The ability to rapidly respond to insect-pest attacks or to provide the assessment of pollination status are of crucial importance in todays agriculture. To do that, it is necessary to identify insects directly in the field, thus allowing timely reaction to save crops, such as using the right insecticides or reducing the use of incorrect ones.
Researchers from BioSense Institute have developed a system for automated insect identification based on pattern recognition and image processing. Namely, the automatic species discrimination is performed based on the detection and extraction of vein junctions in wing venation. The system currently operates for hoverflies as model organism, but is continually being upgraded to other insects groups.
Images of the wing of four selected hoverfly species belonging to two different genera. The constellations of numbered red dots at vein junctions reveal the genera-specific information used for species discrimination.
HOG (Histogram of Oriented Gradients) features were used to identify the vein junctions. Two images were used for visualization: one depicting discrete histograms of gradient orientation on the level of single cells in the image and one which representing its grayscale counterpart. Nine discrete orientations were defined in the range 0–180°. Image gradient computation has been performed using Sobel filter.
The developed system enables connected smart devices to identify insects in the field and to provide means to map and track status of particular insect groups on a wider area, therefore presenting the hearth of the BioSense pest and pollination monitoring tool.
In addition, this approach can be used for other applications, such as species delimitation, or as field trip journal or insect monitoring tool which saves all important insect-related data in the field together with date and time, GPS coordinates, weather conditions and so on.
An example of automatic detection of vein junctions (shown as red dots) in wing images of four selected hoverfly species. Since detector is based on sliding-window, each vein junction is detected several times in the same image. Based on detector’s confidence level, different significance is given to each of the multiple detections corresponding to the same junction.
Species delimitation: geo-phenogram based on wing shape changes among populations of M. avidus complex showing spatial distribution of different phenotypes.