Trash Bot

Sample Screenshots

Inspiration

There is growing environmental awareness across the globe in today's world. Our team also wants to contribute to the saving of our earth. One of the most important environmentally friendly practices is recycling. But a common problem of recycling is that a lot of people (including us) do not know how to recycle their trash properly. We want to share some information about recommended ways of recycling and hence encourage people to start recycling in more meaningful ways.

What it does

Classify trash according to images or text supplied by users
If it can be recycled, recommend nearby recycling locations for the specific trash type in NUS if the user agrees to share their current location. If not, all available recycling points will be shown.
If the item cannot be recycled, share some disposal tips for that item

How we built it

python-telegram-bot: a Python wrapper for Telegram Bot API
Tensorflow: image recognition and trash classification

Challenges we ran into

Our team had zero knowledge about how Telegram Bot works, so we had to google all the way to implement the features we need. For example, how to handle image input and store it in specified locations.

One specific challenge we encountered while developing the bot is to make the user information persistent throughout. After some research, we found out we must make use of ConversationHandle from the Telegram API to create the chat-styled interaction between users and our bot.

Another challenge is about the accuracy of our model. Due to time constraints, we did not have enough time to train our model sufficiently. Only one member of our team have experience with machine learning and the use of Tensorflow so it's a hard time for the rest of the team to catch up but we all learnt a lot from her.

Accomplishments that we're proud of

It's our first hackathon and we managed to finish a working prototype within one day! Starting from scratch! The accuracy in validation set can reach 55% for 17 labels.

What we learned

Image recognition with tensorflow
How to build a telegram bot using python-telegram-bot
Debugging skills

What's next for Trash Bot

We plan to expand the scope of the recycling locations to the whole island. So far, our trash bot returns only recycle locations in NUS. Next, we will collect more data about recycling locations across Singapore, so that every resident in Singapore can easily know where to recycle their trash.

Moreover, the accuracy of the image recognition model can be further improved and the number of types of trash can be increased. Besides, our bot so far only supports jpg format images. We can work on it to accept other photo formats. All these can make our bot more convenient for the user.