Waste Management in the United States and other countries globally has emerged as a major concern over the past few years. The rise in urban population and economic growth in the absence of an effective management mechanism has manifested in the current state of solid waste management in the US which is far from perfect. Given the present situation, the quantum of waste generated in highly populated cities is on the rise. The focus must be on devising effective mechanisms for waste treatment and disposal in urban centers and monitoring the use of banned plastic.

What it does

The proposed solution includes an automated recognition system that uses a Deep learning Algorithm in Artificial Intelligence to classify objects as biodegradable and non-biodegradable, where the system, once trained with a readily available & self-made dataset of more than 50000 images, can identify and classify objects in real-time with high accuracy. Based on this classification, the adjacent community will be told whether or not the discarded garbage from their area has been separated, as well as the existence of prohibited plastic in the classified waste.

How we built it

For developing the prescribed solution we have used Node.Js, Express & Python in the backend. Django with Tailwind CSS is used for developing the user interface. As external tools, OpenCV, Tensorflow, and Keras were used and the database will be managed using MongoDB.

Challenges we ran into

  • Data Collection - We created our own dataset by roaming around during the hackathon and also asked our friends to send garbage photos. The dataset approximately consists of 50,000 images

  • Data cleaning, preprocessing, and labeling was the most tedious job of this project.

  • We have a total of 29 labels across all 50,000 images.

  • Research, trained, and tested various AI models and finally decided to go ahead with Faster RCNN Optimizing and training Faster RCNN to obtain higher accuracy. [86.55% Accuracy Achieved]

Accomplishments that we're proud of

  • Successfully completed our idea and we are very proud in achieving 86.55% accuracy after building so many models.

What we learned

  • Object detection using Keras and Tensorflow
  • Object Labelling
  • Deploying AI model on GCP and connecting the server with the domain. (

What's next for GarbageVision

  • The research employs an augmentation learning approach, in which the model employed is returned on user input after a predetermined time period. This will gradually improve the model's accuracy.

  • We intend to improve the model even further by developing our classification method.

  • We are excited to work with the governing bodies to get this project up and running so that it may be used in all states in everyday life for the welfare of society.

  • We also intend to turn the idea into a full-time company and to scale it significantly.

Built With

Share this project: