Crop loss is a huge problem for sustainability and food security. I use AI and computer vision to help fix it.

• Valeria Kogan is CEO of Fermata, an agricultural software company.

• Her team develops AI tools to reduce crop losses that cause food waste and greenhouse gas emissions.

• This article is part of the “Build IT” series, which explores digital technology trends that are changing the face of industries.

This “as-told” essay is based on a conversation with Valeria Kogan, founder and CEO of Fermata, a company that develops software to monitor plants and crops. The following text has been edited for length and clarity.

When we think of food waste, we usually think of the scale of a never-ending dinner. In reality, agricultural waste is a much bigger problem. It’s estimated that between 20% and 40% of crops are lost to pests and disease, affecting our food systems and the planet; greenhouse gas emissions from food that’s never eaten account for 6% of total emissions.

I didn’t know anything about this about five years ago. I was working in bioinformatics, developing an AI-based platform that helps oncologists diagnose cancer, and I knew so little about agriculture that I couldn’t even keep a houseplant alive.

But someone who ran a commercial greenhouse got in touch to discuss how AI could solve these yield loss problems. I quickly saw an opportunity to translate my work on human health into plant health, and Fermata was born.

Using AI and computer vision where humans fail

Today, many farmers try to detect pests and diseases early by sending workers into the fields every day to check every leaf on every plant for early signs of problems. But people get tired, distracted, and ignore the small changes that suggest a problem.

It reminded me of doctors who spend all day looking at X-rays and miss early signs of cancer. We were using AI image analysis to solve this problem and I felt the same solution could apply here.

Initially, we thought about building a robot that would move through the fields, but we realized that we did not have enough knowledge to achieve this – and that there was a simpler solution.

We developed the Croptimus platform, which uses standard security cameras in greenhouses and fields to take pictures of every plant multiple times a day.

Our AI processes these images to detect any anomalies. If they do, it notifies farmers through our app and suggests what it thinks is the problem. Using data science, we can also provide an overview of what’s happening across the entire facility, whether problems are increasing or decreasing, and whether treatments are working.

With this technology, farmers need less work and can identify problems faster than with humans, which means they can use fewer pesticides. This saves a lot of money and helps produce more sustainable and healthy food.

Performing manual actions while creating a dataset

One of the biggest challenges we had to overcome was establishing ground truth to build a reliable dataset to train our models. Every agronomist or crop scientist has their own opinions and also makes mistakes. We had to not only compile a decent dataset from scratch, but also build machine learning models that could adapt to the errors.

We did a few things to help. First, we built a research lab where we grow plants, infect them with different things, and videotape them. We also hired an in-house team of agronomists to help us label those images.

We also made our product public before it was automated, performing identifications manually and encouraging farmers to provide feedback in the app. This helps us better understand the problem and provides us with a more robust data set. Even as we’ve moved to relying on AI for identification, this feedback loop helps us continually improve our models.

Building relationships to find future technology opportunities

It’s very important for any technologist trying to solve a problem in a new industry—especially in a more conservative field like agriculture—to remain humble and build real relationships with the people they’re trying to help. If I came in as an outsider to the tech industry and told these farmers who’d been doing this for decades that I could teach them how to operate and be more efficient, it wouldn’t work.

Instead, I work to build relationships and trust within the industry. I approach this from the perspective of wanting to learn from my clients and understand how my technical knowledge can help them.

Ultimately, it helped me see even more potential in what we do. I learned that there is so much visual data in agriculture, from understanding whether bees are pollinating to seeing how workers are treating crops. Fermat’s vision is to build a new layer of visual data in the agriculture industry that helps all stakeholders—from farmers to fertilizer sellers to pesticide companies—be more efficient.

Read the original article on Business Insider