cyril1010 / paddyml Goto Github PK

This project implements a deep learning model using Convolutional Neural Networks (CNN) and ResNet50 architecture to classify different diseases in rice crops. The model is trained on a dataset containing images of rice leaves affected by various diseases like blast, blight, spot, and tungro.

• Image Preprocessing: Utilizes Keras for image preprocessing, including data augmentation and normalization.
• Model Training: Loads the pre-trained ResNet50 model and trains a new model on top of it, freezing the base layers and adding custom dense layers for classification.
• Model Evaluation: Evaluates the trained model on a test dataset to measure accuracy and loss.
• Prediction: Predicts the class of rice disease in both local images and images from online links.
• Model Saving and Conversion: Saves the trained model to Google Drive and converts it to TensorFlow Lite (.tflite) format for deployment on mobile devices.

• Access Dataset: Mounts Google Drive to access the dataset containing training, testing, and validation images.
• Load Trained Model: Loads the pre-trained model from Google Drive for inference.
• Image Preprocessing: Preprocesses images using Keras ImageDataGenerator.
• Train Model: Trains the model using the pre-trained ResNet50 base and custom dense layers.
• Test Image Links: Provides links to sample images for testing the model.
• Save Model: Saves the trained model to Google Drive for future use.
• Convert Model: Converts the saved .h5 model to TensorFlow Lite (.tflite) format for deployment.

• TensorFlow
• Keras
• OpenCV
• Matplotlib
• Requests
• PIL

1. Ensure all necessary libraries are installed.
2. Mount Google Drive to access the dataset and save the trained model.
3. Load the pre-trained model or train a new one using the provided dataset.
4. Evaluate the model's performance and save it to Google Drive.
5. Convert the saved model to TensorFlow Lite format for deployment.

This project is licensed under the MIT License.

• This project utilizes the power of deep learning and transfer learning with the ResNet50 architecture.

For any inquiries or suggestions, please contact:

• Email: [email protected]
• LinkedIn: https://www.linkedin.com/in/cyril1010

• Name: Cyril K U