AnthroKrishi uses AI systems to analyse satellite imagery to map the boundaries of individual farms and monitor crops, yields, and farming activities. The capability provides comprehensive data to enable smarter farming, from banks issuing loans based on expected yields to governments optimising subsidy programmes.
“We’ve done pilots with the state of Telangana where they have given us feedback of where there are some mistakes, (here’s where) there is some over segmentation or under segmentation that’s been done, but for the most part, it validated that the models were fairly accurate,” Manish Gupta, Head, Google Research APAC said while addressing a virtual session called Google AI Now.
Gupta added that the search giant has also been doing a pilot in Maharashtra with IIT Bombay wherein it is working with the Department of Land Records and trying to see if usage-based patterns give reasonably good insights about the ownership of farms.
He said that there is a lot of ‘ground truth data’ in India but highlighted that it tends to be ‘very noisy,’ which is one of the foremost challenges when it comes to this project. For instance, some state governments do an annual survey but Gupta said that when the team looked at that data, there tended to be a lot of errors.
Further, there was the issue of not just building the first ever model that provides this landscape understanding and farm segmentation for the entire country but also going ahead and validating it which is where he said the pilots with the governments came in handy.
Discover the stories of your interest
Gupta said that a landscape understanding, which is able to identify farm boundaries and also analyze things like sources of water for each farm and so on was a fundamental building block for a digital agri-stack which is what he said was the vision to be created in countries like India and beyond.”In India, for instance, we have had this very thriving ecosystem for mobile payments,” he explained. “The foundational layer of that entire mobile stack comes from the Aadhaar ID. So in an analogous manner, through this work, we are trying to assign a unique ID to each farm and allow the various solutions based on agriculture to be built on that foundation.”
Gupta added that once each farm has been identified, the next thing to be done is to identify what are the crops which are being grown as well as monitor various farming related activities like sowing of seeds, harvesting and so on.
“For that, we have built that model called AMED – Agricultural Monitoring and Event Detection – and through this we are able to identify, for instance, the crops which are being grown,” he said. “This can unlock a lot of use cases that can benefit farmers. For instance, if a farmer wants to apply for a loan, today it’s a very cumbersome process. But now with a simple click, once the farmer shares their land ownership information, the bank can have a pretty comprehensive view of what the farmer has been growing on their farm as well as estimate their likely income at the time of harvesting and therefore provide that loan or that crop insurance to farmers.”
He went on to say that governments too can start to make decisions on things like subsidies and so on based on what farmers are actually doing on their farm.
