The Defect Detection App is a web application that utilizes machine learning models to predict whether the top part of a submersible pump impeller is defective or not. The models are trained on images of the top part of a submersible pump impeller using a pre-trained ResNet model for feature extraction and a custom defect detection model for classification. The app has been deployed on Streamlite.
Note: Deployed version of the web pages Here
- Image Upload: Users can upload an image of the top part of a submersible pump impeller for defect detection.
- Relevance Check: The app checks if the uploaded image is relevant by comparing its features to a pre-defined clustering model.
- Prediction: After uploading the image, users can click the "Predict" button to get the model's prediction regarding the defect status.
This project has used the some packages such as numpy, tensorflow, which have to be installed to run this web app locally present in requirements.txt file.
To run the project locally, there is a need to have Visual Studio Code (vs code) installed on your PC:
- VS Code: It is a source-code editor made by Microsoft with the Electron Framework, for Windows, Linux, and macOS.
- Clone the project
git clone https://github.com/UmuhireJessie/metal-classification.git
- Open the project with vs code
cd metal-classification
code .- Install the required dependencies
pip install -r requirements.txt- Run the project
streamlit run app.py- Use the link printed in the terminal to visualise the app. (Usually
http://127.0.0.1:5000/)
- defect_detection_model.h5: The main defect detection model trained on top parts of submersible pump impeller images.
- image_validation_model.h5: A model used to validate if the uploaded image is relevant to the defect detection task.
- scaler.pkl: The scaler used for standardizing features during inference.
- kmeans_model.pkl: The KMeans clustering model for checking the relevance of the uploaded image.
- train_features.pkl: Features extracted from the training set for clustering.
- The app is designed to work specifically with images of the top part of a submersible pump impeller.
- Images that do not belong to this category will be rejected.
- Jessie Umuhire Umutesi
- Adrine Uwera
- Evelyne Umubyeyi
- Gabin Ishimwe