Our team is mostly comprised, of Robotics enthusiasts in Africa. We were exploring how we can leverage Robotics in Agriculture to solve problems in agriculture and narrowed down on solving problems specific to corn (maize) as it is one of the staple crops grown in large quantities all over Africa. For this competition we created the team, FLYGRMAIZE: Flying and Ground Robotic is a Multi-Agent system that autonomously inspects maize fields to collect data for disease research. Project FLYGRMAIZE: Flying & ground robots, powered by OpenCV in a Multi-Agent system, autonomously inspect maize fields for disease research. Our team found the Namibia University of Science and Technology Maize Dataset which can be accessed here Maize Dataset Description URL Hyperlink.
Our team detected two main limitations in this research project that we can improve on using Robotics: Data collection methodologies and a limited amount of labels. Data Collection methodology, as described in the pdf document, the collectors of images in the dataset were literally walking around the corn (maize) field and taking pictures with their phones. This was of course very time-consuming. This is where robotics can be leveraged as tools for data collection. In the air, a drone that would be equipped with different kinds of cameras and equipped with on-board processing algorithms such as Normalized difference vegetation index (NVDI) can be used on the collected images to report plant health data back to the farmers. The wonderful OpenCV library has many examples of how to compute the aforementioned index and plenty of others. On the ground, a ground robot would be navigating through the field rows to get up close pictures of the stalks and leaves. The above limitation is apparent because they were only able to get a total of 9034 images. These aren't enough images for a robust machine-learning/ computer vision project as typical datasets contain at least hundreds of thousands of images if not millions. The second problem we identified was that the research was only able to get a limited amount of labels. So they were only able to get two kinds of diseases (Healthy, Fall Army Worm, and Maize Streak Virus). Part of our team is an Agricultural Engineering student whose family has a farm. The farm is located in Kenya, South Rift Region, Kericho County, Londiani SubCounty. The main kind of Maize (Corn) species they grow is H6213/H614/H629. They are commonly affected by some diseases, and bacteria that can also add additional labels to the research:
Stalk rot: is a disease caused by a fungus infection in the stem. The disease can survive in the soil for some years.
Head Smut is characterized by large smut galls that replace ears or tassels. The galls are first covered by fragile, creamy white membranes that eventually rupture to release masses of dark brown spores.
Corn Smut: plant disease caused by the fungus Ustilago maydis, which attacks corn (maize).
Stalk Bores: They are serious insects that affect maize, they are very common and can best be controlled when they are still in their early life cycle.
Aphids: an insect, and a pest of maize and other crops. Among aphids that feed on maize, it is the most encountered and most economically damaging, they mostly spread during the warm season.
Bacterial Leaf Streaks include interveinal streaks that can be yellow to brown with irregular margins. Yellow discoloration may be more evident when backlit by the sun.
Our team will demonstrate the ability to autonomously collect a sufficient amount of data in a short amount of time using Robots and with the aim of eventually contributing to the dataset.
Comments