Vision-guided autonomous underwater vehicles based on remote sensing play an important role in ocean missions. However, some problems exist in underwater visual perception, such as color distortion, low contrast, and fuzzy details, which restrict the applications of underwater visual tasks. Most of the state-of-the-art image enhancement methods are still limited in scene adaptability, recovery accuracy, and real-time processing. To solve these problems, we propose an underwater sensing scene image enhancement method called a multiscale feature fusion network (MFFN). To extract the multiscale feature, the measure merging the feature extraction module, the feature fusion module, and the attention reconstruction module is designed. This measure can also enhance the adaptability and visual effect of the scene. Moreover, we propose multiple objective functions for supervised training to match the nonlinear mapping. Based on the qualitative and quantitative evaluations, the proposed method produces competitive performance compared with some state-of-the-art methods, and the perception and statistical quality of underwater images are enhanced effectively.

Low contrast, color distortion, and blurred details are common problems that perplex vision-guided underwater robots. To this end, we propose an underwater image enhancement framework via self-attention and contrastive learning (UIESC) to solve these problems. In this article, local features and global dependencies are constructed through space and channel dual attention, and criss-cross attention is used to solve the high computational complexity of self-attention. Moreover, contrastive learning is introduced into network training as a loss function, and contrastive regularization ensures that the enhanced images are closer to clear positive samples and away from source negative samples. Finally, smoothed-histogram equalization is adopted for further optimization to accommodate complex and variable underwater scenes. Extensive experiments have shown that our framework outperforms state-of-the-art methods in underwater image enhancement tasks.

Vehicular edge computing (VEC) is an efficient solution to alleviate the limitations of local computing resources in vehicular networks. However, the high mobility of vehicles and the dynamic variability of network topologies make it significantly challengeable. In this work, we make a fusion of model-driven and data-driven intelligence to design a multi-agent deep reinforcement learning (DRL) solution for task offloading in urban VEC networks. First, computational models for task queue, transmission, computation, energy consumption, and expense are meticulously developed for the VEC network that integrates communication and computation. Vehicular tasks vary in type, urgency, size, and timeframe, leading to different latency requirements. Tasks may be executed locally within the vehicle, at a server after V2I offloading via cellular communications, or in a neighboring vehicle after V2V offloading via millimeter-wave (mmWave) communications. Each of these options incurs different levels of latency, energy consumption, and expense. Second, based on these models and the utility function that combines latency, energy consumption, and expense, an optimization problem for task offloading is formulated. This problem can be interpreted as a Markov decision process with a carefully designed reward function. Third, to address the offloading problem, we propose a multi-agent proximal policy optimization-based task and target selection algorithm (MAPPO-TTSA). This algorithm also utilizes convolutional neural networks to extract features from large-scale states, thereby enhancing their correlation. Fourth, comprehensive training is performed on the observational data to determine the optimal parameters for predicting task offloading. Finally, extensive experiments are conducted, and simulation results are provided to demonstrate that the proposed intelligent task offloading scheme offers significant advantages in terms of average task completion delay and utility level across various scenarios.

Hybrid evolutionary approaches have gained significant attention for solving complex optimization problems, but their potential for optimizing the low-dimensional latent space of generative adversarial networks (GANs) remains underexplored. This paper proposes a novel improved crisscross optimization (ICSO) algorithm, a hybrid evolutionary approach that integrates crisscross optimization and perturbation mechanisms to find the suitable latent vector. The ICSO algorithm treats the quality and diversity as separate objectives, balancing them through a normalization strategy, while a gradient regularization term (i.e., GP) is introduced into the discriminator’s objective function to stabilize training and mitigate gradient-related issues. By combining the global and local search capabilities of particle swarm optimization (PSO) with the rapid convergence of crisscross optimization, ICSO efficiently explores and exploits the latent space. The extensive experiments demonstrate that ICSO outperforms state-of-the-art algorithms in optimizing the latent space of various classical GANs across multiple datasets. Furthermore, the practical applicability of ICSO is validated through its integration with StyleGAN3 for generating unmanned aerial vehicle (UAV) images, showcasing its effectiveness in real-world engineering applications. This work not only advances the field of GAN optimization but also provides a robust framework for applying hybrid evolutionary algorithms to complex generative modeling tasks.