The research team, led by Alireza Sanaeifar of California State University together with Eve Laroche-Pinel, virologist Marc Fuchs, and Luca Brillante, worked across four commercial vineyards in Napa Valley, California, growing Cabernet Sauvignon and Cabernet Franc. Over two growing seasons, they geolocated 714 vines with satellite navigation, collected petiole samples from each, and confirmed infection status using endpoint multiplex PCR. The resulting dataset contained 399 infected and 315 non-infected vines, an approximately balanced distribution that strengthens the reliability of the classification results.

To capture the spectral fingerprints of infection, the team flew a DJI Matrice 600 Pro drone carrying a Senop HSC-2 snapshot hyperspectral camera with 29 spectral bands spanning 520 to 820 nanometers. Flights at 30 meters altitude and 5 meters per second produced imagery with a ground resolution of 2 by 2 centimeters per pixel. Because the camera records its bands sequentially, the researchers orthorectified each band individually and verified that residual misregistration stayed below roughly 5 centimeters. Reflectance was calibrated using a white reference panel placed in the field, and orthomosaics built in Agisoft Metashape allowed each PCR-tested vine to be matched precisely to its canopy pixels.

The spectral analysis revealed how subtle the disease signal really is. Infected and healthy vines showed nearly identical overall reflectance curves, with statistically significant differences concentrated in the green region between 540 and 580 nanometers, where healthy vines reflected more light, likely because infection degrades chlorophyll and alters pigment composition. Infected vines also showed localized reflectance increases around 740 to 760 nanometers, hinting at changes in internal leaf structure. Notably, the widely used NDVI vegetation index failed to separate the classes, while the Green NDVI and the Anthocyanin Reflectance Index both discriminated strongly, the latter consistent with the anthocyanin accumulation that gives red blotch disease its name.

Conventional vegetation indices, however, proved too blunt for reliable diagnosis, which is where the deep learning model enters. The researchers designed CompactSpectralViT, a lightweight patch-based vision transformer tailored to hyperspectral data. Each vine’s variable-sized canopy cube was standardized to 64 by 64 pixels across 29 bands through a three-stage pipeline of proportional resizing, adaptive center cropping, and final adjustment. The cube was then divided into 64 patches of 8 by 8 pixels, each carrying all 29 spectral values, and passed through a two-stage embedding that compressed each patch to just 48 dimensions. Four transformer encoder blocks with three attention heads each then let the model weigh relationships across the entire canopy simultaneously, capturing long-range spectral-spatial dependencies that convolutional networks process only locally.

Evaluated with 10-fold stratified cross-validation, CompactSpectralViT achieved a mean accuracy of 75.3 percent, precision of 74.5 percent, recall of 87.2 percent, an F1-score of 79.8 percent, and a ROC-AUC of 0.744. The high recall is operationally significant: in disease monitoring, a missed infected vine can seed further spread by the three-cornered alfalfa hopper, the virus’s insect vector, whereas a false alarm costs comparatively little. By tuning the classification threshold per fold, the model shifted its decision boundary toward sensitivity, accepting more false positives in exchange for catching nearly nine in ten infected vines.

The compact architecture also delivered a striking computational advantage. Benchmarking on an NVIDIA RTX 3090 showed that CompactSpectralViT required only 16.6 million multiply-accumulate operations per inference, roughly 7.6 times fewer than a convolutional baseline and a hybrid CNN-transformer comparator, and used just 138 megabytes of GPU memory versus 376 and 440 megabytes for the alternatives. It also outperformed both deep learning baselines in accuracy, recall, and ROC-AUC, suggesting that direct patch-level tokenization lets self-attention operate across the whole canopy from the first layer rather than relying on late-stage reasoning bolted onto convolutional features.

Interpretability analysis added a second layer of insight. SHAP-based feature importance and permutation importance, two independent attribution methods, agreed on nine of their ten most informative wavelengths. When the model was retrained on only the ten SHAP-selected bands, accuracy held at 75.6 percent, essentially matching the full 29-band system. The selected wavelengths clustered in three physiologically meaningful regions: the green band from 520 to 570 nanometers, tied to chlorophyll stress; the orange-red region from 610 to 620 nanometers, linked to anthocyanin accumulation; and the near-infrared from 780 to 820 nanometers, sensitive to leaf structure and water content. The authors caution that these findings guide sensor design but do not prove that a true multispectral camera would perform identically, since bandwidth and noise characteristics differ.

Against traditional machine learning on the same ten-wavelength input, the gap was decisive. A support vector machine reached only 62.9 percent accuracy and a random forest 65.1 percent, compared with 75.6 percent for the transformer, which also showed lower variability across folds. Part of this advantage reflects input representation, since the classical models received flattened feature vectors that discard spatial structure, but the result still underscores that self-attention over spectral-spatial patches extracts disease signatures that conventional classifiers miss. The accuracy itself is comparable to earlier ground-based hyperspectral studies of the virus, but the new work achieves it while imaging entire vineyard blocks by drone rather than one vine at a time, and across four sites and two cultivars.

The study’s authors are candid about limitations. Because sites and seasons were pooled across validation folds, transferability to unseen vineyards remains untested, and the infected class mixed symptomatic and asymptomatic vines, so the reported accuracy reflects detection under mixed symptom expression rather than purely asymptomatic diagnosis. Still, the work demonstrates that lightweight transformer architectures can perform credibly with limited labeled data under real field conditions, and it charts a practical path forward: automated vine segmentation, longitudinal monitoring of individual vines, integration with precision roguing programs, and eventually multi-disease frameworks that distinguish viral, fungal, and abiotic stresses from the air. For an industry where infection costs can reach tens of thousands of dollars per hectare over a vineyard’s lifetime, a drone and a compact neural network may soon become standard tools in the fight against an invisible pathogen.

Subject of Research: UAV hyperspectral imaging and a lightweight vision transformer for detecting grapevine red blotch virus in vineyards

Article Title: CompactSpectralViT: A lightweight vision transformer for grapevine red blotch virus detection from UAV hyperspectral imagery

Article References: Sanaeifar, A., Laroche-Pinel, E., Fuchs, M., & Brillante, L. (2026). CompactSpectralViT: A lightweight vision transformer for grapevine red blotch virus detection from UAV hyperspectral imagery. Smart Agricultural Technology, 15, Article 102599. https://doi.org/10.1016/j.atech.2026.102599

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102599

Keywords: grapevine red blotch virus, hyperspectral imaging, vision transformer, UAV remote sensing, precision viticulture, deep learning, plant disease detection, Napa Valley, PCR validation, feature importance, SHAP, multispectral sensors

Grapevines across central and southern Europe face persistent pressure from two green leafhopper species, Jacobiasca lybica and Hebata vitis. These insects feed on phloem sap and produce a distinctive injury known as hopperburn, which begins as yellowing along leaf margins and can progress to reddening or yellowing between the veins, depending on whether the cultivar produces red or white grapes. Under heavy infestation, leaves may dry out entirely, reducing photosynthetic activity and ultimately compromising both yield and berry quality. Because the two leafhopper species generate visually indistinguishable symptoms, the researchers designed their system to detect leafhopper-induced damage in general rather than to identify the culprit species.

Diagnosing hopperburn in the field is notoriously difficult. The symptoms overlap with those caused by viral diseases such as grapevine leafroll-associated viruses, phytoplasma diseases like Flavescence doree, fungal pathogens, nutritional deficiencies, and herbicide drift. Current monitoring relies on yellow sticky traps for adults and laborious visual inspection of leaves for nymphs, but adult captures do not correlate with actual plant damage, so control decisions must be based on leaf infestation levels that are extremely time-consuming to assess across large areas. An automated system capable of flagging symptomatic leaves could transform this workflow.

The research team collected images over two growing seasons, from June to October 2021 and July to August 2022, across 23 commercial vineyards in Sardinia on 21 sampling dates. The vineyards encompassed 15 white and red wine cultivars differing in age, training system, and pest management strategy. Images were captured with commercial smartphones, such as an iPhone 12, at roughly 50 centimeters from symptomatic leaves under natural and deliberately varied light conditions, with backgrounds including healthy leaves, woody organs, and landscape elements. In total, 316 symptomatic images were selected in 2021 and 2,561 in 2022. Expert entomologists reviewed every image, and any showing ambiguous symptoms that might instead reflect potassium deficiency, viral infection, or other disorders were excluded to keep the ground truth clean.

Each image was manually annotated with bounding boxes around symptomatic leaves using the VGG Image Annotator software, producing four symptom classes: early and severe symptoms on white cultivars, and early and severe symptoms on red cultivars. To confirm which leafhopper species was present, the team deployed yellow sticky traps and identified adult males under a light microscope at 100x magnification using genitalia morphology. Across 115 microscope slides, every identifiable specimen proved to be J. lybica, and no males of H. vitis were recorded in the investigated vineyards.

Three modern object detection architectures were benchmarked: YOLOv5 as a widely adopted baseline, the newer YOLOv11, and RF-DETR, a transformer-based detector. All models were trained with identical datasets, annotation protocols, and training schedules on a Google Colab virtual machine equipped with an Nvidia Tesla P100 GPU. Data augmentation, including horizontal flipping, blurring, contrast enhancement via CLAHE, and HSV color adjustments, was applied to compensate for the limited dataset size. Performance was measured using standard object detection metrics, with the intersection over union threshold set at 50 percent, and macro mean Average Precision serving as the primary metric because it jointly evaluates localization and classification across all symptom classes.

The most striking result concerned the evaluation strategy itself. When images were randomly split into training, validation, and test sets, all three models performed superbly, with RF-DETR reaching a macro-mAP of 0.975, YOLOv5 0.944, and YOLOv11 0.899. But random splitting allows nearly identical images from the same vineyard, date, and cultivar to appear in both training and test data, inflating apparent accuracy. When the researchers instead partitioned the dataset so that no vineyard-cultivar combination was shared between training and test sets, performance dropped dramatically, with macro-mAP falling by 30 to 38 percentage points to between 0.517 and 0.678. This decline mirrors earlier findings in grape disease detection, where models trained on one cultivar achieved a true positive rate of 0.98 for Flavescence doree on Chardonnay but collapsed to 0.08 when tested on Ugni-Blanc.

Crucially, the analysis revealed that the residual errors under realistic partitioning stemmed mostly from misclassification among visually similar symptom classes rather than from failure to detect damage. When all hopperburn categories were treated as a single detection class, average precision exceeded 0.96 for all three models on the four-class partitioned dataset and remained above 0.95 on a three-class variant. Merging the two early symptom classes into a single category also boosted performance substantially, raising overall macro-mAP by 18 to 26 percentage points. Severe symptoms on red cultivars, with their pronounced red marginal and interveinal discoloration, were detected with high accuracy regardless of architecture or partitioning strategy, while the subtle white early class proved hardest to discriminate. Statistical comparisons using Fisher’s exact tests with Holm correction confirmed that dataset partitioning exerted a stronger influence on performance than the choice of detector architecture.

The practical implications are significant. Because misclassifications among symptom classes are far less consequential than missed detections or false alarms in an infestation-monitoring context, the study suggests that the operational value of these detectors as early warning systems may be underestimated by strict class-based metrics. The authors envision integration into mobile applications or autonomous ground platforms for real-time in-field monitoring and spatial mapping during routine vineyard operations. Such tools would fit a broader agricultural context in which mating disruption has reduced insecticide use against other key pests, the European Union’s Farm to Fork Strategy targets major pesticide cuts by 2030, and only four new synthetic active substances have been registered since 2011.

The study is not without limitations. All images were collected within a single viticultural region, and the team emphasizes that further validation under different environmental conditions and viticultural contexts is still required. Future work should expand the dataset to additional grape-growing regions with more balanced representation of cultivars, symptom severities, infestation-free vineyards, and visually similar biotic and abiotic disorders. Nevertheless, by pairing the first dedicated multi-temporal, multi-varietal field dataset of hopperburn with a rigorous, cultivar-independent evaluation framework, the research provides a foundation for objectively comparing future computer vision tools and marks a concrete step toward automated, sustainable pest management in the world’s vineyards.

Subject of Research: Deep learning detection of grapevine leafhopper hopperburn symptoms in vineyards

Article Title: Automatic detection of grapevine hopperburn symptoms using deep learning under field conditions

Article References: Cocco, A., Ghiani, L., El Moussaoui, S., Mannu, R., Sassu, A., Deidda, A., Santoro, F., Lentini, A., & Gambella, F. (2026). Automatic detection of grapevine hopperburn symptoms using deep learning under field conditions. Smart Agricultural Technology, 15, Article 102576. https://doi.org/10.1016/j.atech.2026.102576

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102576

Keywords: deep learning, object detection, grapevine, leafhoppers, hopperburn, precision viticulture, YOLO, RF-DETR, pest monitoring, computer vision, Sardinia, sustainable agriculture