DNN & Time-Series Prediction
App Entry Page
App Rank Page
Map based on 311 Data
Simple webapp: upload
Simple webapp: detection
What can every citizen of Montreal do to make our city more sustainable? What are the small things in life that, when everyone does them, make a big change to this planet? With our app we want to make climate change actionable. The treatment of personal household waste and recycling have a big impact on both CO2 emissions and well-being in public spaces. But how to make it fun? How to learn what your personal impact is? We want to increase awareness for the impact of proper waste treatment and make recycling fun and easy. And on the side, we also make Montreals streets and public spaces more enjoyable. Take our app for a walk, collect trash and get rewarded, and directly see the impact you're having on your environment, and connect with like-minded people.
The data collected through the individual use of the app, will also allow the city of Montreal to develop a better waste management and optimize the placement of waste and recycle bins. The city of Montreal is spending millions of dollars on waste management and the cost is increasing each year. In 2018, the complaints related to compost, recycling and trash was 2170 and in 2019, the number is already at 1380 complaints. This issue is not prevalent in only in Montreal however, many cities around the world are not effective in their waste management systems. This is due to lack of city resource, citizen education and lack of incentives to make a change on an individual level. Our goal is to save the money spent on waste management hope that it could be directed into planting more trees, increasing urban environment green initiatives and further education of the citizens.
What it does
We introduce to you YOCCO, an app that allows you to take a picture of your waste and identifies whether it should be placed in the trash, compost or recycle bin. The app uses data from climatedata.ca and city of Montreal 311 complaints data to determine the areas of the city that have the highest trash, compost, recycling complaints. The app incentivizes the user by having a ranking system which keeps track of CO2 emissions saved.
How we built it
To detect the objects, we used 2 different pre-trained image detection models: DenseNet-121 and MobileNet. DenseNet-121 was pre-trained on ImageNet and can distinguish 1000 classes of objects. MobileNet was furthermore fine-tuned to distinguish cardboard, glass, metal, paper, plastic, trash. The results of the classification connects to a database that lets the user know if the item should be recycled, composted, put into the general trash or needs special treatment (hazardous materials). We also used a Multi-Variate Time-Series algorithm (VAR) to predict the future complaints to the 311 City of Montreal hot-line. For the front-end, we used HTML and created a prototype of the app and its features.
Challenges we ran into
Although we were able to get quite good accuracy with object detection and time-series prediction, the integration of our app with the ML system and 311 data was where we faced the biggest hurdle. Regarding the data collection, there is a vast amount of data and gathering the most relative as well as minimize it to the scale of our project was a challenge. On the machine learning side, getting a high accuracy and suitable confidence was the biggest challenge. When there was more than one item in the picture, only one item was identified, since the algorithms lacks segmentation. Regarding the front end, the application we used to generate the HTML code for the application generated a very complex code that was not easy to integrate with the machine learning code. Hence, we had to create a simplified version of the application to be able to integrate it with the machine learning code.
Accomplishments that we're proud of
Our garbage material detection algorithm has an accuracy of ~90%! We worked very hard as a team, learning different skills to bring together the project right until the last minute. We were able to use our strength, individually, and share them with each other.
What we learned
We learned the importance of teamwork: working together is essential because people complete each other. We also learned methods of classification for unlabeled data by using object detection algorithms. Meanwhile, we found out how to search the web for useful data to feed the algorithm. In web development we learned how to design and develop a web app in a very short amount of time. We learned how to think creatively as a team, brainstorming and producing creative ideas. We gained some general knowledge while checking out the data used and were generally impressed with the complexity of relations and processes regarding CO2 emissions and waste management, and the vast amount of information on climate change available online.
What's next for YOCCO (You Only Climate Change Once)
In the future, we have many plans for the YOCCO app! We plan to create an "effective walking path" that lets the users know which path is the best path to take to their desired destination that would provide the highest yield of garbage pick up. This would be integrated with the 311 data and time-series analysis. Furthermore, the ML trash detection would benefit from an automatic segmentation model that will allow us to detect and separate multiple objects in one picture. In addition, we will include an 'expert mode', where trusted users can correct the ML garbage detection if it's correct. This is how we will augment our initial data set and improve the AI accuracy further.
Additionally, we hope to deploy the app on Android and iOS to allow users to create a community and use the gamification method of the platform to decrease CO2 emissions individually while working together with their local communities.