A deep learning-powered web application that predicts plant diseases based on leaf images. The system uses image classification techniques and deploys the trained model via Streamlit for real-time user interaction.
- Data Collection: Gathered a labeled dataset of plant leaf images representing 38 different diseases.
- Image Preprocessing: Resized, normalized, and augmented images to enhance training efficiency and generalization.
- Data Separation: Divided the dataset into training, validation, and testing sets.
- Data Generators: Used
ImageDataGeneratorfor training, validation, and testing. - Parameter Tuning: Configured key training parameters including:
- Batch Size
- Learning Rate
- Number of Epochs
- Model Architecture: Implemented ResNet50 with a GlobalAveragePooling2D layer for robust image classification.
- Training: Used TensorFlow/Keras with generators to efficiently train the model on the processed data.
- Performance Metrics: Achieved
- ✅ Training Accuracy: 95.54%
- ✅ Test Accuracy: 94.78%
- Visualization: Plotted training/validation accuracy and loss curves to monitor performance.
- Inference: Used the trained model to predict disease categories on unseen test images.
- Evaluation: Compared predictions with true labels for accuracy assessment.
- Streamlit Web App:
Deployed the model using Streamlit for an intuitive web interface. - Features:
- Upload plant leaf images
- Get real-time disease predictions
- Simple and interactive UI
cv2,numpy,pandas,matplotlibTensorFlow,Keras,ResNet50StreamlitImageDataGenerator(for image augmentation)
- ✅ End-to-end pipeline: data → model → deployment
- 🌿 Focused on agricultural disease identification
- 📷 Image-based classification using transfer learning
- 🌐 Real-time web deployment for user accessibility
Developed by Meena M
Open to collaboration, feedback, and improvements!