ScrapScan

• 

  Image Identifier interface

• 

  classification report of machine learning model running on a test dataset

• 

  List of proper disposable methods based on material classification

Inspiration

We really wanted to create an app that could promote sustainability in an interactive and easy way. After thinking about sustainability in our everyday lives, we fell upon a question that we surprisingly ask ourselves often: How should I dispose of this? While disposable methods for trash can be searched online, we were inspired to simplify this process for users using image recognition.

What it does

The main feature of the web application is to, given an image of some type of trash uploaded by the user, use a trained ML model to provide a prediction of the trash's material and display its corresponding proper disposal method. The app also provides descriptions of each material category and its disposal method for users who are more confident in their item's material.

How we built it

We built this app using Javascript with a ReactJS frontend. We used machine learning models in python with TensorFlow and Keras. We used Flask to incorporate our Javascript and Python.

Challenges we ran into

We had difficulty setting up a ML algorithm to best accurately represent the data. We also had trouble exporting the trained model from Google Colab to the Python script while retaining the prediction algorithm. We found it challenging to get ReactJS and Flask to work together.

Accomplishments that we're proud of

We are proud of the implementation of some of the frontend-heavy features, considering the limited time we had. We are proud to have incorporated machine learning into out project to create something pretty interactive.

What we learned

We learned that time management and proper division of tasks are very important in such short time. We learned more about image recognition ML and the connections between the many components of the project. We learned more about functional component-based rendering using ReactJS.

What's next for ScrapScan

We hope to go back improve the connection between the initial training of the machine learning model to the image upload function. We would also like to improve the experience for mobile users.

Built With

• cors
• flask
• google.colab
• javascript
• keras
• node.js
• numpy
• pandas
• python
• react
• sklearn
• tensorflow