The combine harvester equipped with attitude-adjustment functionality significantly enhances its adaptability to complex terrain but often struggles to maintain the reliability of its mechanisms. Therefore, investigating the dynamic load characteristics of the attitude-adjustment mechanism becomes imperative. This article employed the DEM–FMBD (Discrete Element Method–Flexible Multibody Dynamics) bidirectional coupling simulation method to establish a multibody dynamic model of a tracked combine harvester. The study delved into the interaction mechanism and dynamic stress response characteristics between the tracked chassis and the complex terrain under various height adjustments, lateral adjustment angles, longitudinal adjustment angles, and different field-ridge crossing methods. Finally, the accuracy of the coupled simulation model was validated through a constructed stress detection system. The research findings revealed that the displacement and tilt angle deviation of the hydraulic cylinders utilized to execute the chassis adjustment actions in the constructed coupled simulation model was less than 5%, and the deviation between the simulation results and the actual maximum dynamic stress under multiple working conditions ranged from 7% to 15%. This verification confirmed the effectiveness of the DEM–FMBD coupled simulation method. Under different adjustment conditions, the maximum stress position was consistently distributed in the same area of the left-front and left-rear rotating arms. The primary and secondary effects of the various parts of the adjustment mechanism on the overall reliability of the chassis were as follows: left front > right front > left rear > right rear. By implementing the middle height with the adjustment strategy, the dynamic stress extreme value of the adjustment mechanism can be effectively reduced by 21.98%, thereby enhancing the structural stability of the chassis.

It is difficult to extract small and dense objects with random state, such as grain and impurity, in image of vehicle-mounted dynamic rice grain flow on combine harvester. Therefore, this paper improves Deeplabv3+ by constructing MobileNetv2 in coding layer and adding ECA(Efficient Channel Attention) to Encoder and Decoder to improve extraction accuracy of high-dimensional features in images with a large number of objects with random state. In addition, the YOLOv4 is improved by using Mixup in preprocessing, constructing Mish in Neck and Head, adding ECA to Neck and Prediction of BackBone to improve training precision of small and dense objects and reducing effect of gradient disappearance. And the impurity/breakage rates are assessed based on relationship model between pixel area and quality, improved Deeplabv3+ and YOLOv4. The proposed method was verified by experiments with images acquired on intelligent combine harvester. Compared with existing Deeplabv3+, YOLOv4, U-NET, BP, the extraction accuracy by improved method increased by more than 4.01%. The average relative error and time of impurity/breakage assessment by proposed method were 7.69% and 1.56s. The proposed method can accurately and rapidly assess impurity/breakage rates for dynamic rice grain flow on combine harvester, and further realize closed-loop control of intelligent harvesting operation.

The precise, real-time delineation of rice lodging areas constitutes a fundamental prerequisite for the adaptive operation of unmanned combine harvesters. However, existing deep learning methods struggle to resolve a critical limitation: achieving an optimal equilibrium between robust regional morphological perception—which is crucial for irregular lodging patterns—and the ultra-low computational overhead demanded by resource-constrained edge terminals. To address this specific constraint, StarNet-RiceSeg is proposed as a lightweight semantic segmentation network explicitly tailored for unmanned harvesters. Initially, the architecture incorporates the minimalist StarNet as its backbone. By leveraging the unique “Star Operation,” it implicitly maps features into a high-dimensional nonlinear space, thereby significantly augmenting feature discriminability while drastically curtailing computational overhead. Furthermore, to mitigate the misdetection issues stemming from the textural similarity between lodged and upright rice, the Rice Spatial Attention (RSA) module was designed. By intensifying feature interaction within the spatial dimension, this module steers the network to focus on the cohesive morphology of lodged regions while effectively suppressing background noise. Experiments conducted on a self-constructed high-resolution rice lodging dataset demonstrate that StarNet-RiceSeg achieves a mIoU of 94.42%, significantly outperforming mainstream models such as U-Net, DeepLabV3+, SegNet and HRNet. Notably, the model maintains a compact footprint with only 8.01 million parameters and a computational load as low as 9.32 GFLOPs. Following optimization with TensorRT, the system achieved a real-time inference speed of 32.51 FPS on the NVIDIA Jetson Xavier NX embedded platform. These results indicate that StarNet-RiceSeg provides a high-precision, low-latency solution for perceiving rice lodging areas in complex field environments, facilitating unmanned precision harvesting.