Vision-based Aphrophoridae foam detection for sustainable management of Xylella fastidiosa

Xylem sap-feeding insects, such as adult Aphrophoridae, commonly known as spittlebugs, are vectors of the plant pathogenic xylem-limited bacterium Xylella fastidiosa, a causal agent of a number of severe diseases, among which the Olive Quick Decline Syndrome (OQDS) has resulted in an unprecedented decimation of olive trees in the Mediterranean area. Aphrophoridae life cycle and behavior are characterized by a weak stage, i.e., the juvenile stage, during which the insects live solitary on stems covered by a self-produced foamy fluid (froth) protecting them from dehydration and thermal stress. Juvenile vectors are the perfect target for a control action to mitigate the transmission due to adults. In this work, an automated vision-based system to detect the nymph froth directly in the field is proposed. It exploits a semi-supervised DeepLabv3+ architecture with ResNet18 backbone to segment images acquired by a consumer-grade camera and automatically recognize the spittles. The system is intended to guide the action and assess the efficiency of an aeraulic machine able to generate an airstream with shape and thrust proper to manage the target organisms by ex ante and ex post control action data comparison. Experimental results carried out in an area covered by wild tall grass show that the proposed approach allows for recognizing froth instances with recall and precision of about 76% and 66%, respectively, despite the low quality of the input images as well as the challenges arising from small item size, object resemblance in color, occlusions, and changing lighting conditions.