They monitor the whale shark in BC with drones and artificial intelligence models

Researchers from the Ensenada Scientific Research and Higher Education Center (Cicese) tested the combined use of drones and artificial intelligence (AI) models to automatically identify and track whale sharks (Rhincodon typus) in Bahía de los Ángeles, Baja California. These tools promise to reduce the physical effort of monitoring, reduce risks for observers and reduce costs.

“We wanted a monitoring alternative that required less physical effort and that could be used by the community, which is the one that carries out this work in Bahía de los Ángeles, in addition to reducing expenses,” he explained in an interview with The Day Paola Judith Delgado García, specialist in marine megafauna and author of the research.

The whale sharks that arrive each year in Bahía de los Ángeles, in the Gulf of Baja California, are young populations that are characterized by their enormous size. “Some reach 6 to 9 meters in this region,” and feature a distinctive pattern of white spots and light lines on a dark gray background.

Because in Mexico they are classified as a threatened species, according to NOM-059-Semarnat, correct and constant monitoring is crucial for their conservation, since it allows us to know the size of populations, if boat traffic affects their presence and if there are critical areas that require better management measures.

Compared to monitoring carried out from boats, these tools showed their advantages. There is a saving in gasoline, the logistics of the boats are dispensed with and the obtaining of shark sizes is strengthened.

“We wanted an alternative that would allow us to carry out monitoring from the coast, because the people who carry it out are volunteers, so many times they have to choose between going to monitor the whale shark and not generating profits or giving a tour tourism,” explained Delgado García.

In order to train the AI ​​models to identify the images collected by the drones, a collaboration was established with the Cicese Data Science and Machine Learning Laboratory, where they used more than 2 thousand high-definition images, collected by Delgado García, and the application of data augmentation techniques to feed the AI.

“The AI ​​models were manually told where the shark was within the video frames so that it could learn what the species is like and could be identified without human intervention.”

Delgado García worked with two different techniques. One of them based on the DeepLabCut software, an open source tool designed to track animals on video, although not particularly sharks, and the Multi-Scale Patch (MSP) model, developed from scratch to recognize the whale shark more efficiently than DeepLabCut.

After testing both techniques, they concluded that MSP was more efficient and made improvements until it not only detected the shark, but also followed its trajectory frame by frame, even under adverse environmental conditions.

“With what we have at the moment, the model could be used for counting, which allows you to estimate populations, but it can also cover endless options,” Paola anticipated, including estimating sizes or studying behaviors.

Limitations

Although these tools have shown their advantages, they also have limitations. So far, said Delgado García, “we have not been able to collect all the information that is done by the traditional method.”

Some of the obstacles that the drone monitoring model faces is that it is carried out from the beaches, and some have been privatized. Also, the high temperatures in the region, which reach 40 degrees, cause the drone’s batteries to heat up and the overflight cannot exceed 15 minutes.

Despite these limitations, Delgado García considers that the use of artificial intelligence in the study and protection of threatened marine species can be a strategic ally for the species monitoring system, less invasive, more constant, less expensive and automated.

“These tools can provide us with information in a more efficient way to make decisions that allow better management for the conservation of the species,” he concluded.