Low Earth Orbit (LEO) satellite communication constellations provide crucial support for the future of ubiquitous connectivity. However, the high dynamic nature of LEO satellites’ relative positions affects the stability and reliability of inter-satellite communication links while imposing stricter latency requirements. Fast adaptive beamforming technology offers a viable solution to address these challenges. To achieve fast adaptive beamforming, this paper proposes a two-step beamforming solution based on a Transformer neural network model and explores the feasibility of applying meta-heuristic algorithms for hyperparameter optimization in neural network (NN) models. Experimental results demonstrate that the angle-of-arrival predictor optimized using the Polar Lights Optimizer (PLO), a meta-heuristic algorithm inspired by the aurora phenomenon, significantly outperforms the unoptimized model, with its error consistently remaining within the half-power beamwidth of the proposed planar antenna array. Meanwhile, the beamforming accuracy improves by 15.2% compared to other NN-based models, and in all test scenarios, our algorithm reduced the average response time by approximately 82.1% compared to the null-steering beamforming algorithms. This study provides a novel solution for achieving low-latency, high-reliability 3D beamforming in LEO inter-satellite communication, integrating both speed and accuracy, thereby contributing to the advancement of 6G and future ubiquitous connectivity.

The frequency-offset-governed range-angle-dependent dynamic beam pattern of frequency diverse array (FDA) constitutes a critical mechanism for enhancing spatial-domain covertness in wireless communication systems. This paper investigates FDA-assisted satellite covert communication. By considering covertness constraints based on average Kullback-Leibler (KL) divergence, frequency offset limitations, and power budget, we jointly optimize the satellite transmit power and FDA element frequency offsets to simultaneously maximize the time-varying covert rate and minimize the time-varying covert angle. To address practical channel dynamics, we develop a stochastic channel model incorporating Doppler compensation and channel aging effect mitigation for the FDA scheme. Closed-form and approximate expressions for the beam peak angle and half-power beamwidth (HPBW) with respect to frequency offsets are derived and analytically verified. The principle of linear uniform FDA frequency offset interval is established, with covertness quantified through KL divergence, enabling an optimized FDA frequency offset design from an angular perspective. To solve it, an improved deep reinforcement learning (DRL) with soft actor-critic (SAC) algorithm is proposed. Simulation results demonstrate that the interception probability is improved, and the proposed dynamic FDA algorithm exhibits better covert performance than other DRL-based methods.

Text-video retrieval establishes accurate similarity relationships between text and video through feature enhancement and granularity alignment. However, relying solely on similarity to associate intra-pair features and distinguish inter-pair features is insufficient, \textit{e.g.}, when querying a multi-scene video with sparse text or selecting the most relevant video from many similar candidates. In this paper, we propose a novel Dual Uncertainty Quantification (DUQ) model that separately handles uncertainties in intra-pair interaction and inter-pair exclusion. Specifically, to enhance intra-pair interaction, we propose an intra-pair similarity uncertainty module to provide similarity-based trustworthy predictions and explicitly model this uncertainty. To increase inter-pair exclusion, we propose an inter-pair distance uncertainty module to construct a distance-based diversity probability embeding, thereby widening the gap between similar features. The two components work synergistically, jointly improving the calculation of similarity between features. We evaluate our model on six benchmark datasets: MSRVTT (51.2%), DiDeMo, MSVD, LSMDC, Charades, and VATEX, achieving state-of-the-art retrieval performance.

Aquatic species play crucial roles in global ecosystems but are increasingly threatened by factors such as overfishing, coastal development and climate change . Existing deep learning methods address these challenges by employing powerful networks and large-scale, diverse datasets, separately tackling species recognition and trait identification during ongoing monitoring. However, they often exhibit limited generalization ability. Inspired by the human ability to quickly identify fish species and their locations with just a glance at an underwater image or scene, we introduce FishDetectLLM—a framework built on the lightweight TinyLLaVA architecture. FishDetectLLM utilizes the powerful reasoning capabilities and vast world knowledge of large language models (LLMs) to address the fish detection problem, providing both fish classification results and predicted bounding boxes for fish. Specifically, we create instruction dialogues for fish detection that connect fish taxonomy with classification descriptions and map location descriptions to the corresponding coordinates of bounding box in the input images from the recently released large-scale FishNet dataset. Then, we pretrain and fine-tune FishDetectLLM to achieve fish detection using the created dataset, leveraging the principle of augmenting human knowledge. Our results show that FishDetectLLM significantly outperforms existing multimodal LLMs and task-specific methods. Unlike conventional detection architectures that struggle to generalize beyond the training data, FishDetectLLM exhibits strong generalization capabilities, achieving robust performance on unseen data. This innovation paves the way for future applications of MLLMs in full research and offers valuable tools for the conservation of fish biodiversity.