The system combines deep learning with environmental sensing and cloud connectivity, achieving high accuracy while sending instant alerts to farmers via a customizable dashboard, mobile app, or SMS.
Designed for early intervention and reduced crop loss, it offers a scalable approach to sustainable farming with minimal human input.
Background
Plant diseases remain a serious threat to global food security, accounting for up to 30 % of crop losses each year.
Traditional detection methods, which are largely manual, are slow, labor-intensive, and often fail to catch outbreaks early. While deep learning and IoT technologies offer a way forward, most current solutions are either too costly, lack mobility, or aren't practical for real-time field deployment. This is even more true for smaller-scale farms.
Aiming to tackle those limitations head-on, this study developed a solar-powered autonomous robot that integrates a deep convolutional neural network (CNN) for real-time image-based disease diagnosis, paired with environmental sensors.
The result is a practical, cost-effective system designed for precision agriculture in resource-limited settings.
System Design and Methodology
The system follows a secure, multi-layer IoT architecture:
- Perception Layer: High-resolution cameras and environmental sensors capture plant images and collect soil and climate data.
- Network Layer: Data is securely transmitted using MQTT protocols.
- Edge Processing: A Raspberry Pi processes data locally using an optimized ResNet-50-based CNN for rapid classification.
- Application Layer: Alerts and analytics are made accessible through a farmer-friendly cloud dashboard, app, and SMS.
The robot operates on a ground-based platform for stable, close-range monitoring. It uses a modular pipeline: images are preprocessed (e.g., noise reduction, contrast enhancement), then classified in real time by the CNN. The model was optimized for embedded hardware using quantization and pruning, achieving low-latency inference (220 ms/image) while maintaining 95.8 % classification accuracy in field conditions.
The entire system also runs on a solar-powered battery, ensuring extended, off-grid operation. In field tests on a tomato farm, the robot successfully navigated semi-arid terrain, maintained high detection accuracy, and sent geotagged alerts. All in all, the system was able to successfully demonstrate its effectiveness in real-world conditions.
Implementation and Evaluation
Model training was conducted on a high-performance computing system using a custom dataset of 87,000 leaf images, including publicly available PlantVillage images and augmented versions to improve diversity.
The dataset covered 38 classes of healthy and diseased plants, with preprocessing steps such as resizing and segmentation to standardize inputs.
A significant portion of the study focused on optimizing model performance:
- Baseline Accuracy: 19.7 % (without optimization)
- With Particle Swarm Optimization (PSO): 29.4 %
- PSO + ADAM Optimizer: 41.8 %
- ADAM Alone: 99.1 %
- Grey Wolf Optimizer (GWO) + ADAM: 99.3 %
The final model achieved strong generalization:
- Training Accuracy: 99.39 %
- Validation Accuracy: 97.47 %
- Testing Accuracy: 97.13 %
- F1 Score: 99.46 %
Most misclassifications occurred between visually similar diseases like early and late blight. But error rates remained under 2 %. The team addressed class imbalance with data augmentation and mitigated overfitting using L2 regularization and dropout.
Limitations include the system’s current inability to detect unknown diseases outside the training set, sensitivity to lighting and image quality, and navigation constraints in more complex terrains. These are areas targeted for future updates, including continuous learning and multi-modal sensing.
Conclusion
This research presents a solar-powered autonomous robot that effectively merges deep learning and IoT for real-time crop disease detection.
With over 97 % accuracy and real-world validation, the system stands out as a mobile, cost-effective tool for precision agriculture. Farmers receive instant alerts via cloud dashboards, apps, or SMS, allowing for timely intervention.
Future enhancements will focus on navigating complex environments, expanding the range of detectable diseases, and improving robustness in varied lighting and weather conditions. The prototype achieved 86 % average energy efficiency and is projected to decrease in cost from $650 to $450–$500 with scaled production, making it an accessible solution for small to medium-sized farms.
Journal Reference
Talaat, F. M., Ibrahim, M. A., Karim, A. A., Elsonbaty, H. K., & Al-Zoghby, A. M. (2026). IoT-Integrated robotic system for automated plant disease detection and environmental monitoring. Scientific Reports, 16(1). DOI:10.1038/s41598-025-32624-4. https://www.nature.com/articles/s41598-025-32624-4
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.