On behalf of Rijkswaterstaat, Aquila Ecology investigated whether it is possible to inventory road verges from a moving car. With this project we are pushing the boundaries of what is possible when creating images and analyzing them with AI. And with results. Although there are still plenty of challenges in recognizing plants and vegetation types, it appears to be quite possible to recognize larger plant species. For example, we were able to detect 4 different exotic species on a short route: Japanese knotweed, Tree of Heaven, Giant Hogweed and Pampas Grass, while the two trained ecologists in the car only noted Japanese knotweed and Giant Hogweed. We do this by taking photos with a short shutter speed while driving at 80 – 100 km/h. These photos are then analyzed by AI.


Photographing roadsides

The photos that can be taken from a moving car with this method are of surprisingly high quality. Small species such as the soft stork's bill can also be clearly recognized in the photos. To achieve this, we tested two different photography methods and also created, trained and tested a GAN (Generative Adversarial Network) to combat blur. This technique is known from websites that allow entire images to be composed based on some text, but can also be used to improve images. We have also built a platform that can be slid out and in by the driver of the car, with the camera mounted on it, which can also be aimed up and down. In principle, one person could inventory roadsides.


Recognition with AI

We took approximately 150 photos per kilometer, which were then analyzed by AI. This showed that large species were well recognized, but that smaller species were not always identified correctly. It is very likely that the training data behind the AI ​​model does not correspond sufficiently with the image created from the car. A quick test showed that recognition could be greatly improved by improving training data.

Although not all plants are recognized correctly, the amount of data is so large that general patterns can be extracted from it: the correctly recognized plant species are often represented enough to compensate for the incorrect recognitions. Using a method developed by Aquila Ecology, we converted the recognized plant species into vegetation types according to the RWS typology. This showed that the correct vegetation types can be predicted with around 50% accuracy. The data can be read directly into GIS programs.



The developed technology makes it possible to monitor hundreds of kilometers of roadsides at low cost. In any case, the distribution of large exotic species and other large, recognizable plants can be clearly visualized. In the long term it will probably also be possible to obtain an even more detailed picture of the vegetation, even the presence of grass species such as glossy oats and white bulb.