Título:	Automated detection of european canker (Neonectria ditissima) in apple trees via multispectral sensors and computer vision.
Autoria:	ALVES, S. A. M. SILVA NETO, E. C. da MARCHIORETTO, L. de R. GEBLER, L.
Afiliação:	SILVIO ANDRE MEIRELLES ALVES, CNPUV; EDUARDO CARVALHO DA SILVA NETO, CNPS; LUCAS DE ROSS MARCHIORETTO, EMBRAPA UVA E VINHO; LUCIANO GEBLER, CNPUV.
Ano de publicação:	2026
Referência:	In: INTERNATIONAL CONFERENCE ON PRECISION AGRICULTURE, 17., CONGRESSO BRASILEIRO DE AGRICULTURA DE PRECISÃO DIGITAL, 11., 2026, Porto Alegre, RS. Anais...Porto Alegre: AsBraAP; AP&D, de 13 a 16 de julho de 2026.
Conteúdo:	European canker, caused by the fungus Neonectria ditissima, represents one of the major economic challenges for Brazilian pomiculture, severely affecting ‘Gala’ and ‘Fuji’ cultivars. The disease manifests primarily in woody tissues, such as trunks and branches, although it can also cause fruit rot during the pre-harvest stage. Infection occurs obligatorily through wounds, whether natural (leaf scars) or resulting from management practices (pruning and harvesting). Effective disease management depends on early detection and the immediate elimination of inoculum sources. However, conventional manual inspection is time-consuming, costly, and limited in identifying incipient lesions or those hidden by foliage. In this context, this study aims to develop faster and more efficient detection methods through an automated system based on image processing. To enable this system, an image database of apple tree branches was established under field conditions, monitoring the stages of disease development associated with pruning wounds. The experimental design consisted of: an Inoculated Group, comprising 50 branches (across ten plants) subjected to pruning cuts and inoculation with a N. ditissima fungal suspension; and a Control Group, with 50 branches subjected only to cutting and treated with distilled water. Image captures were performed at approximately seven-day intervals using three distinct sensors: conventional RGB, multispectral RedEdge (725 nm), and RG-NIR. A sample of 132 images was annotated with segmentation masks and used to train a YOLOv8nseg model to assess the feasibility in finding European canker on branches. The database obtained from this experiment comprises more than 4,000 images. Preliminary results indicate that physiological changes resulting from infection generate detectable contrasts in the images, especially at wavelengths above 700 nm. This spectral signature highlights the potential of this range for the early identification of diseased branches. The generated database will allow for detailed spectral analysis and the training of convolutional neural networks (CNNs) for the automated recognition of infection patterns. The preliminary training showed promising results, achieving mAP50-95(M) of 0.63, Precision(M) of 0.89, and recall(M) of 0.91, even with a relatively low number of annotated images. Future stages will involve testing different segmentation models, the development of equipment for the detection, quantification, and geospatial localization of diseased branches in the orchard. Such tools are fundamental for the consolidation of precision fruit farming, enabling localized interventions, rapid diagnostics, and a reduction in operational costs for growers.
Thesagro:	Malus Domestica
NAL Thesaurus:	Plant pathology Spectroscopy
Palavras-chave:	Apple disease
Tipo do material:	Artigo em anais e proceedings
Acesso:	openAccess
Aparece nas coleções:	Artigo em anais de congresso (CNPUV)