AI-Driven Sugarcane Diseases Detection & Remedy Suggestion with Flask App Final Year Project

The following contents are Downloadable immediately after the successful payment

• Source Code

• Flask Web App

• Dataset

• Trained Model

• Project Report (Both PDF and Word files)

• Instructions to install the necessary Software and Libraries

• Step-by-step instructions to execute the project (readme file)

Project Description

• Data Augmentation: Enhanced dataset variation through rotation, flipping, zooming, contrast changes.
• Preprocessing: Images resized to 224×224, normalized for DenseNet201 input.
• Feature Extraction: DenseNet201 pretrained on ImageNet extracts 1920-dimensional vectors capturing leaf features.
• Classification: Support Vector Machine with RBF kernel classifies features into Healthy, Red Rot, or Red Rust.
• Training: Combined DenseNet201+SVM trained with stratified 5-fold cross-validation; achieved training accuracy of 98% and validation accuracy of 97.78%.
• Deployment: Flask web app integrates model inference, image upload interface, and remedy suggestions.

System Architecture

• Model Layer: DenseNet201 extracts features, SVM performs classification.
• Backend: Flask handles image upload, inference, and serves results.
• Frontend: HTML templates for image upload and displaying disease prediction with remedies.
• Remedies: Actionable management tips linked to online product purchases.

Installing required Software and Running the Project

Install Anaconda Python

Open Anaconda Prompt, navigate to project folder:
cd path_to_your_project_folder
Example:
C:\>D:
C:\>cd D:\Sugarcane-Leaf-Disease-Detection

Create and activate environment:
conda create -n sdd python=3.9
conda activate sdd

Install dependencies
pip install -r requirements.txt

Train the model (Optional):
jupyter notebook
Open Sugarcane-Leaf-Disease-Detection.ipynb and run all cells.

Launch the Flask web app:
python app.py

Results & Discussion

• Classification report shows ~94% precision, recall, and F1-score overall.
• Healthy leaves were classified most accurately; some overlap between Red Rot and Red Rust due to symptom similarity.
• Compared to baseline CNN models (78-91% accuracy), DenseNet201+SVM achieved superior generalization.
• Data augmentation was key to performance on small datasets.
• Web app enables real-time use by farmers, bridging diagnosis to action.

Limitations

• Dataset limited to 225 images and three classes.
• Disease coverage excludes some sugarcane diseases (Smut, Mosaic Virus, etc.).
• Remedies currently provided in English only.
• Model robustness yet to be tested extensively under varied real-world conditions.

Future Work

• Expand datasets with diverse regions, more diseases, and severity levels.
• Introduce mobile and IoT-based deployment for field usage.
• Incorporate explainable AI visualization methods for interpretability.
• Develop multilingual user interfaces for wider accessibility.
• Cloud deployment for scalable monitoring and analysis.

Happy Learning

Still need help to set up and execute the project

• Setup and modification are paid services based on requirements.