<p>Precise delineation of agricultural parcels from aerial images is a central issue for land use monitoring, agricultural planning and the optimization of cultivation practices. In rural areas, the lack of up-to-date, interoperable parcel boundary data hinders the implementation of effective agricultural policies and the modernization of farming practices. This study proposes a processing workflow integrating deep learning within a Spatial Data Infrastructure (SDI), to automate parcel delineation from RGB orthoimages at 1-meter resolution. This SDI constitutes a structuring framework that guarantees the interoperability, sharing and continuous updating of results, while facilitating their exploitation. Six image segmentation models U-Net, DeepLabV3+, TransUNet, MFCA-Net, HRNet-OCR and BiSeNetV2 were evaluated using RGB orthoimages with 1-meter spatial resolution. The U-Net and DeepLabV3 + models deliver the best results by achieving a Dice coefficient of 0.9384 and 0.9355 and an F1 Score of 0.9416 and 0.9385 and an IoU of more than 0.87. The second-best results were obtained by MFCA-Net and TransUNet which demonstrate good shape recognition abilities although they sometimes produce over-segmentation. The performance of HRNet-OCR and BiSeNetV2 is limited mainly because they struggle with small objects detection which results in high Loss values (0.1168 and 0.0932) and IoU below 0.83. The U-Net model demonstrates the highest stability for applications that need precise contour detection and strong generalization capabilities. The implementation of this intelligent data into an SDI system creates a practical solution which drives scientific research and technological advancement and enables specific agricultural and territorial management innovations.</p>