Viticulture and wine trade are continuously increasing worldwide and require advanced and effective pest control tools for early pest detection. In Sardinia (southern Italy), the grape production is affected by a number of pests, including Empoasca vitis (Göthe) and Jacobiasca lybica (Bergevin & Zanon) (Hemiptera: Cicadellidae). Once on grapevine, they cause leaf discoloration (hopperburn) and reduce the quality of berries. In this study, the objective was to develop a model for the automatic detection of leafhopper symptoms using convolutional neural networks under field conditions and test the accuracy of predictions. Pictures of symptomatic leaves were acquired in 2021 and 2022 in different areas of Sardinia and classified according to the berry color (white or red) and severity of symptoms (early or severe). The models developed with data of 2022 provided overall high accuracies (F1 score > 0.7), although testing of models of 2022 with data of 2021 showed lower accuracy index values. Symptoms were attributed to J. lybica, according to species identification based on male genitalia morphology. In conclusion, the first attempt of automatic detection of symptoms of green grapevine leafhoppers provided promising results to develop useful tools for early detection of leafhopper infestation.