To address the problems of low recognition accuracy and slow processing speed when identifying harvest-stage cabbage heads in complex environments, this study proposes a lightweight harvesting period cabbage head recognition algorithm that improves upon YOLOv8n. We propose a YOLOv8n-Cabbage model, integrating an enhanced backbone network, the DyHead (Dynamic Head) module insertion, loss function optimization, and model light-weighting. To assess the proposed method, a comparison with extant mainstream object detection models is conducted. The experimental results indicate that the improved cabbage head recognition model proposed in this study can adapt cabbage head recognition under different lighting conditions and complex backgrounds. With a compact size of 4.8 MB, this model achieves 91% precision, 87.2% recall, and a mAP@50 of 94.5%—the model volume has been reduced while the evaluation metrics have all been improved over the baseline model. The results demonstrate that this model can be applied to the real-time recognition of harvest-stage cabbage heads under complex field environments.

Reducing damage and missed harvest rates is essential for improving efficiency in unmanned cabbage harvesting. Accurate real-time segmentation of cabbage heads can significantly alleviate these issues and enhance overall harvesting performance. However, the complexity of the growing environment and the morphological variability of field-grown cabbage present major challenges to achieving precise segmentation. This study proposes an improved YOLOv8n-seg network to address these challenges effectively. Key improvements include modifying the baseline model’s final C2f module and integrating deformable attention with dynamic sampling points to enhance segmentation performance. Additionally, an ADown module minimizes detail loss from excessive downsampling by using depthwise separable convolutions to reduce parameter count and computational load. To improve the detection of small cabbage heads, a Small Object Enhance Pyramid based on the PAFPN architecture is introduced, significantly boosting performance for small targets. The experimental results show that the proposed model achieves a Mask Precision of 92.2%, Mask Recall of 87.2%, and Mask mAP50 of 95.1%, while maintaining a compact model size of only 6.46 MB. These metrics indicate superior accuracy and efficiency over mainstream instance segmentation models, facilitating real-time, precise cabbage harvesting in complex environments.

Semantic segmentation of rural roads presents unique challenges due to the unstructured nature of these environments, including irregular road boundaries, mixed surfaces, and diverse obstacles. In this study, we propose an enhanced PP-LiteSeg model specifically designed for rural road segmentation, incorporating a novel Strip Pooling Simple Pyramid Module (SP-SPPM) and a Bottleneck Unified Attention Fusion Module (B-UAFM). These modules improve the model’s ability to capture both global and local features, addressing the complexity of rural roads. To validate the effectiveness of our model, we constructed the Rural Roads Dataset (RRD), which includes a diverse set of rural scenes from different regions and environmental conditions. Experimental results demonstrate that our model significantly outperforms baseline models such as UNet, BiSeNetv1, and BiSeNetv2, achieving higher accuracy in terms of mean intersection over union (MIoU), Kappa coefficient, and Dice coefficient. Our approach enhances segmentation performance in complex rural road environments, providing practical applications for autonomous navigation, infrastructure maintenance, and smart agriculture.

To address the issues of low efficiency, insufficient coverage, and high energy consumption in wheat sowing path planning for large-scale irregular farmland, this study proposes an improved hybrid particle swarm optimization algorithm (TLG-PSO) for autonomous operational path planning. Building upon the standard PSO, the proposed method introduces a Tent chaotic mapping initialization mechanism, a Logistic-based dynamic inertia weight adjustment strategy, and adaptive Gaussian perturbation optimization to achieve precise control of the agricultural machinery’s driving orientation angle. A comprehensive path planning model is constructed with the objectives of minimizing the effective operation path length, reducing turning frequency, and maximizing coverage rate. Furthermore, cubic Bézier curves are employed for path smoothing, effectively controlling path curvature and ensuring the safety and stability of agricultural operations. The simulation experiment results demonstrate that the TLG-PSO algorithm achieved exceptional full-coverage operation performance across four categories of typical test fields. Compared to conventional fixed-direction path planning strategies, the algorithm reduced average total path length by 6228 m, improved coverage rate by 1.31%, achieved average labor savings of 96.32%, and decreased energy consumption by 6.45%. In large-scale comprehensive testing encompassing 1–27 field plots, the proposed algorithm reduced average total path length by 8472 m (a 5.45% decrease) and achieved average energy savings of 44.21 kW (a 5.48% reduction rate). Comparative experiments with mainstream intelligent optimization algorithms, including GA, ACO, PSO, BreedPSO, and SecPSO, revealed that TLG-PSO reduced path length by 0.16%–0.74% and decreased energy consumption by 0.53%–2.47%. It is worth noting that for large-scale field operations spanning hundreds of acres, even an approximately 1% path reduction translates to substantial fuel and operational time savings, which holds significant practical implications for large-scale agricultural production. Furthermore, TLG-PSO demonstrated exceptional performance in terms of algorithm convergence speed and computational efficiency. The improved TLG-PSO algorithm provides a feasible and efficient solution for autonomous operation of large-scale agricultural machinery.

With the advancement of agricultural automation, the demand for road recognition and understanding in agricultural machinery autonomous driving systems has significantly increased. To address the scarcity of instance segmentation data for rural roads and rural unstructured scenes, particularly the lack of support for high-resolution and fine-grained classification, a 20-class instance segmentation dataset was constructed, comprising 10,062 independently annotated instances. An improved StyleGAN2-ADA data augmentation method was proposed to generate higher-quality image data. This method incorporates a decoupled mapping network (DMN) to reduce the coupling degree of latent codes in W-space and integrates the advantages of convolutional networks and transformers by designing a convolutional coupling transfer block (CCTB). The core cross-shaped window self-attention mechanism in the CCTB enhances the network’s ability to capture complex contextual information and spatial layouts. Ablation experiments comparing the improved and original StyleGAN2-ADA networks demonstrate significant improvements, with the inception score (IS) increasing from 42.38 to 77.31 and the Fréchet inception distance (FID) decreasing from 25.09 to 12.42, indicating a notable enhancement in data generation quality and authenticity. In order to verify the effect of data enhancement on the model performance, the algorithms Mask R-CNN, SOLOv2, YOLOv8n, and OneFormer were tested to compare the performance difference between the original dataset and the enhanced dataset, which further confirms the effectiveness of the improved module.