Sign up for our daily Newsletter and stay up to date with all the latest news!

Subscribe I am already a subscriber

You are using software which is blocking our advertisements (adblocker).

As we provide the news for free, we are relying on revenues from our banners. So please disable your adblocker and reload the page to continue using this site.
Thanks!

Click here for a guide on disabling your adblocker.

Sign up for our daily Newsletter and stay up to date with all the latest news!

Subscribe I am already a subscriber

Light on phone makes image recognition better

Students of the HAS University of Applied Sciences in Den Bosch have investigated the application of automatic image recognition based on phone pictures. Bruce Schoelitsz reports on behalf of the team of students about the project in the framework of PPS 'Weet wat er Leeft' (Know what's lives) at Glastuinbouw Nederland.

In the first study, having several species in one photo proved to be a tricky situation. This was solved in study 2 by strategic analysis. From September onwards, new student projects will start in which the prediction model for recognizing species will be trained further to determine whether the quality can be improved further.

The first study showed that the target species (greenhouse whitefly, tobacco trips, green peach aphid, the predatory mites Montdo and Swirskii, and the parasitic wasp Aphidius colemani) can be recognized with automatic image recognition, but that it becomes more difficult if several species are on the photos at the same time. Of these species, trips were recognized least well, probably because of their small size.

In the second study, which was partly a follow-up of the first study, the problem of multiple species in one photo was solved by analyzing the photos in a strategic manner. As a result, more details are visible, and species are recognized considerably better.

Image recognition program more consistent
In addition, a portable lamp designed by the students themselves was tested. The phone can be attached to the lamp in such a way that there is no direct reflection in the camera, and the entire glue trap can be photographed at once. The effect of the lamp is that the exposure is always the same, regardless of the light conditions in the greenhouse. Trips and whitefly, in particular, were better recognized as a result, and the image recognition program worked more consistently when the lamp was used. This means that the results on different days and with different light conditions in the greenhouse were good as well as comparable.

The accuracy of image recognition without the lamp, on the other hand, differed greatly between different days, as a result of which, for example, it may appear that there is much more whitefly present, while this is not the case. The lamp, therefore, really adds value to the students' own prediction model with all species.

Other glue traps
From September onwards, new student projects will start in which the prediction model for recognizing species will be trained more to determine whether the quality can be improved further. For example, it will be investigated whether very detailed photographs with special cameras/scanners can be used and can improve image recognition. Furthermore, it will be investigated how well the insects are recognized when they are caught in other glue traps, with different colors and wet glue instead of dry glue. But also to what extent accurate trends can be determined. Within this trial, the first comparison with other techniques, such as eDNA, is also made.

This research was carried out by Maxim Lisi, Marith Mol, Ella Ruizendaal, Corné van der Spek, Marijn de Waard, Jorn Walravens, Henk Heijmen, Berkay Helvaci and Robbert-Jan Seeleman.

Source: Glastuinbouw Nederland

Publication date: