UAV remote sensing monitoring of winter wheat tiller number based on vegetation pixel extraction and mixed-features selection

Tiller number serves as a crucial indicator of total yield in agriculture. Accurately monitoring tiller numbers is essential for guiding variable fertilization to optimize inputs and enhance final yields. Currently, most of the existing winter wheat tillering sensing monitoring methods cannot meet the three requirements of fast, accurate and efficient at the same time. In order to improve the monitoring accuracy and efficiency, this paper proposes a fast UAV remote sensing method for winter wheat tiller number estimation, combined with self-adaptive segmentation framework and multi-feature selection method. The self-adaptive segmentation framework is designed from the perspective of amplifying the difference of spectral information of ground objects. This framework fully considers the influence of complex environment, especially vegetation coverage, on parameter setting while using mathematical function means and polarization ideas. It aims to solve the problem of insufficient applicability of traditional segmentation methods under different environmental conditions. Subsequently, an optimized grey wolf search algorithm CMI-IGWO with multi-strategy fusion is proposed. Its novel evaluation criteria and iterative mechanism provide important algorithm support for the determination of the best inversion feature combination. The search algorithm combines statistics and information theory, comprehensively considers the linear and nonlinear relationships between variables, and guides the gray wolf to select the optimal features. Experiments show that the mixed features covering spectral, texture, color and shape information can make the final tiller number prediction model work best, achieving an R2 value exceeding 0.75. It affirms the efficacy and robustness of the proposed method in estimating winter wheat tiller numbers, providing valuable guidance for precise topdressing. This study provides a new feasible solution for predicting crop tiller number with high accuracy and efficiency using UAV remote sensing technology.