Happy Earth Day! Today millions of people around the world will be discussing emerging green technologies and environmental innovation–we wanted to get in on the conversation too! We’ve written in the past about computer vision based wildlife conservation efforts, but for this post we want to pick up a topic that’s a little less fluffy and arguably more smelly: computer vision for waste management. 

The Trash Problem

Waste management might not be the first thing you think about when someone mentions computer vision, but it’s actually an excellent use case for the technology. Identifying and sorting trash is, at its most basic, an exercise in object detection. Object classification also comes into play when, for example, waste-managing computers recognize a piece of trash as not only a soda bottle, but a Coca-Cola bottle or a bottle made of polyethylene terephthalate (PET) plastic.

Computer vision labels distinguish different kinds of trash on a conveyor belt.

Annually, human beings produce 2.01 billion tons of waste (that’s over 4 trillion pounds) and according to the World Bank, in high-income countries like the United States almost 51% of all waste is recyclable. In a 2018 study, the Environmental Protection Agency (EPA) reported that of the 292.4 million tons of municipal solid waste produced in the US, only 69 million tons was recycled, coming in around 24% of the total waste. This means roughly 76 million tons of recyclable materials were not correctly processed and most likely ended up in a landfill.

What is Being Done? 

Currently, only eight states have banned single-use plastic bags and two states, Oregon and Maine, have adopted laws requiring producers of packaging materials to share in the cost of recycling programs, which is a common practice in Europe. In March of 2022, Washington-based coffee giant Starbucks pledged to ditch single-use cups by 2025.  

In November 2021, President Biden signed the Infrastructure Investment and Jobs Act into law. The bill, passed with bi-partisan support, allocates $350 million dollars to the Recycle Act which calls for the EPA to establish a program for allocating funding to improve recycling programs at the local and state levels across the country. While these are steps in the right direction, the sheer size of the problem will require building and maintaining a recycling infrastructure and leaning on technology and innovation to ensure it constantly evolves and grows to proactively, strategically address waste. 

The current waste processing system in the US is a combination of locally regulated collection streams (for example, different colored trash bins designating different types of waste) and material recovery centers where waste is sorted by hand and weight on conveyor belts. For this option to prove viable, the existing waste processing facilities need to become much more accurate and efficient at sorting out recyclable materials. Government agencies and companies need to think beyond what’s been manually impossible and embrace what’s now technically possible with vision AI.

Computer Vision and Trash

Sorting trash is a dirty and hazardous job, one which would be perfectly suited for an AI  intervention. But how do you teach a machine to understand the difference between waste that is recyclable and waste that is not? The answer lies in computer vision models.

Computer vision models consist of algorithms that teach computers to “understand” the images that they “see.” In order to do this, the algorithms need visual datasets that act almost like learning flashcards, training the algorithm to recognize an object through repetition. 

Creating a waste dataset requires manually labeling images, often totalling in the thousands. In this case, images of trash could be labeled for everything from material type, to object shape, to product brand. Plainsight’s vision AI platform offers a no-code, end-to-end model building environment for quickly labeling and automatically training datasets with images or video of waste or visual data of any kind.

Examples of WasteNet’s Waste Database and Sub-Databases. Image courtesy of Recycleye’s Website.

Material Recovery centers are not the only area where waste detection models can be useful. Even The Great Ocean Cleanup project is collecting visual data to combat waste. They are currently looking for volunteer citizen scientists to help add to their database by taking photos of local polluted waterways and uploading them for analysis and comparison. The application closes at the end of this month if you are interested in getting involved. 

Cleaning up with Vision AI  

As global waste is projected to grow to 6.8 trillion pounds produced annually or 3.4 billion tons by 2050; it’s awesome to know we can count on vision AI to play a role in the fight to clean up our planet. No-code platforms like Plainsight On-Demand make collecting data, creating datasets, and building custom vision AI models fast, easy, and affordable. With free, unlimited labels and dataset versioning tools, Plainsight makes creating robust and accurate datasets a no-brainer. If you’ve got specific questions about how Plainsight vision AI can transform your computer vision projects, reach out to schedule a free demo today 

 View All Blogs