Scaling Vision Transformers (ViTs) has yielded remarkable advancements in diverse vision tasks, albeit at the cost of escalating computational, memory, and parameter demands. Existing efficiency techniques typically address only one dimension, computation, memory, or parameters, lacking a cohesive approach. In this paper, we introduce Vision-MoR, a novel ViT architecture that unifies parameter sharing, spatially adaptive computation, and memory-efficient design into a single framework. Vision-MoR employs a spatial-aware router with shifted-window attention to dynamically assign per-patch recursion depths, coupled with a recursive Transformer loop enabling token-wise early exiting. This facilitates content-adaptive processing and recursive parameter reuse while preserving spatial locality. On ImageNet-1K, Vision-MoR Small attains 74.6% Top-1 accuracy with 140M FLOPs and 5.7M parameters, outperforming EfficientViT-M2 (70.8%) and SHViT-S1 (72.8%) at superior throughput. The Vision-MoR X-Large variant achieves 80.4% Top-1 and 95.2% Top-5 accuracy using 14.3M parameters and 2044M FLOPs, surpassing ResNet-50 and EfficientNet-B1. On COCO object detection, Vision-MoR X-Large yields 39.1 AP with the lowest latency among comparable models. These results underscore Vision-MoR's state-of-the-art accuracy-efficiency trade-offs, positioning it as a scalable, deployment-friendly backbone for real-time vision applications.

We introduce DreamLand, an interactive 4D visualization framework that integrates WebGL with Supersplat rendering. It transforms static images and text into coherent 4D scenes through four core modules and employs a foveated rendering strategy for efficient, real-time multi-modal interaction. This framework enables adaptive, user-driven exploration of complex 4D environments.

Radiation source (RS) localization is crucial in electromagnetic environmental monitoring. Existing methods require prior knowledge of the environment and have overlooked the impact of multipath effects. Their positioning accuracy is penalized in practical applications. This paper presents an environment cognition-based variational Bayesian positioning method, which can achieve precise multi-source localization in uncooperative multipath environments using only the measurements of received signal strength (RSS). We begin by leveraging the signal propagation properties to design a model-data integrated unsupervised learning network, which partitions the positioning area according to the transmission states of different multipath signals. To estimate the number of RSs and mitigate the multipath effect, we identify the data samples of multipath RSS and divide them into groups, each corresponding to an RS. Then, a new variational Bayesian positioning algorithm is developed to operate without prior channel information and locate multiple RSs accurately across varying signal propagation models. Simulations show that our positioning method can effectively improve the accuracy of multi-RS localization by 10% in uncooperative multipath environments, compared to the state-of-the-art RSS-based localization techniques.

Specific emitter identification (SEI), known as radio frequency fingerprint (RFF) identification, is one of the key techniques to provide effective protection for the low-altitude security. However, most existing SEI methods cannot achieve satisfactory identification performance in low signal-to-noise ratio (SNR) environments. By fusing background information of the environment, this paper presents a new deep learning-based SEI method that can accurately identify emitters in the environments with severe noises. We first construct a dual convolutional neural network (DCNN) and a U-shaped Convolutional Network (UNet) to extract the RFFs and background information features, respectively. Then, a background-fingerprint attention fusion network (BFAFN) is designed to fuse the background information with RFF features. Using this network, we can obtain detailed emitters information through the fused signals, improving the identification accuracy. Experimental results show that our proposed SEI method outperforms other methods in performance on both open-source and collected unmanned aerial vehicles (UAVs) datasets, with an improvement in identification accuracy of 2% to 5%.