A sticky trap for insect monitoring, placed on a eucalypt (left), and after removal and initial insect identification (right). The two insects of interest were manually labeled (‘A’ and ‘B’) in the trap image, for training of the deep learning model. Photo by Achiad Sviri.

Insects are everywhere, profoundly impact human lives, and are also dramatically affected by human activities. Unfortunately, our knowledge about insect population trends and locations is generally poor, and often only descriptive. This limits our ability to predict insect outbreaks, migrations or declines, and to plan our farms, cities and conservation policies accordingly. Identifying insects in large quantities and with high precision is difficult and time-consuming, generating a bottleneck in data collection. To open this bottleneck, this research combins a “low-tech” method for capturing flying insects with an image processing technology that automatically and rapidly identifies and counts individual insects.

Over the last year, Tamar Keasar and Avi Bar-Massada from the department of Biology at Oranim, and members of the University of Haifa’s Data Science Research Center, used this approach to monitor a sap-sucking insect and its natural enemy, a tiny wasp. Both insects are of Australian origin, and have recently invaded eucalypt forests in Israel. The sap-feeder damages the forest trees, while the wasp preys on it and may be able to reduce its populations. 

The researchers are capturing the insects using sticky traps placed in the forests (see photos above). A Convolutional Neural Network identifies the insects from photo images of the traps, and measures their body sizes. Their deep learning model currently discriminates the two insects from one another, as well as from other elements such as leaves and other insects, with > 90% accuracy. Based on these data, they will predict the future distribution of the forest pest in Israel, and suggest management strategies to reduce its populations.