Skip to main content

Resultados relacionados

Resultados totales (Incluyendo duplicados): 3
Encontrada(s) 1 página(s)

IRTA Pubpro. Open Digital Archive
Publication Artículo científico (JournalArticle)

Individual Segmentation of Intertwined Apple Trees in a Row via Prompt Engineering

  • Metuarea, Herearii
  • Laurens, François
  • Guerra, Walter
  • Lozano, Lidia
  • Patocchi, Andrea
  • Van Hoye, Shauny
  • Dutagaci, Helin
  • Labrosse, Jeremy
  • Pejman, Rasti
  • Rousseau, David
Computer vision is of wide interest to perform the phenotyping of horticultural crops such as apple trees at high throughput. In orchards specially constructed for variety testing or breeding programs, computer vision tools should be able to extract phenotypical informa tion form each tree separately. We focus on segmenting individual apple trees as the main task in this context. Segmenting individual apple trees in dense orchard rows is challenging because of the complexity of outdoor illumination and intertwined branches. Traditional methods rely on supervised learning, which requires a large amount of annotated data. In this study, we explore an alternative approach using prompt engineering with the Segment Anything Model and its variants in a zero-shot setting. Specifically, we first detect the trunk and then position a prompt (five points in a diamond shape) located above the detected trunk to feed to the Segment Anything Model. We evaluate our method on the apple REFPOP, a new large-scale European apple tree dataset and on another publicly available dataset. On these datasets, our trunk detector, which utilizes a trained YOLOv11 model, achieves a good detection rate of 97% based on the prompt located above the detected trunk, achieving a Dice score of 70% without training on the REFPOP dataset and 84% without training on the publicly available dataset.We demonstrate that our method equals or even outperforms purely supervised segmentation approaches or non-prompted foundation models. These results underscore the potential of foundational models guided by well designed prompts as scalable and annotation-efficient solutions for plant segmentation in complex agricultural environments., This research was funded by the European Union’s Horizon Europe Research and Innovation Programme under PHENET project, Grant Agreement No. 101094587., info:eu-repo/semantics/publishedVersion
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

NRT-GSF: A novel near-real-time ground-satellite fusion algorithm to retrieve daily green area index at field scale

  • Li, Wenjuan
  • Buis, Samuel
  • Verger, Aleixandre
  • Jay, Sylvain
  • Ren, Zihan
  • Wu, Wenbin
  • Jiang, Jingyi
  • Comar, Alexis
  • Solan, Benoit De
Near-real-time (NRT) daily crop monitoring at the field scale is crucial for precision agriculture, yet remains challenging due to limitations in the spatial or temporal resolution of existing remote sensing methods. While Sentinel-2 provides adequate spatial resolution for field-level applications, its temporal resolution is insufficient for capturing rapid crop dynamics, especially in cloudy regions. Existing spatiotemporal fusion techniques require multiple clear-sky images and lack true NRT capability, while ground-based sensors offer continuous monitoring but with limited spatial coverage. To address these limitations, this study develops the Near-Real-Time Ground-Satellite Fusion (NRT-GSF) algorithm, a novel approach based on a Bayesian dynamic linear model and Kalman filtering. The algorithm uniquely integrates Sentinel-2 imagery with continuous measurements from Internet of Things for Agriculture (IoTA) systems to generate daily 10-m Green Area Index (GAI) products. Its recursive framework supports both forward prediction in NRT mode following satellite overpasses and backward updating to refine historical profiles. Implemented over French wheat fields using 34 IoTA systems and Sentinel-2 time series from 2019, the algorithm effectively enhanced spatiotemporal completeness and accuracy (R = 0.75–0.98, RMSE = 0.1–0.49). A comprehensive leave-one-out Sentinel-2 evaluation demonstrated its superiority over the current Consistent Adjustment of the Climatology to Actual Observations (CACAO) algorithm. Ground validation using handheld RGB cameras further confirmed the accuracy of the GAI products from the new algorithm (RMSE = 0.5). The NRT-GSF framework offers a robust and operationally solution for daily, high-resolution crop GAI mapping in NRT mode, and it can be extended to other traits or applications in the near-real-time context., Wenjuan Li acknowledges the support from National Key Research and Development Program of China (2023YFD2300500) and National Natural Science Foundation of China (42201388). Marie Weiss acknowledges the support from the European Union's Horizon Europe Research and Innovation Programme under grant agreement N° 101094587 for the writing of this paper. Aleixandre Verger acknowledge support from Spanish Ministry of Science, Innovation and Universities grant PID2023-152114OB-I00. This work was initially funded by the French FUI IOTA Project (FUI 24 – D0S0071619)., Peer reviewed
Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

Individual Segmentation of Intertwined Apple Trees in a Row via Prompt Engineering

  • Metuarea, Herearii
  • Laurens, François
  • Guerra, Walter
  • Lozano, Lidia
  • Patocchi, Andrea
  • Van Hoye, Shauny
  • Dutagaci, Helin
  • Labrosse, Jeremy
  • Pejman, Rasti
  • Rousseau, David
Computer vision is of wide interest to perform the phenotyping of horticultural crops such as apple trees at high throughput. In orchards specially constructed for variety testing or breeding programs, computer vision tools should be able to extract phenotypical informa tion form each tree separately. We focus on segmenting individual apple trees as the main task in this context. Segmenting individual apple trees in dense orchard rows is challenging because of the complexity of outdoor illumination and intertwined branches. Traditional methods rely on supervised learning, which requires a large amount of annotated data. In this study, we explore an alternative approach using prompt engineering with the Segment Anything Model and its variants in a zero-shot setting. Specifically, we first detect the trunk and then position a prompt (five points in a diamond shape) located above the detected trunk to feed to the Segment Anything Model. We evaluate our method on the apple REFPOP, a new large-scale European apple tree dataset and on another publicly available dataset. On these datasets, our trunk detector, which utilizes a trained YOLOv11 model, achieves a good detection rate of 97% based on the prompt located above the detected trunk, achieving a Dice score of 70% without training on the REFPOP dataset and 84% without training on the publicly available dataset.We demonstrate that our method equals or even outperforms purely supervised segmentation approaches or non-prompted foundation models. These results underscore the potential of foundational models guided by well designed prompts as scalable and annotation-efficient solutions for plant segmentation in complex agricultural environments., This research was funded by the European Union’s Horizon Europe Research and Innovation Programme under PHENET project, Grant Agreement No. 101094587., info:eu-repo/semantics/publishedVersion