In this letter, we investigate the content caching problem within large-scale integrated satellite-terrestrial networks, focusing on a fusion scenario of future large-scale remote sensing constellations and communication satellite networks. Our investigation relies on deep reinforcement learning techniques aimed at minimizing the long-term average content delivery delay. To address the inherent challenge of convergence in single-agent algorithms, we propose clustering intelligent remote sensing satellites, with each cluster headed by an intelligent agent. Based on the characteristics of the model, we modify the multi-agent proximal policy optimization (MAPPO) algorithm by integrating long short-term memory (LSTM) to capture the correlation of the state information of different agents in the time domain. Simulation results show that the proposed LSTM-MAPPO outperforms the benchmarks, exhibiting faster convergence speed and lower standard deviation.

This paper seeks to address the limitations of conventional remote sensing data dissemination algorithms, particularly their inability to model fine-grained multi-modal heterogeneous feature correlations and adapt to dynamic network topologies under resource constraints. This paper proposes multi-modal-MAPPO, a novel multi-modal deep reinforcement learning (MDRL) framework designed for a proactive data push in large-scale integrated satellite–terrestrial networks (ISTNs). By integrating satellite cache states, user cache states, and multi-modal data attributes (including imagery, metadata, and temporal request patterns) into a unified Markov decision process (MDP), our approach pioneers the application of the multi-actor-attention-critic with parameter sharing (MAPPO) algorithm to ISTNs push tasks. Central to this framework is a dual-branch actor network architecture that dynamically fuses heterogeneous modalities: a lightweight MobileNet-v3-small backbone extracts semantic features from remote sensing imagery, while parallel branches—a multi-layer perceptron (MLP) for static attributes (e.g., payload specifications, geolocation tags) and a long short-term memory (LSTM) network for temporal user cache patterns—jointly model contextual and historical dependencies. A dynamically weighted attention mechanism further adapts modality-specific contributions to enhance cross-modal correlation modeling in complex, time-varying scenarios. To mitigate the curse of dimensionality in high-dimensional action spaces, we introduce a multi-dimensional discretization strategy that decomposes decisions into hierarchical sub-policies, balancing computational efficiency and decision granularity. Comprehensive experiments against state-of-the-art baselines (MAPPO, MAAC) demonstrate that multi-modal-MAPPO reduces the average content delivery latency by 53.55% and 29.55%, respectively, while improving push hit rates by 0.1718 and 0.4248. These results establish the framework as a scalable and adaptive solution for real-time intelligent data services in next-generation ISTNs, addressing critical challenges in resource-constrained, dynamic satellite–terrestrial environments.

Fine-grained aircraft classification in optical remote sensing images relies on subtle difference of local parts. However, aircrafts in low-resolution (LR) images are depicted as small objects with blurred details and limited appearance information, resulting in the dramatic classification performance drop. We propose a novel dynamic hybrid knowledge distillation (DHKD) method to enhance fine-grained aircraft classification in LR remote sensing images. The core idea is to construct a teacher network for high-resolution (HR) images and a student network for LR images, and dynamically distill the key discriminative region attention-guided feature-based and logit-based object prior knowledge acquired from the teacher to the student to instruct its training, thereby improving the student’s fine-grained representation capability and classification performance for LR images. The distillation weights of different knowledge types are adaptive to their optimization direction similarity with that of the primary fine-grained classification task during the iterations. Extensive experiments on the MAR20 dataset demonstrate that DHKD consistently outperforms state-of-the-art distillation methods on fine-grained classification for aircrafts with absolute sizes of 32, 20, and 12 pixels in LR remote sensing images.