Artificial intelligence could soon become the latest tool in the battle against redbanded stink bugs and other soybean pests.

A pair of projects led by Ivan Grijalva, an LSU AgCenter assistant professor in precision pest management, in collaboration with Jeff Davis and Kevin Hoffseth, focuses on developing decision-support tools. These tools are designed to help recognize insect damage to soybean seed and also to detect and number insects for more efficient field sampling.

Drones, smartphone cameras and other technologies have the capacity to capture high-resolution images, Grijalva said; however, artificial intelligence — particularly machine learning models — must still be taught to recognize insects and crop damage.

“We have the technology and the sensors capable of detecting small insects,” Grijalva said. “We need to develop the model so it can be deployed to a decision-support tool. For example, a model trained with drone imagery will be able to process the imagery and tell you that in a specific area there are 20 stink bugs, and then you can decide if you make an insecticide application or not.”

Both artificial intelligence projects are in their first year of development.

One of the first goals for the insect detection project is to create a model that can potentially monitor the top of the crop canopy with drones and under the canopy with smartphone cameras until redbanded stink bugs reach target populations.

Using an artificial intelligence approach known as machine learning, computer systems can then recognize the shapes and pixel features of these insects and how they appear in the field. This process helps the model refine its decisions about determining which shapes and features are stink bugs and which are not.

Grijalva plans to create a prototype application that soybean producers can use to upload images and detect stink bug presence.

“You gather all this information, then the machine learning model makes an insect detection,” Grijalva said. “You apply a lot of image analysis, preprocessing, train the model and then you test it in different images that the model didn’t see before.”

One goal, Grijalva said, is to develop a smartphone application for producers to use while scouting fields instead of manual insect counts. Eventually, Grijalva envisions creating a robotic vehicle that can crawl through fields and scout for insect populations and make management decisions.

“The farmer can invest their time in other activities instead of sampling,” he said.

Grijalva is working on a second artificial intelligence project with AgCenter engineer Kevin Hoffseth to develop a machine learning model that will assist soybean inspectors in assessing insect damage on soybean seeds. Hoffseth said that Grijalva’s work complements the objectives of Hoffseth’s lab.

In 2020, Hoffseth began developing software and hardware to assist soybean inspectors with the Federal Grain Inspection Service who grade soybeans and other grains postharvest. Hoffseth does not think computers can replace inspectors, but they can improve the process, he said.

“If you can make their life easier by automating a lot of the process, but leaving the grading decision up to the person that’s been trained, then I think that’s a really good goal,” Hoffseth said. “That’s what I’m trying to work towards.”

Grading grains is difficult work, Hoffseth and Grijalva said, and inspectors often consult photographs while performing their work.

“We want to apply these approaches because most of the visual inspections of soybean seeds are performed by humans, and that takes a lot of time,” Grijalva said. “It’s very time consuming and very subjective. When you apply machine learning, you can automate this process, and then you can be more efficient in sampling. Also, it will be more accurate and faster.”

The proposed machine learning model would be used to identify and categorize insect-damaged seeds. The first step in the process will involve teaching the model to recognize patterns of insect damage. Eventually, the research team would create an internet-based pilot tool that inspectors could access while working in the field.

Images, such as thes annotated photos of soybean seed, are used to train insect damage detection models. Photo by Saurav Upadhyaya and Peyton A. Falterman

This image of stink bugs is used to help train artificial intelligence models detect insects. Photo by Joshua Broussard