Enhancing the estimation of winter wheat leaf area, photosynthetic pigment, and nitrogen content utilizing UAV image features and machine learning optimized by bio-inspired algorithms

• UAV data with bio-inspired algorithms significantly improved wheat inversion accuracy. • Mixed image features inputs average improved R² by 6.2% over spectral inputs alone. • NGO optimization averaged improved the R² of SVM and BPNN models by 16.9%. • NGO-BPNN_VIs+TFs+CFs model whole performed best with R²of 0.811 and RMSE of 0.152. Fast and precise estimation of leaf/canopy-scale information is essential for winter wheat field management. This study aimed to improve the estimation accuracy of winter wheat leaf area, photosynthetic pigment and nitrogen contents by integrating UAV remote sensing data with bio-inspired optimized machine learning models. Data collections were at different growth stages during the 2020-2021 and 2021-2022 winter wheat growing seasons. This study extracted 17 Vegetation indices (VIs), eight texture features (TFs), and three color features (CFs) from UAV images, and their relationships with winter wheat parameters were systematically analyzed. The performance of multiple machine learning models on winter wheat leaf/canopy-scale information were evaluated, including traditional machine learning models and bio-inspired optimization models, using two types of input features (the optimally selected VIs alone and combined VIs+TFs+CFs). The results showed that multi-features fusion (VIs+TFs+CFs) significantly improved estimation accuracy compared with using VIs alone. Among the evaluated models, back propagation neural network (BPNN) generally outperformed linear (multiple stepwise regression, MSR) and kernel-based models (support vector machine, SVM). And bio-inspired optimization further enhanced model performance, with Northern Goshawk Optimization (NGO)-based optimization showing the greater improvement than Genetic Algorithm (GA). Across all parameters, canopy chlorophyll b was highest ( R² =0.811, RMSE =0.152), while the estimation accuracy of leaf nitrogen content was lowest ( R² =0.670, RMSE =0.193). Overall, the integration of UAV multi-features fusion and bio-inspired algorithm optimized machine learning model (NGO-BPNN_VIs+TFs+CFs model) significantly improved the inversion accuracy of leaf area, photosynthetic pigment and nitrogen contents. This provided a reliable and effective method for precision monitoring of winter wheat leaf/canopy information.