Leaf age and plant centre are important phenotypic information of weeds, and accurate identification of them plays an important role in understanding the morphological structure of weeds, guiding precise targeted spraying and reducing the use of herbicides. In this work, a weed segmentation method based on BlendMask is proposed to obtain the phenotypic information of weeds under complex field conditions. This study collected images from different angles (front, side, and top views) of three kinds of weeds (Solanum nigrum, barnyard grass (Echinochloa crus-galli), and Abutilon theophrasti Medicus) in a maize field. Two datasets (with and without data enhancement) and two backbone networks (ResNet50 and ResNet101) were replaced to improve model performance. Finally, seven evaluation indicators are used to evaluate the segmentation results of the model under different angles. The results indicated that data enhancement and ResNet101 as the backbone network could enhance the model performance. The F₁ value of the plant centre is 0.9330, and the recognition accuracy of leaf age can reach 0.957. The mIOU value of the top view is 0.642. Therefore, deep learning methods can effectively identify weed leaf age and plant centre, which is of great significance for variable spraying.

This letter presents a sample-efficient deep reinforcement learning (DRL) based method to address the intricate task of autonomous navigation in dynamic, constrained environments. The proposed method leverages a graph neural network with feature-wise linear modulation modules to effectively extract features from observations modeled as heterogeneous graph. By transferring offline model predictive control (MPC) experience data, which can be generated fully in parallel, the training process can be jump-started to cope with the sparse rewards. Simulation results demonstrate a 98% navigation success rate, surpassing baselines by at least 6%, while halving the training steps. Furthermore, we successfully deploy our policy network on a real robot and contribute our code to the community through open-sourcing: github.com/TIB-K330/drl_planner.