Automatically detect insect pests

  • Processing of large catches and up-to-date evaluations thanks to efficient analysis

  • Large-scale monitoring thanks to reproducible and scalable evaluations

  • High classification quality even for harmful insects with subtle morphological differences and when bycatch and dirt occur, thanks to the latest artificial intelligence (AI) processes

previous arrow
next arrow

Aphid-Identification with Artificial Intelligence

What it’s about

Aphids are not only a problem in the home garden. They have also been increasingly used in agriculture in recent years. Aphids damage infested plants by sucking on them and transmitting viruses that can cause plant diseases. In order to be able to protect arable crops from harmful insects in practice, it is therefore very important to know exactly when and where they appear. This is the only way to combat harmful insects in the best possible way and avoid the need for prophylactic use of plant protection products.

Insect pests are currently recorded in practice by examining the crop plants in the field or by catching the insects from the air using suction traps or yellow bowls. Correct identification of the trapped insect pests is then extremely important in order to be able to decide what to do. However, identifying aphids based on their external characteristics is very complex and therefore very time-consuming and expensive. However, the well-trained staff required for this is becoming fewer and fewer. In practice, this presents many plant protection services with personnel bottlenecks that can hardly be solved, especially when the aphids are in flight.

In order to enable the insect samples to be evaluated as efficiently and promptly as possible, we have developed a software-hardware solution in cooperation with the Julius Kühn Institute (JKI) which, on the basis of current AI methods, enables automated detection of over 20 aphid species relevant to agriculture, some with high morphological similarity. The software is characterized by its high classification quality, its analysis speed and robustness to sample contamination and bycatch that occur in practice.

The development is made possible by funding from the Federal Ministry of Food and Agriculture (BMEL) as part of the AI2 project.

Sample recording with linear shift

The structure of the recording situation is shown in the video below. A Basler camera mounted on a three-axis linear shift serves as the recording device. The insect sample is recorded automatically by moving the camera device once over the entire sample. The individual images recorded here are combined so that a recording of the entire recording is available.

Segmentation and Classification of Aphids

The videos demonstrate how the AI works. In the microscope images of the samples, the aphids are first segmented into individual instances. The detected specimens are marked by an ellipse. For each of the aphids found, a classification of the associated species is made, with the results being given as a histogram of the corresponding probabilities. The classification takes place as an assignment to the aphid species with the highest assignment probability. In the vast majority of cases, there is a clear assignment to a species, which reflects the robustness of the classification results.

Did we caught your interest?

If you would like more information or are interested in a similar application scenario, please contact us. We’re happy to help: