysis of Albanian Field Images
AU  - Kumar, Rajesh
AU  - Vajjhala, Narasimha Rao
AU  - Ramollari, Ervin
AU  - Joca, Earta
JO  - Information
PY  - 2026
DA  - 2026/08/02
VL  - 17
IS  - 8
SP  - 751
PB  - MDPI
SN  - 2078-2489
DO  - 10.3390/info17080751
UR  - https://www.narasimharao.net/research/deep-learning-weed-classification-deepweeds-albania/
LA  - en
ER  - 

Links and files

• DOI: https://doi.org/10.3390/info17080751
• Publisher’s page: https://www.mdpi.com/2078-2489/17/8/751
• Full text (PDF): https://www.mdpi.com/2078-2489/17/8/751/pdf
• This page as text: https://www.narasimharao.net/research/deep-learning-weed-classification-deepweeds-albania/paper.md
• Permanent address of this page: https://www.narasimharao.net/research/deep-learning-weed-classification-deepweeds-albania/