Deep Learning Based Polyculture Structural Phenotyping


Yilmaz A., Erdem Y., BOZMA AYDIN H. I.

35th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2025, İstanbul, Türkiye, 31 Ağustos - 03 Eylül 2025, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/mlsp62443.2025.11204237
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: deep learning, plant segmentation, point cloud analysis, polyculture, structural phenotyping
  • Boğaziçi Üniversitesi Adresli: Evet

Özet

Structural phenotyping in polyculture farming is challenging due to the morphological diversity across plant species. We propose a novel end-to-end pipeline for autonomous estimation of four structural traits-length, width, height, and canopy area-from RGB-D data. The system employs a deep learning model for plant detection and segmentation, trained on a custom-collected polyculture RGB-D dataset spanning two months. Segmented data are converted into point clouds for per-plant structural analysis. Unlike prior methods, our pipeline generalizes across multiple species, enabling accurate phenotype extraction in diverse crop beds. We also introduce a new RGB-D polyculture dataset and validate the pipeline's performance through experiments in a real-world polyculture garden, demonstrating robust, species-agnostic phenotyping.