Popillia Japonica Newman Detection Through Remote Sensing and AI Computer Vision

Food production is one of the most important human activities. Fighting against animals and vegetable weeds is necessary, as climate change and globalisation allow for the diffusion of exotic pests and illnesses. ICTs can provide solutions for a sustainable struggle against these problems. This research aims to detect and estimate the number of Popillia Japonica Newman in vineyards near Novara, Italy. To do that, the pest and vine leaf spectrometric signatures are studied with the goal of identifying different behaviour in the Near InfraRed (NIR) portion of the electromagnetic spectrum. The results are exploited to define a data collection protocol which includes UAS equipped with optical sensors (RGB and NIR). These data are used to create two datasets employed in two AI models, RGB and NIR. The algorithm belongs to the YOLO family and implements the semantic segmentation approach. The performances were compared, and the NIR-based model provided more accurate results, proving that remote sensing and artificial intelligence represent a solution for quick and autonomous Popillia Japonica Newman detection.