This study addresses the common challenges in medical image segmentation and recognition, including boundary ambiguity, scale variation, and the difficulty of modeling long-range dependencies, by proposing a unified framework based on a hierarchical attention mechanism. The framework consists of a local detail attention module, a global context attention module, and a cross-scale consistency constraint module, which collectively enable adaptive weighting and collaborative optimization across different feature levels, thereby achieving a balance between detail preservation and global modeling. The framework was systematically validated on multiple public datasets, and the results demonstrated that the proposed method achieved Dice, IoU, Precision, Recall, and F1 scores of 0.886, 0.781, 0.898, 0.875, and 0.886, respectively, on the combined dataset, outperforming traditional models such as U-Net, Mask R-CNN, DeepLabV3+, SegNet, and TransUNet. On the BraTS dataset, the proposed method achieved a Dice score of 0.922, Precision of 0.930, and Recall of 0.915, exhibiting superior boundary modeling capability in complex brain MRI images. On the LIDC-IDRI dataset, the Dice score and Recall were improved from 0.751 and 0.732 to 0.822 and 0.807, respectively, effectively reducing the missed detection rate of small nodules compared to traditional convolutional models. On the ISIC dermoscopy dataset, the proposed framework achieved a Dice score of 0.914 and a Precision of 0.922, significantly improving the accuracy of skin lesion recognition. The ablation study further revealed that local detail attention significantly enhanced boundary and texture modeling, global context attention strengthened long-range dependency capture, and cross-scale consistency constraints ensured the stability and coherence of prediction results. From a medical economics perspective, the proposed framework has the potential to reduce diagnostic costs and improve healthcare efficiency by enabling faster and more accurate image-based clinical decision-making. In summary, the hierarchical attention mechanism presented in this work not only provides an innovative breakthrough in mathematical modeling but also demonstrates outstanding performance and generalization ability in experiments, offering new perspectives and technical pathways for intelligent segmentation and recognition in medical imaging.

This study addresses the problem of pre-harvest fruit commodity grade prediction in intelligent orchards. To overcome the limitations of conventional post-harvest grading systems, including delayed quality evaluation, the limited representation capability of single-modality approaches, and the difficulty of modeling heterogeneous agricultural information collected throughout the growing season, this study proposes STF-Former, a spatiotemporal multimodal Transformer framework for pre-harvest fruit quality prediction. Rather than relying solely on static visual observations, the proposed framework formulates fruit quality prediction as a multimodal spatiotemporal learning problem by jointly exploiting phenotypic evolution, environmental dynamics, and field management information across different growth stages. Through stage-aware temporal modeling and cross-modal representation learning, STF-Former captures the dynamic interactions between fruit developmental processes and environmental water–fertilizer conditions, providing a unified prediction framework for both commodity grades and key quality indicators. This design improves the interpretability of quality formation and establishes a methodological framework for integrating heterogeneous agricultural data in precision orchard management. Experimental results demonstrate that the proposed method achieves significant advantages in both classification and regression tasks. In the commodity grade classification task, STF-Former achieves an Accuracy of $0.887$, a Precision of $0.875$, a Recall of $0.861$, a Macro-F1 score of $0.868$, and an AUC of $0.924$, substantially outperforming traditional machine-learning methods (Random Forest and XGBoost) as well as mainstream unimodal deep-learning models. In the regression task for key quality indicators, superior performance is consistently achieved across multiple agronomic metrics, where the mean absolute error (MAE) is $3.41$ mm for fruit diameter, $11.85$ g for single fruit weight, $0.057$ for coloration index, $0.81$ °Brix for soluble solid content, and $2.68$ N for firmness, with an overall $R^2$ reaching $0.846$. These results validate the effectiveness and robustness of multimodal spatiotemporal learning for accurate pre-harvest fruit quality prediction. The proposed framework provides a practical technical solution for intelligent orchard management, harvest planning, and data-driven precision agriculture, while offering a scalable paradigm for integrating multimodal sensing and temporal learning in smart agricultural systems.