The combination of Google Cloud, NVIDIA GPUs, and TensorFlow machine learning has helped agriculture innovator Taranis reduce crop loss for farmers all around the world.
Up to 40 percent of the world’s crops are lost every year to issues like pests, weeds, diseases, and nutrient deficiencies. To feed a rapidly growing population in an increasingly unstable climate, farmers need to find ways of preventing these issues to maximize yields and lower production costs.
But how do you spot these issues early and deal with them proactively when farms are spread over hundreds of acres?
Agriculture technology provider Taranis gives farmers unprecedented insight into the health of their crops. It monitors more than 20 million acres via its platform, which processes multitudes of high-resolution photographs from drones, planes and satellites.
Processing this many files in such a high level of detail naturally takes huge amounts of computing power and, because Taranis is uploading data from remote sites all over the world, reliable connectivity.
“We collect vast amounts of data from all over the world, in places with not much connectivity including Russia, Eastern Europe, US midwest and South America,” says Eli Bukchin, Taranis Co-Founder and CTO. “Developing methods to upload data in those conditions is a challenge.”
A fast, powerful infrastructure – accessible from anywhere
Each drone flight collects around 10,000 photographs, and each image can be up to 20MB. Taranis needed a fast, reliable system for uploading and processing these images – which didn’t require millions in infrastructure investment.
So Taranis migrated to Google Cloud, using NVIDIA Tesla V100 GPUs to provide the massive amounts of power it needs to process its images through Compute Engine.
Agriculture is seasonal, so there are always peaks and troughs in demand during the year – alongside daily peaks in activity that also need to be managed. As its NVIDIA GPUs are hosted in Google Cloud, the system can scale easily from 1,000 to 4,000 V100s (and back down again) when new images are uploaded.
This processing power also enables Taranis to train its machine learning platform, TensorFlow. With a team of 30 agronomists tagging images with points of interest to feed into the models, the AI can process millions of images to improve its automated identification function.
Without the cloud-based processing power, Taranis wouldn’t be able to manage this process as effectively. “Each photo might have up to a thousand items of interest, such as insect damage or leaf discoloration,” says Eli. “So the data volumes are really significant.”
Taranis has built a full ecosystem using Google Cloud and NVIDIA that includes multiple optimized services As Eli explains: “Information is deduced from the images, which is then uploaded to a Cloud SQL database. We also have an image processing pipeline on Kubernetes Engine for our satellite images, in addition to using Cloud Functions and Cloud Pub/Sub.”
Intelligent agriculture with Google Cloud and NVIDIA
The power and connectivity of the company’s new environment has enabled Taranis to work more effectively than ever before. Now, images from the drones, planes and satellites can be processed far faster – and in greater volumes.
“Before, uploading was three or four times slower and could take up to a day rather than a few hours. The difference is really substantial,” says Eli. “In total, we have processed around 100 million distinct features in around 700,000 images.”
Not only is the solution more effective, it’s cheaper, too. The average cost per image is now 10 times lower, thanks to the combination of cloud-based services and scalable GPUs reducing operational costs.
Together, this is all enhancing the company’s ability to provide useful, proactive insights and advice to the farmers of the world – so they can produce higher yields using exactly the same fields as before. As Eli explains, these technologies “enable us to identify problems before they affect the crops, so farmers can intervene earlier in a more targeted way.”
The Taranis team is now looking for more ways to integrate data analytics into its cloud environment: “We’re currently considering how Cloud Bigtable, BigQuery, and Cloud Dataflow might fit with our business needs,” says Eli. “We’re building the Google Cloud infrastructure to gain BI insights into our own system.”