An Expert System for Disease Prediction and Fertilizer Recommendation using Transfer Learning

Abstract

The agricultural production of tomatoes is vulnerable to a range of diseases. To make the matters worse, imprecise fertilizers application also aggravates diseases that results in poor harvests for farmers. It however does not only connote but also exists a tomato plant disease detection system with different advanced deep learning models. Their only problem is that they always give disease detection but not fertilizer recommendations. This paper will try to fill this gap by coming up with a hybrid expert system for disease detection and user-friendly efficient fertilizer recommendations for tomato plants. The proposed expert system will employ advanced deep learning techniques such as MobileNetV2 and DenseNet121, DenseNet201 for disease recognition and feature extraction. Moreover, fertilization recommendation for detected disease based on rules. Dataset of “tomato village” is used for training and validation. The expert system proposed for disease diagnosis and fertilizer recommendation aims to transform tomato cultivation, enhancing crop yield, reducing environmental impact, and promoting agricultural sustainability. Its implementation promises a brighter future for tomato production, offering holistic solutions to farmers' challenges and advancing the agricultural industry.

Specification

Description:Detailed Description of the Invention
1. Introduction: The invention is a hybrid expert system designed to improve tomato cultivation through the integration of advanced deep learning models for disease detection with a rule-based system for precise fertilizer recommendations. This system addresses common challenges faced by farmers, such as diagnosing plant diseases and optimizing fertilizer application, to enhance crop yield and sustainability.
2. Components of the System:
a. Disease Detection Module:
• Deep Learning Models:
o MobileNetV2: A lightweight deep learning model optimized for mobile and embedded vision applications. It provides efficient and accurate disease detection by using depthwise separable convolutions, reducing computational load while maintaining performance.
o DenseNet121 and DenseNet201: These models utilize dense connections between layers to improve feature propagation and reduce vanishing gradients. DenseNet121 and DenseNet201 are used for more accurate disease recognition and feature extraction, leveraging their deeper architectures for enhanced diagnostic capabilities.
• Image Input: The system accepts images of tomato plants taken under various conditions. These images are processed by the deep learning models to identify symptoms indicative of specific diseases.
• Disease Classification: The models analyze the images to classify the diseases based on learned features and patterns. The output includes a diagnosis of the disease affecting the plant, providing critical information for subsequent steps.
b. Fertilizer Recommendation Module:
• Rule-Based System:
o Integration with Disease Detection: Once a disease is identified, the system uses a rule-based approach to recommend appropriate fertilizer treatments. The rules are based on a database of disease-specific requirements and fertilizer efficacy.
o Recommendation Engine: The engine processes the disease diagnosis and generates recommendations for fertilizer types and application rates. These recommendations are designed to address both the nutritional needs of the plant and the disease management requirements.
• User Interface: A user-friendly interface allows farmers to input images of their crops and receive recommendations. The interface is designed to be accessible, with clear instructions and actionable advice for fertilizer application.
c. Data Utilization and Training:
• Dataset: The "tomato village" dataset is used for training and validating the deep learning models. This dataset includes a diverse range of images of tomato plants affected by various diseases, providing a robust foundation for model development.
• Training Process: The models are trained using a supervised learning approach, where the images are labeled with disease types. The training process involves optimizing model parameters to improve accuracy and reduce errors in disease detection.
3. Workflow of the System:
• Image Capture: Farmers capture images of their tomato plants using a camera or smartphone.
• Disease Detection: The captured images are uploaded to the system, where they are processed by the deep learning models to detect and classify any diseases present.
• Fertilizer Recommendation: Based on the disease diagnosis, the rule-based system generates tailored fertilizer recommendations. These recommendations are presented to the farmer through the user interface.
• Actionable Advice: Farmers apply the recommended fertilizers and follow any additional guidance provided by the system to manage the identified diseases and improve plant health.
4. Benefits:
• Improved Accuracy: The use of advanced deep learning models ensures high accuracy in disease detection, reducing the likelihood of misdiagnosis.
• Optimized Fertilization: The integration of disease detection with fertilizer recommendations helps optimize nutrient application, leading to better plant health and higher yields.
• Sustainability: By providing precise recommendations, the system reduces the environmental impact of over-fertilization and promotes more sustainable agricultural practices.
• Farmer Support: The system offers a comprehensive tool that simplifies the process of managing tomato crops, supporting farmers in achieving better outcomes with less effort.
5. Conclusion: This invention represents a significant advancement in agricultural technology by combining sophisticated disease detection with targeted fertilization recommendations. The hybrid expert system provides a holistic solution to the challenges of tomato cultivation, aiming to enhance crop productivity, support sustainable farming practices, and improve overall agricultural efficiency.








Operation:
1. System Setup and Initialization:
• System Installation: The hybrid expert system is installed on a compatible device or server with sufficient computational resources to run deep learning models. It includes a user interface (UI) for farmers to interact with the system.
• Model Training: The deep learning models (MobileNetV2, DenseNet121, DenseNet201) are pre-trained using the "tomato village" dataset. This dataset consists of annotated images of tomato plants affected by various diseases. The models are optimized for accurate disease detection and classification.
2. Image Capture:
• Image Acquisition: Farmers capture images of their tomato plants using a camera or smartphone. These images should be clear and taken under good lighting conditions to ensure accurate detection.
3. Image Upload and Preprocessing:
• Uploading Images: The captured images are uploaded to the system through the user interface. The system supports various image formats and allows for batch uploads if needed.
• Preprocessing: Before analysis, the images undergo preprocessing to enhance quality and ensure compatibility with the deep learning models. This may include resizing, normalization, and noise reduction.
4. Disease Detection:
• Image Analysis: The preprocessed images are fed into the deep learning models (MobileNetV2, DenseNet121, DenseNet201). Each model analyzes the image to detect and classify the presence of diseases based on learned features and patterns.
• Disease Classification: The models output a diagnosis indicating the type of disease affecting the tomato plant. The system combines results from different models to improve accuracy and reduce the likelihood of false positives or negatives.
5. Fertilizer Recommendation:
• Disease Diagnosis Integration: The detected disease information is passed to the rule-based recommendation engine. This engine uses a database of rules that correlate specific diseases with appropriate fertilizer treatments.
• Recommendation Generation: Based on the disease diagnosis, the recommendation engine generates tailored fertilizer advice. This includes the type of fertilizer, application rate, and any additional instructions for optimal effectiveness.
• User Interface Display: The recommendations are displayed on the user interface, providing clear and actionable advice for the farmer. The system may also include visual aids or charts to help farmers understand the recommendations better.
6. Implementation of Recommendations:
• Farmer Action: Farmers use the recommendations to apply the appropriate fertilizers to their tomato plants. The system may also offer guidance on how to apply the fertilizers and any other necessary management practices to address the detected diseases.
7. Monitoring and Feedback:
• Follow-Up: The system may include options for farmers to provide feedback on the effectiveness of the recommendations. This feedback can be used to refine the recommendation rules and improve system performance over time.
• Continuous Learning: The system may incorporate mechanisms for continuous learning, where new data and feedback are used to update and enhance the deep learning models and recommendation rules.
8. Reporting and Analytics:
• Data Collection: The system collects data on disease diagnoses, fertilizer recommendations, and crop outcomes. This data can be used for reporting and analysis to monitor system performance and provide insights into disease management practices.
• Analytics: Advanced analytics can be applied to identify trends, assess the impact of recommendations on crop yield and health, and optimize the system's overall effectiveness.
9. User Support:
• Help and Documentation: The system includes help resources, user guides, and support options to assist farmers in using the system effectively.
• Technical Support: For any issues or queries, technical support is available to ensure smooth operation and address any challenges faced by users.
, Claims:1) The hybrid expert system of claim 1, wherein the disease detection module utilizes one or more deep learning models
selected from the group consisting of MobileNetV2, DenseNet121, and DenseNet201
2) The hybrid expert system of claim 1, wherein the fertilizer recommendation module includes:
• a rule-based engine configured to generate fertilizer recommendations based on the identified disease types; and
• a user interface for displaying the fertilizer recommendations to the farmer.
3) The hybrid expert system of claim 1, wherein the disease detection module integrates results from multiple
deep learning models to improve accuracy and reduce false positives or negatives.

Documents