Trash-optimizer

When someone wants to find the right place to recycle their trash they face two major obstacles: classifying the item to know the deposit location and find the closest place to their home to do so. To resolve those pains we launched a projet called trash-optimizer. Based on an user's picture the app would classify an item within 19 classes of trashes based of the classification of Nantes Metropole ( https://data.nantesmetropole.fr/explore/dataset/244400404_jeter-dechet-nantes-metropole/table/ ). The app generates the shortest road to the user to throw those different items based on available locations to do so. To create the app the team started from an existing image recognition model called EfficientNetB0 which could classify 1000 different objects. Out of our 19 trash classes only 5 were available which pushed the team to fine-tune and re-train the model based on 14 new labelled datasets. The updated model is generating 90% accuracy in classifying items correctly. To find the places to deposit items the team collected data from different sources (CSV, Open Data API, Webscrapping & Geoservice API). This data was reprocessed, consolidated, transformed and pushed to Big Query to generate one single source of truth. To find the shortest road to the user the app is calling the database in batch and calculating the distance between each points and optimizing the final road and displaying it to the user.