Related Experiment Videos
Comparative evaluation of deep learning models for plant disease classification with edge-aware performance analysis
Zirije Hasani1, Jakup Fondaj2, Dionesa Bytyqi1
1Faculty of Computer Science, University "Ukshin Hoti" Prizren, Prizren, Kosovo.
Plos One
|June 18, 2026
Summary
For precision farming, YOLOv8-cls offers the best balance of accuracy and speed for plant disease classification, outperforming other deep learning models like ResNet50 and DenseNet121 on resource-constrained platforms.
Area of Science:
- Agricultural Science
- Computer Vision
- Machine Learning
Background:
- Precision farming requires efficient agricultural disease monitoring.
- Deploying computer vision on resource-constrained platforms presents challenges.
- Deep learning models are increasingly used for plant disease classification.
Purpose of the Study:
- To comparatively evaluate deep learning architectures for multi-class plant disease classification.
- To assess performance trade-offs between accuracy and inference latency.
- To provide a statistically validated benchmarking framework for edge deployment.
Main Methods:
- Evaluated ResNet50, DenseNet121, Binarized Neural Network (BNN), and YOLOv8-cls.
- Utilized the PlantVillage dataset (15 classes).
- Performed 5 runs of stratified experiments with statistical validation (mean ± std dev).
Main Results:
- DenseNet121 achieved high accuracy (99.48%) but with high latency.
- BNN had minimal latency but poor performance (88.31%).
- YOLOv8-cls demonstrated the best trade-off: 99.64% accuracy with 3.3 ms latency.
Conclusions:
- YOLOv8-cls significantly outperforms ResNet50 with lower inference time.
- The PlantVillage dataset has generalization limitations for real-world edge deployment.
- A statistically grounded, edge-aware benchmarking framework for plant disease classification is presented.