seeds of innovation
Author(s)
roberto masturzi
Date Issued
2026
Type
conferenceObject
Abstract
Hazelnut (Corylus avellana L.) is increasingly recognized as a major crop in Italy and worldwide. Despite its relatively high resilience, hazelnut production can be significantly affected by several insect and mite pests. Among these, the eriophyid mite Phytoptus avellanae Nalepa has gained growing importance due to the severe yield reductions it causes in various hazelnut-producing regions. This mite attacks both vegetative and reproductive buds, inducing irreversible tissue alterations that lead to the formation of swollen galls, commonly referred to as “big buds”. Infestation levels are typically assessed during winter by counting big buds on randomly selected plants, while mite dispersal—primarily wind-mediated—is monitored using sticky and water traps. However, sampling and identifying the mite fauna( mite size <200 µm) present in hazelnut orchards is extremely labor-intensive and time-consuming, requiring multiple steps such as washing, counting, taxonomic identification, and slide preparation.
This study, conducted in 2025 in the main hazelnut-growing area of Viterbo (central Italy), investigated the feasibility of streamlining and improving mite identification through the application of a YOLO (You Only Look Once) object detection framework. YOLO algorithms enable the simultaneous recognition of multiple objects within images with high accuracy and reduced computational time, making them particularly suitable for the automatic detection of microarthropods.
A dedicated image database of laboratory-acquired samples was constructed, processed, and manually labeled to support YOLO model training. The trained system allows the automatic discrimination of phytophagous mites—further classified into vermiform and fusiform morphotypes—from predatory phytoseiid mites. The proposed approach also evaluates the contribution of data augmentation and image enhancement techniques to the detection of microscopic objects, providing a rapid, non-invasive, and low-cost diagnostic tool for the identification of hazelnut-associated mites.