Automating counting of Roseau cane scale using computer vision

Roseau cane (Phragmites australis (Cav) Trin. Ex Steud.) is the dominant plant species in the Mississippi River Delta, contributing to the region's stability with its extensive root system and substantial above-water biomass. Since 2016, populations of cane have been experiencing die-offs due to a number of seasonal stressors, one of which is the infestation of roseau cane scale (RCS), Nipponaclerda biwakoensis (Kuwana). The effect of RCS infestation is so significant that even 20 live scales per meter of stem can result in a 28% reduction in cane biomass. Monitoring the spread of RCS is essential for the preservation of roseau cane. However, traditional sampling methods are labor-intensive and require repeated visual inspection. This research addresses this challenge by applying the emerging field of object detection using computer vision. We trained an object detection model capable of detecting this scale using collected field imagery with a precision of 78.90%, a recall of 73.40%, and a mean average precision of 80.80%. This model could potentially accelerate sampling efforts, enabling scalable collection on RCS distribution and density. This technological advancement could also support the use of robotics or applications for spatial monitoring of scale dynamics.