Most mobile vision systems adopt background subtraction or task offloading, but inference performance remains constrained by background dynamics. Dynamic deep neural networks (DNNs) scale computation without relying on historical information. However, a dynamic DNN tuned to a specific spatial granularity cannot excel across all images. For example, resolution level models perform well on images with sparse and large objects but poorly on images with dense and small objects. Changing system overheads make the choice of an optimal edge-assisted scheme for dynamic DNNs difficult. In this paper, we propose an efficient reinforcement-learning-based model adaptation and partition scheme for mobile vision empowered by dynamic DNNs, which jointly adapts the backbone network and the partition point of the inference model to accelerate inference without degrading inference accuracy. The scheme selects hyperparameters for dynamic DNNs, such as spatial granularity and complexity level, based on image characteristics and measured performance without relying on previous frames. A model partition algorithm is proposed to split the changing inference model in polynomial complexity using the Ford–Fulkerson algorithm. Experimental results based on object detection and image classification show that the proposed scheme reduces inference latency and energy consumption by over 71.2% and 31.8%, without degrading inference accuracy compared to the benchmarks.

Reinforcement learning (RL) based routing enables unmanned aerial vehicles (UAVs) to choose the next hop to transmit the packets, which has performance degradation against jamming and gray-hole attacks that disrupt communication and drop packets selectively, resulting in a substantial reduction in packet delivery ratio. In this paper, we propose an RL based jamming-resilient routing against gray-hole attacks in maritime UAV networks, which optimizes both the next hops to build multiple available routing paths with enhanced failure resilience and the transmit power to improve the communication reliability. Besides the channel gain and the hop count, based on the location of neighboring UAVs, the measured wind speed and jamming power, the service type and the moving speed, the multi-path routing is selected to enhance the packet delivery ratio with reduced latency and routing energy consumption following the quality of service constraints in the routing policy distribution. The UAV trust level is also exploited in the routing policy distribution formulation to enable rapid route repair against gray-hole attacks. The feedback recovery based on routing experience replay applies the future performance feedback from the destination to reconstruct the experience pool under fast changing maritime radio channel. Simulation results in offshore scenarios show the performance gain over the benchmarks.

Mobile device can deploy high-complexity vision tasks by reducing spatial redundancy in the inference of deep neural networks (DNNs) and offloading computations to remote servers via wireless channels. However, a trained DNN model with fixed network architecture faces challenges in dynamic scenarios with less overlap between previous and current frames. In this paper, we propose a RL-based edge-assisted mobile vision scheme based on dynamic DNN models to balance inference latency and capacity without relying on historical information for dynamic visual scenarios. This scheme optimizes the hyperparameter settings and complexity level of dynamic neural networks as well as the collaborative server and the partition points of the DAG structured inference model. Based on the image embeddings, channel gain, and previous inference performance, the inference policy is selected to optimize the utility function, which is a weighted sum of inference latency, local energy consumption, and inference accuracy. Experimental results demonstrate the performance improvements of our proposed scheme over the benchmarks.

Existing adaptive modulation and coding schemes suffer from performance degradation under time-varying underwater acoustic channels and jamming attacks, resulting in increased video jitter and energy consumption, as well as reduced peak signal-to-noise ratio (PSNR). In this paper, we first propose a reinforcement learning (RL)-based underwater video transmission scheme, which jointly optimizes the channel coding methods, subcarrier modulation order, quantization parameter, and transmit power against jamming. The direct sequence spread spectrum mechanism is used to reduce the jamming power in each subcarrier frequency band, which ensures that the transmitter obtains a reliable feedback message. In addition, a multi-frame statistical scheme is designed to support transmission policy selection, which mitigates the impact of channel instability by averaging the observed transmission performance and channel gain. To further enhance communication performance and robustness under large state-action spaces and dynamic channel conditions, we propose a deep RL (DRL)-based anti-jamming video transmission scheme to compress the excessive state-action space, mitigate quantization errors, and improve transmission stability. Two target networks are employed in DRL to ensure learning stability by mitigating fluctuations in Q-value and R-value updates caused by the time-varying channel. In addition, the performance bounds of transmission delay and energy consumption related to modulation order and channel coding rate are derived. Simulation and experimental results demonstrate that our schemes improve the video transmission performance by reducing transmission delay, energy consumption, and video jitter while increasing PSNR and video quality compared with the benchmarks.