AI Case Study
Researchers at Stanford University aim to detect illegal polluting from farms using a convolutional neural network
Researchers at Stanford University have applied a deep convolutional neural network to high-resolution satellite images of ‘concentrated animal feeding operations’ (CAFOs). Until now, practices to identify and locate these facilities have included manual work using maps and conducting field visits. The researchers believe that their method could be a cost-effective and accurate approach to identifying CAFO locations and ensuring their compliance to environmental regulations, and in turn to detecting pollution into waterways. When the system scanned satellite images of North Carolina, the algorithm was able to detect 15% more poultry CAFOs in comparison to the existing data in just two days, compared to the six weeks time period required manually.
Industry
Basic Materials
Agriculture
Project Overview
"Artificial intelligence is using satellite images to observe farms. The technique is being tested in the US to detect farms that may be illegally polluting the waterways, and has been trialled across Europe to monitor and inspect farmland.
There’s no federal government system that tracks the number and size of these farms, so environmental groups laboriously compile their own by manually scanning satellite images.
To locate farms more efficiently, Daniel Ho and Cassandra Handan-Nader at Stanford University, California trained a neural network to scan publicly available satellite images for CAFOs. It learned to identify certain hallmarks, such as multiple rectangular barns or outdoor manure pits.
Locating these farms is the first step for regulators to identify polluters. Ho believes that computer algorithms will be able to detect actual pollution into waterways in the future.
Similar techniques are being used on agricultural land across Europe. Algorithms are monitoring the health of vineyards in southern Italy, and observing farmers in Lithuania and Estonia who receive government subsidies for keeping their land in good condition.
The system used in Estonia detects whether farmers are mowing their fields as required, reducing the need for physical visits from inspectors. It is estimated to save €500,000 every year in manual inspection costs and false payments made to non-compliant farmers."
Reported Results
"Scanning satellite imagery of North Carolina for poultry CAFOs, the algorithm found 15 per cent more farms than were previously known – an increase of 589 farms. It took the AI two days to complete the task, which would have taken a person six weeks.
The previous manual count was based on earlier imagery, so the facilities could have been constructed in the meantime, contributing to the increase. But it is likely that individuals missed farms while scanning through troves of images, says Ho."
Technology
Function
Legal And Compliance
Compliance
Background
"In the US, facilities known as concentrated animal feeding operations (CAFOs) comprise around 40 per cent of the country’s livestock. These intensive farms often contain as many as 2500 pigs or 125,000 chickens per facility and generate around 335 million tonnes of waste per year.
Manure forms a large proportion of this waste, which often makes its way into waterways untreated. Under the US’s Clean Water Act, anyone who wants to dump waste into a waterway requires a federal permit, but the Environmental Protection Agency estimates that nearly 60 per cent of CAFOs don’t have one."
Benefits
Data
High-resolution satellite images