We noticed that a lot of people, including us, are often confused as to what waste is recyclable and what isn't. Millions of dollars are spent every year on sorting waste, and a lot of it can be made easier and cheaper just by increasing awareness about classifying waste.
What it does
We decided to make this easier and interesting for you using our app RecycleMe! Our app allows you to take pictures of objects in your environment and classify them as 'Recyclable', 'Landfill', or 'Compost', so that you can sort them accordingly. Have a positive impact and change society one step at a time.
How we built it
We used Microsoft Azure's Cognitive Services to build a Custom Vision Trainer to classify the waste into the three respective classes. The dataset for this machine learning model was obtained using a Python script that took advantage of the Bing Search API to pull images of various common objects. We downloaded the top 150 searches returned by the API for each object specified in a text file, and fed them to the trainer. Our Android application, built using Android Studio, allows users to take a picture, which is sent using HTTP to the Azure model. A JSON file is returned with details of the predicted tag for the image.
Challenges we ran into
Although image collection was automated, we did not always get satisfactory data. We had to manually filter some of the images which was time-consuming, especially when we were given only 24 hours. The Android Studio application was the most challenging part of the hack. Apart from being slow and time-consuming, the high level of abstraction made the debugging process cumbersome.
Accomplishments that we're proud of
Our classifier works very well for common objects in our vicinity, such as plastic wrappers, foils, and bottles, which we would often have confusion in classifying.
What we learned
We learnt to use Azure and work with APIs, such as the Bing Search and Custom Vision. We also integrated a camera into our app for the first time. Having been our first Hackathon, we learned how to manage our time and work efficiently by splitting work among our members. Two of us had prior Android app development experience while the other two took on the task of building the vision model.
What's next for RecycleMe
We hope to expand and implement this idea all over the world. This application could be embedded into a system that would automate the sorting process at waste facilities. We can further increase the accuracy of the model, and add more classes, like e-waste, bio-hazards, and even sub-classes under recycling for paper, plastics, and metal. Further steps would include obtaining statistics to track how user awareness of waste classification changes with time. We would also provide more information about the object, its decomposition time, its carbon footprint statistics, etc. We feel that this idea can save public tax money which can be used to target other social causes.