Jan 13 2021
Researchers are merging artificial intelligence (AI) and the latest computer technology with biological know-how to identify insects that possess supernatural speed. This paves the way for new possibilities to illustrate unidentified species and to pursue the life of insects across time and space.
Insect monitoring cameras in a remote area in East Greenland. Image Credit: Toke T. Høye
Although insects constitute the most diverse group of animals on Earth, just a small percentage of these have been discovered and officially described. There are so many species that finding them all in the near future would be improbable.
Such a vast diversity among insects also suggests that they possess very varied life histories and play different roles in the ecosystems.
For example, the life of a hoverfly in Greenland is very different from that of a mantid in the Brazilian rainforest. However, even within each of these two groups, several species can be found, each with their own unique characteristics and environmental roles.
To study the biology of each species and its interactions with other species, it is essential to catch, identify, and count numerous insects. It is obviously a very laborious process, which, to a great degree, has inhibited the ability of researchers to gather insights into how external elements impact the life of insects.
New research reported in the Proceedings of the National Academy of Sciences describes how latest computer technology and AI can rapidly and efficiently identify and count insects. It is a massive step forward for the researchers to be able to comprehend how this critical group of animals transforms through time—for instance, in reaction to climate change and loss of habitat.
Deep Learning
With the help of advanced camera technology, we can now collect millions of photos at our field sites. When we, at the same time, teach the computer to tell the different species apart, the computer can quickly identify the different species in the images and count how many it found of each of them. It is a game-changer compared to having a person with binoculars in the field or in front of the microscope in the lab who manually identifies and counts the animals.
Toke T. Høye, Study Lead and Senior Scientist, Arctic Research Centre, Department of Bioscience, Aarhus University
The international study team included statisticians, biologists, as well as electrical, mechanical, and software engineers.
The approaches illustrated in the study go by the broad term deep learning and are forms of AI typically used in other areas of research such as in the creation of driverless cars. But presently, the team has shown how the technology can be a substitute for the painstaking task of manually tracking insects in their natural setup, as well as the tasks of classifying and identifying insect samples.
We can use the deep learning to find the needle in the hay stack so to speak - the specimen of a rare or undescribed species among all the specimens of widespread and common species. In the future, all the trivial work can be done by the computer and we can focus on the most demanding tasks, such as describing new species, which until now was unknown to the computer, and to interpret the wealth of new results we will have.
Toke T. Høye, Study Lead and Senior Scientist, Arctic Research Centre, Department of Bioscience, Aarhus University
Moreover, there are a number of tasks ahead with respect to the exploration of insects and other invertebrates, under the field of entomology. One thing is the absence of solid databases to compare unidentified species to those which have already been defined, but also because a proportionally bigger share of scientists focus on recognized species like mammals and birds. The team hopes to use deep learning to quickly advance knowledge about insects significantly.
Long-time Series are Necessary
To comprehend how insect populations transform through time, observations have to be carried out in the same place and in the same manner over extended periods of time. It is crucial with long-time series of data.
Certain species become more plentiful and others rarer, but to appreciate the mechanisms that cause these variations, it is important that the same observations are carried out each and every year.
A simple technique is to place cameras in the same location and capture pictures of the same local area. For example, cameras can click a picture each minute. This will render tons of data, which over the years can offer insights regarding how insects react to warmer climates or to the variations caused by land management. Such data can become a vital tool to guarantee a proper equilibrium between human use and safeguarding of natural resources.
There are still challenges ahead before these new methods can become widely available, but our study points to a number of results from other research disciplines, which can help solve the challenges for entomology. Here, a close interdisciplinary collaboration among biologists and engineers is critical.
Toke T. Høye, Study Lead and Senior Scientist, Arctic Research Centre, Department of Bioscience, Aarhus University
Journal Reference
Høye, T, T., et al. (2021) Deep learning and computer vision will transform entomology. Proceedings of the National Academy of Sciences. doi.org/10.1073/pnas.2002545117.