Introduction

The consumption and creation of single-use plastics continue to be at an all-time high, with only 10% of plastics recycled – the other 90% continuing to pollute the planet. The current plastic recycling process is tedious, requiring the sorting of the 7 types of plastics before recycling, due to different methods and properties. This sorting process is currently done manually; however, automation would allow for efficient results with less energy. We set out to do this through camera detection and machine learning artificial intelligence. For an AI to produce accurate results, it must be trained on masses of data – datasets that are typically interpreted by humans. This requires a collective crowdsourcing method, for many people to get involved as we work. Hence, the importance of an accessible method of collecting data.

Software

For the software, we used Figma to mock-up the UI/UX, and Flutter/Dart as the programming language, known for its popularity in multi-platform applications. A cloud database is used to send the information to the yolo machine learning program. The web application is showcased on a desktop, but in theory, should be available on a mobile device as well.

Projects

User interface using Figma

In earlier planning stages of the application, Figma is used as a mock-up tool. Here, understanding UI/UX is crucial before creating the rest of the application.

After several sketches and general brainstorming, we settled on a simple design of a home page, an upload page, an annotation page, as well as an information page. The home page consists of general information on the purpose of the application, along with the importance of plastic recycling. It gives a quick explanation of how the datasets work and its relation to machine learning and plastic recycling. The uploading page allows users to upload their pictures of plastic, while the annotation page allows users to either choose uploaded pictures or take pictures live, before annotating and labeling with a bounding box. The information page offers further information on the types of plastic, issues with plastic recycling, and

In UI/UX planning, a color scheme was created of greens and blues, as well as fonts, buttons, and navigation bars. The web application aimed to be simple and easy to navigate – concise pages with all the crucial information needed. In the data collection aspects of uploading and annotation, the contents are bareboned – working well with Flutter, which develops applications in the form of building blocks.

Bounding Box Annotation Tool

The created machine learning program the dataset is meant to train for requires the image to be annotated and labelled through a bounding box. It takes into consideration the file name used, the center X and Y coordinates, the width and height of the box, as well as the label (of the plastic type), all in one text file. In doing so, the function Stack is used to overlap two widgets, one with the uploaded picture and another user-made box, identifying the object. Taking into account the window pixels and border radius, the coordinates, width, and height are found.

Conclusion, Application and Future Outlook

As said previously, the data collected in this web application will be applied to train a machine learning AI. This AI is put into practice in a prototype named SortBotz, identifying plastic types through a live camera feed, showing where the plastic must be dropped. For an AI to produce accurate results, it must be trained on masses of data – datasets that are typically interpreted by humans. Due to the number of datasets that are needed, data collection must be a collective effort. The accessibility of annotation and collection become key in crowdsourcing and working our way towards a more sustainable future.

This web application for data collection can be used in other contexts as well. The UI/UX, especially regarding the bounding box, can be applied universally for other data-collecting purposes – not only in plastic detection. As machine learning and AI continue to be on the rise, accurate data collection continues to be of equal importance.