To address the multiple challenges in existing remote sensing images detection methods, including insufficient localization accuracy, imprecise category recognition, and high false positive and false negative rates, this paper proposes RS-YOLOv11 (Remote Sensing-YOLOv11), an improved object detection algorithm specifically designed for remote sensing applications based on the YOLOv11 framework.This study introduces the fine-grained SPD-Conv module to optimize backbone network downsampling, effectively preserving feature information and enhancing small object detection. The detection head employs Dynamic Head architecture with integrated multi-dimensional attention mechanisms, significantly improving model performance.To reduce network complexity, Faster_Block replaces the Bottleneck design, decreasing C3K2 module computational cost and addressing YOLOv11 deployment challenges. This improvement achieves lightweight design while maintaining performance and balancing Dynamic Head overhead. Additionally, WIoU loss function replaces CIoU to suppress gradient issues from low-quality images.Experiments on the VisDrone2021 dataset demonstrate that our improved model achieves a 3.9% increase in mAP50 compared to the YOLOv11n baseline, while maintaining comparable computational complexity and parameter efficiency.

Autism Spectrum Disorder (ASD) is a developmental disorder whose incidence has been increasing, significantly impacting patients' daily lives. Recent research trends focus on leveraging large-scale, multi-center neuroimaging datasets to enhance clinical applicability and statistical validity. However, the lack of reliable biomarkers and data heterogeneity across datasets limit classification effectiveness. This paper proposes a resting-state functional MRI (rs-fMRI) based approach to improve classification accuracy for ASD by integrating multi-site data.

In this study, a MultiModal Deep Attention (MMDA) model is introduced for ASD identification, which effectively combines rs-fMRI features with demographic characteristics through three modules: feature extraction, feature learning, and deep perception. The feature extraction module uses autoencoders to clean rs-fMRI time series data; the feature learning module employs multi-head attention mechanisms and convolutional neural networks to uncover intrinsic data structures; and the deep perception module integrates multimodal features to produce final ASD classifications. Simulation results demonstrate that the MMDA model outperforms benchmark algorithms in ASD identification.

In summary, this research advances ASD diagnostic accuracy by integrating multimodal data with neural networks, offering a promising tool for objective auxiliary diagnosis and providing new insights into ASD pathomechanisms.