BEATRIZ ZANFORLIN

Product | 2020 | Gaivota

Nematodes are small worms that live in the soil and attack the roots of agricultural crops.

A microscopic creature, a macroscopic problem

Nematodes attach to soybean roots and prevent them from absorbing nutrients from the soil. According to the USDA, the soybean cyst nematode is the most serious pest of soybean throughout the world (USDA, 2017). In Brazil, they are responsible for losses of R$35 billion (USD6 billion).

They spread through soil carried on boots, tractors, and also move slowly on their own. In 20 years, they expanded from 0 to 25% of the most productive grain-producing state in Brazil (Embrapa, 2009). In affected fields, they have been shown to cause yield losses of up to 30% (Schmitt & Baker, 1981).

A huge problem in food production.

Nematodes come in many types…

There are many nematode species. Some are harmful, while others are not. Each type requires a different management approach.

Can you tell if these two are of the same species?

What is usually done about it?

First, you need to identify the type of nematode affecting your farm. To do this:

• you send soil samples to a lab, following a specific sampling procedure
• the lab processes the soil and extracts a liquid solution containing nematodes
• this solution is then placed on a slide, where a highly skilled specialist – trained for years – manually counts each nematode, identifying them by type.

The problem: is slow and imprecise. There are very few specialists capable of doing this correctly. Brazil has 5M farms, in 5k cities, and less than 200 specialists. Chances are, there isn’t one near your farm. Get the wrong diagnosis, mistreat it, they will spread further.

We sent trial samples to different labs and found no consistency.

Each color represents a different specialist analyzing the sample. The numbers indicate the nematode counts in a small portion of the sample.

Inspiration and innovation

Telemedicine.

In the 1990s, telemedicine emerged. Skilled doctors weren’t always available to analyze medical images, but could now make a remote diagnosis**.** What if… we applied the same idea to nematology? What if… new technology at the time enabled automatic image recognition?

Process innovation, supported by technology

Our solution began with a detailed task analysis of the process, identifying areas that could be unbundled. Unbundling is not always straightforward, it requires shifts in the divided process and careful design of interfaces between the now disconnected steps.

Meso-level task analysis of the usual process

The new process still required a lab analyst to prepare the sample, but it no longer needed a specialist on-site for counting. Using their own cellphone and a purpose-built app, analysts captured images, which were then uploaded to the cloud. The system intelligently managed and redirected the images to nematologists within the network for analysis.

Mobile Application: lab assistant guided by the app to collect images.

Intelligent Distribution: to remote nematologists with predefined redundancy.

The physical world

Changes in process required changes in hardware.

We built 3D-printed cellphone mounts that attach to microscopes. Cheap and effective.

We also designed a new slide to guide analysts as they scan the visual field. With high magnification, it’s easy to lose track of the area being analyzed and we had to avoid double counting. We developed slides with system-detected markings to aid the user.

Final slide design

System detected marks

Machine Learning

The next step was integrating machine learning models to visually identify and support nematode counting. We used collected images to train and test the model. One of the challenges was making the system function without relying on an internet connection, with a lightweight model running locally – back in 2020. The model was efficient in counting nematodes and identifying certain species.

This project was an example of solving a problem by rethinking the distribution of tasks in the process, identifying opportunities to unbundle activities, redesign workflows, and integrate AI. We first optimized the human processes, with remote diagnosis, and system guidance, before strategically applying ML to enhance identification and counting. Nematode analysis could become more widely available and much faster.