SustainabLA

Inspiration

The United States discards nearly 40 million tons - 80 billion pounds of food every year. That’s estimated to be 30-40 percent of the entire US food supply ( 219 pounds of waste per person ). Imagine every person in America throwing more than 650 average-sized apples right into the garbage. Apart from causing major losses to wholesale food producers, this impacts the environment heavily by increasing landfill sizes, transportation logistics, and crop cycle requirements.

Food spoilage is one of the biggest reasons people throw out food. What if there was a way we could optimize food production and make it more sustainable?

What it does

Our application aims to solve food sustainability issues in LA. Often, fresh produce is discarded due to improper management, and here is where we come in.

Grab your phone and take a quick picture of the food product. The app analyses the image and categorizes it accordingly into four buckets: Fresh, Ripe, Spoilt, and Rotten.

Rotten produce usually ends up in a landfill due to improper waste management. But now, our application will suggest the nearest agricultural unit based on your location. This way, even rotten fruits and vegetables can be used as manure and become useful biodegradable waste.

If the product is in the spoiled stage, based on your current location the app suggests a list of nearest food manufacturers and production units that require it. For example, spoilt bananas would not have a huge retail customer base, but can be sent to a smoothie production facility to be utilized appropriately.

At the end of the day, if the product is Ripe, then it is unlikely that any customers will purchase it the next day (after 12 hours or so). This time is crucial as it is on the verge of becoming spoilt. So now, based on your current location, the application suggests a list of the nearest food camps, shelters, and restaurants that need the food immediately and can make better use of it. Donating your produce to meal camps even earns you goodie points!

What would previously end up in a landfill directly will now be routed appropriately to ensure maximum utilization and zero wastage. Apart from sustainability, our app also adds monetary value by increasing the savings of wholesale food producers and reducing their potential loss.

How we built it

Machine Learning :

Cleaned and preprocessed ~5k images of stale and fresh produce images
Built a classifier model using transfer learning from InceptionV3 architecture
Trained to predict if the food image is rotten, spoiled ripe, or fresh Frontend:
A React progressive web app that streams images to the backend and receives the prediction, location, and statistical data
Uses OpenStreet maps to display nearby options
Uses custom charts to display user summary report Backend:
REST API to upload images as multipart/form-data.
DB to store the locations of the nearest calculated targets to redirect the produce.
Analytics to track the distribution of redirection and vendor goodie score.

Challenges we ran into

Scaling the application
Rendering maps with real time proximal data
Training a CNN model with limited resources

Accomplishments that we're proud of

Brainstorming and coming up with a novel idea. Building an end to end solution for the same in 24 hours.
Devising a high accuracy model for image classification
Smooth integration of frontend and backend.

What we learned

Building an end to end application in 24 hours.
Data Analytics to highlight major trends in data
Transfer learning and extending an existing model to more data

What's next for SustainabLA

Installing cameras in various supermarkets along the lines of Amazon Go to supplement the execution of our idea.
Extrapolating the existing system for other produce available in supermarkets.