kvpratama/crop-disease-prediction

Crop Disease Prediction Web App

Crop Disease Prediction Web App

Video Demo: https://youtu.be/o4mBi-VXlFU

Description:

This web app, built using Flask, allows users to predict crop diseases from images. It utilizes a machine learning model hosted on Hugging Face, which was trained on a dataset obtained from Kaggle. The main features include image uploading, predicting crop diseases, viewing prediction history, submitting feedback, and an admin page for managing user feedback.

Features

• Crop Disease Prediction: Users can upload images of crops, and the web app predicts the likelihood of crop diseases using a pre-trained machine learning model.

• Prediction History: Users can view a history of their uploaded images alongside the corresponding predictions.

• Feedback Submission: Users can submit feedback if they believe the prediction is incorrect.

• Admin Page: Administrators can access an admin page to review user feedback about predictions.

Dataset

The dataset used for training the machine learning model was sourced from Kaggle. It consists of images labeled with various crop diseases.

Model Training

The machine learning model was trained on the dataset, and the resulting model was uploaded to the Hugging Face model repository.

Web App

• The Flask web application allows users to upload images of crops.
• The uploaded images are sent to Hugging Face through the Gradio API to obtain predictions using the pre-trained model.

Predict Function

• Handles the prediction process for uploaded crop images.
• Upon receiving a POST request with an uploaded image, the function:
  • Checks for the presence of the image file.
  • Generates a timestamp, creates a new filename, and saves the file to the uploads folder.
  • Utilizes the Hugging Face Gradio API to obtain predictions for the uploaded image.
  • Inserts the image information, prediction, and JSON result into the database.
  • Extracts labels and confidences from the prediction result.
  • Renders the result page with the image, labels, and a chart displaying prediction confidences.
• Returns:
  • Renders the 'result.html' template with the prediction result.

History Page

• Displays the prediction history for the logged-in user.
• Retrieves and displays the user's uploaded images and their corresponding predictions from the database. Additionally, retrieves information about available classes for potential user's feedback.
• Returns:
  • Renders the 'history.html' template with the user's image history and available classes.

Detail Page

• Displays detailed information about a specific image prediction.
• If the request method is POST, retrieves the image details based on the provided image_id.
• Extracts information such as file path, labels, and confidences from the prediction result JSON.
• Renders the 'result.html' template with the detailed information.
• Returns:
  • Renders the 'result.html' template with the detailed information for the selected image.

Suggest function

• Handles user suggestions for correcting predictions.
• If the request method is POST, retrieves the image_id and suggested prediction from the form.
• Updates the database with the user-suggested prediction for the corresponding image_id.
• Redirects the user to the '/feedback' page.
• Returns:
  • Redirects to the '/feedback' page after processing the user suggestion.

User's Feedback

• Displays the feedback history for the logged-in user.
• Retrieves and displays images for which users have provided feedback, indicating a disagreement with the prediction.
• Only includes images with non-null values in the 'prediction_user' column.
• Returns:
  • Renders the 'feedback.html' template with the user's feedback history.

Admin's user's feedback page

• Displays the feedback history for all users (admin-only view).
• Retrieves and displays images for which users have provided feedback, indicating a disagreement with the prediction. -Includes additional information such as the image path, original prediction, user-suggested prediction, and the username of the uploader. -Returns:
  • Renders the 'feedback-users.html' template with the feedback history for all users.

Design Choices

• Hugging Face Integration: Leveraging the Hugging Face model repository allows for easy model deployment and updates without the need to host the model locally.

• Gradio API: The use of Gradio simplifies the integration with the Hugging Face model and provides a user-friendly interface for predictions.

• Feedback System: Including a feedback system enables users to contribute to model improvement, enhancing the accuracy of predictions over time.

Acknowledgments

• Kaggle for providing the dataset.
• Hugging Face for hosting the machine learning model.