Related Experiment Videos
Deep convolutional models for robust multi-crop disease recognition in real-world conditions.
Anubhava Srivastava1, Amit Bhola2, Jaroslav Frnda3,4,5
1Department of Computer Science and Engineering, Faculty of Engineering Sciences & Technology, Adani University, Ahmedabad, India.
Scientific Reports
|May 12, 2026
Summary
A new deep learning system identifies crop diseases from leaf images using MobileNetV3. This web-based tool provides disease names and treatments, aiding farmers globally.
Area of Science:
- Agricultural Science
- Computer Science
- Artificial Intelligence
Background:
- Crop diseases cause significant agricultural losses globally.
- Limited availability of diagnostic experts exacerbates the problem in many regions.
- Deep learning offers potential for automated plant disease identification from images.
Purpose of the Study:
- To develop and deploy a web-based deep learning system for detecting diseases in multiple crops.
- To compare the performance of different deep learning architectures for crop disease classification.
- To create a user-friendly, multilingual platform for farmers to access disease diagnosis and treatment information.
Main Methods:
- Compared MobileNetV3, EfficientNetB4, and ResNet50 using transfer learning on public datasets.
- Integrated MobileNetV3 as the primary model due to its balance of accuracy, speed, and deployability.
- Incorporated a super-resolution preprocessing step (Real-ESRGAN) to handle varying image quality.
- Utilized Grad-CAM for visualizing disease-affected areas, enhancing model interpretability.
- Developed a multilingual Progressive Web Application (PWA) for user image submission and AI-driven results.
- Employed a Large Language Model (LLM) to generate treatment options based on structured disease metadata.
Main Results:
- MobileNetV3 demonstrated superior performance considering classification accuracy, prediction time, and deployment feasibility.
- The super-resolution step improved handling of diverse image qualities.
- Grad-CAM effectively highlighted disease-specific regions in leaf images.
- The PWA successfully integrated the deep learning model and LLM for disease prediction and treatment recommendations.
- The system achieved good performance on test sets, despite acknowledged dataset biases.
Conclusions:
- The developed system serves as a practical digital agriculture decision-support tool.
- The system demonstrates the feasibility of deploying deep learning for crop disease detection via a PWA.
- Further field-level validation and model improvements are necessary to address real-world complexities and dataset limitations.