Remote sensing imagery presents vast, inherently unstructured spatial data, necessitating sophisticated reasoning to interpret complex user intents and contextual relationships beyond simple recognition tasks. In this paper, we aim to construct an Earth observation workflow to handle complex queries by reasoning about spatial context and user intent. As a reasoning workflow, it should autonomously explore and construct its own inference paths, rather than being confined to predefined ground‑truth sequences. Ideally, its architecture ought to be unified yet generalized, possessing capabilities to perform diverse reasoning tasks through one model without requiring additional fine-tuning. Existing remote sensing approaches rely on supervised fine-tuning paradigms and task‑specific heads, limiting both autonomous reasoning and unified generalization. To this end, we propose RemoteReasoner, a unified workflow for geospatial reasoning. The design of RemoteReasoner integrates a multi-modal large language model (MLLM) for interpreting user instructions and localizing targets, together with task transformation strategies that enable multi-granularity tasks, including object-, region-, and pixel-level. In contrast to existing methods, our framework is trained with reinforcement learning (RL) to endow the MLLM sufficient reasoning autonomy. At the inference stage, our transformation strategies enable diverse task output formats without requiring task-specific decoders or further fine-tuning. Experiments demonstrated that RemoteReasoner achieves state-of-the-art performance across multi-granularity reasoning tasks. Furthermore, it retains the MLLM's inherent generalization capability, demonstrating robust performance on unseen tasks and categories.

Reconfigurable intelligent surfaces (RIS) have received significant attention in the field of communication and wireless networks, due to their numerous advantages, such as low cost, low energy consumption, miniaturization, programmability, and ease of deployment. This paper proposes an RIS based on liquid crystal (LC) for constructing wireless communication systems. Tunable LC embedded between metal microstructures supported by quartz substrates allows for precise control of the reflected wave phase. The experimental results demonstrate that RIS has more than 3600 of phase modulation capability and four stable digital coding states at the optimal operating frequency of 100.2 GHz. In addition, the RIS has a 100∘ beam scanning capability, providing a solid foundation for tracking and targeting during communications. The communication capability of RIS was then demonstrated in the sub-terahertz band by constructing a 16QAM constellation diagram test system and a video real-time transmission system with varying deflection angles. Our work provides excellent solutions for communication and wireless network applications in the millimeter wave and terahertz (THz) frequency bands.

Disparity refinement is an essential step of local stereo matching methods to produce fine dense disparity maps. The inherent defect of local stereo methods results in erroneous disparity in occluded areas. In this paper, we present a novel post processing method which can effectively improve the accuracy of dense disparity maps by rectifying disparity errors iteratively. Invalid disparities are first detected by left-right consistency check and color-disparity consistency check. For each invalid pixel, supports from valid pixels in the neighborhood are collected to determine the plane parameters of the local window. An iterative strategy is adopted to gradually propagate disparity information from valid pixels to invalid areas. We apply the proposed method to disparity maps produced by two recent stereo matching methods, and compare the refining results with other post processing methods. Experimental results show the effectiveness of our method in improving dense disparity maps.

Image-based road vanishing point (VP) detection is a challenging problem as the algorithm needs to deal with different road type with varying imaging conditions in real time. We propose a hierarchical VP detection method with the ability to choose efficient algorithms for different road types. Firstly, road scenes are classified into 4 types based on holistic road features abstracted from responses of Gabor filters, which also coarsely locate the road. Then, the distinct edge or line-like texture of each road type is adaptively extracted by a new road-line-segment detection method. Finally, VP is estimated by road-type-specific voting and verification. Experiments on 1000 general road images show that, the proposed method can detect VP under complex road scenes, achieving improvements for both accuracy and efficiency, compared with the state of the art methods relying on single features.

Image-based road scene understanding is a critical issue for intelligent vehicles and autonomous mobile robots. It is challenging to deal with varying road conditions in a dynamic environment in real time. This paper presents an effective while simple approach to classify road types and locate the road-related elements through the analysis of a holistic visual road feature. The feature is abstracted from responses of Gabor-filter-set and grouped into super-pixel grids, consisting of road-scene textural context and dominant orientation distribution. From this feature we successively deduce the information of horizon line, road type and coarse locations of road surface, lanes, on-road obstacles and off-road regions. Experiments show that the proposed analyzing method based on the new holistic feature is beneficial to road estimation and vehicle detection in complex road scenes, achieving improvements in both accuracy and efficiency.

Nilsimsa algorithm plays an important role in the digest-based anti-spam technique. Based on the research about Nilsimsa digest, we propose an approach to publish and lookup for Nilsimsa digests inDHT: DHTnil. In this paper, we discuss DHTnil's publication and lookup algorithm in detail, and evaluate the effectiveness of DHTnil through simulation experiments. For random generated digests, DHTnil can guarantee that the publication of a group of similar Nilsimsa digests involves no more than 3% of all subspaces, and the similar lookup of a digest involves no more than 3.5% of all subspaces to guarantee a recall rate of 80%. For real junk email samples collected from Internet, DHTnil has an even better performance.