DrainWatch
The Problem
More than 8 million tons of plastics sneak into our oceans every year. While approximately 20% of this is lost directly into the sea from boats and shoreline activities, the rest slips into our sewage, waterways, and eventually oceans under our noses. Runoff from precipitation sweeps away both litter and even plastic secured in outdoor bins and landfills. Microbeads hide undetected in many of our personal hygiene products, which we wash down the drain by turning on a faucet with our own hand. It is well-known that this plastic harms wildlife, and potentially humans by the extension of our seafood-inclusive diets. Sea turtles suffocating inside a plastic bag. Pelicans choking on plastic wrapping. Over 50 species of fish eating microplastic beads as food and dying from malnourishment when these morsels fail to provide essential nutrients, or passing the associated toxins up along the food chain. The effort to clean up this mess requires a superhuman effort as it is, yet every day more plastic infiltrates our aquatic ecosystem.
Our Proposal
DrainWatch is an effort to give our drains eyes - and a voice - to help us keep our plastic out of the ocean. At its core is an AI that will be trained to recognize plastic that is at risk of infiltrating the waterways. Storm drains could be equipped with a sensor which is able to detect the entry of objects carried in by the water, and which triggers a simple camera to take a picture. The image is uploaded to be processed by the AI, which identifies litter items by type. The system keeps track of the amount and types of plastic litters that enter the storm drainage, as well as their points of entry. Plastic type and point of entry data can be used to develop a strategy to mitigate infiltration. Alternatively, the sensor and camera system can be placed at the point where storm drainage is emptied into a body of water. If the image upload and analysis can be made to occur in less than 2 or 3 seconds, we could potentially even equip drains with grates that engage when plastic is detected to prevent it from falling in. DrainWatch could also identify microplastic content of a personal care product on the store shelf. A conscientious consumer, armed with a picture-taking mobile app, could share the image with the DrainWatch bot. The AI would analyze the product packaging to match the product on the shelf with information on its content, most notably on whether the product contains microplastic. This would enable the consumer to identify microplastic-free alternatives.
We Assume that...
1. An established, widely available machine learning framework (e.g. TensorFlow) can identify plastic items in an image taken of what is entering a storm drain at a given time.
2. A sensor exists that can identify when an object of significant size passes through the storm drain entry.
3. A camera can be outfitted that is durable and cheap enough to be installed at the mouth of a storm drain, as well as able to take pictures of high enough detail to be analyzed by an AI.
4. Information visible on the packaging of a personal care product is enough to distinguish it from other products.
Constraints to Overcome
As with any machine learning application, a major component will be collecting and effectively conditioning image data to train the algorithm. Training the system to identify garbage items that will be largely submersed in water and moving at high speeds will be a challenge. The microplastic identification system would require developing data on the microplastic content of a wide variety of personal care products, to which the algorithm would match photographed packaging on store shelves. Outfitting or developing a camera that is sturdy and cheap enough to install in a stormwater drain - as well as able to upload images reliably to the cloud - may contribute to another component of challenge.
Current Work
- Capture and characterize images of 1,000 pieces of plastic flowing into storm drain. - Identify one quality of storm drain location that draws high rates of plastic waste. - Identify 3 forms of plastic litter that most frequently are washed into storm drains. - Provide microplastic content information for 300 products that consumers find on shelves.