Analysis of morphological features and stages of ontogenesis of weeds in automated recognition tasks

This article systematizes the morphological features of weeds relevant to automated image recognition tasks. Four main classes of features–geometric, architectural, surface, and dynamic–are described, and their complementarity in the formation of stable descriptors is demonstrated. The need to account for ontogenetic variability and standardize phenophases using the BBCH and Zadoks scales, as well as adapt these scales for weed species, is substantiated. Based on the analysis, priority weed groups for Belarusian agrocenoses are identified, and criteria for selecting developmental stages for representative training samples are proposed.

1. Weed recognition using deep learning techniques on class-imbalanced imagery / A. S. M. Hasan, F. Sohel, D. Diepeveen [et al.] // Crop and Pasture Science. – 2022. – Vol. 74. – № 6. – P. 628–644. – DOI: 10.1071/CP21626.

2. Meier, U. The BBCH system to coding the phenological growth stages of plants –history and publications / U. Meier, H. Bleiholder, L. Buhr [et al.] // Journal für Kulturpflanzen. – 2009. – Vol. 61. – № 2. – P. 41–52.

3. Справочник сорных растений / Национальная академия наук Беларуси, Научно-практический центр НАН Беларуси по земледелию, Институт защиты растений; сост. А. В. Сташкевич [и др.]; под общ. ред. Е. А. Якимович. – Мн.: Колорград, 2025. – 121 с.