The Ocean Internet of Things (OIoT) has promoted the development of the ocean devices, which generate a large amount of underwater acoustic data. The transmission of underwater acoustic data requires a large amount of spectrum resources. Aiming at the problems of improving the utilization rate and security of spectrum resources, an underwater acoustic spectrum sensing algorithm based on personalized federated learning (PFL) and RepViG is proposed. A security protection framework is established for underwater acoustic data based on PFL. The common features of each underwater acoustic data are extracted through the meta-model. Each client only needs to fine-tune the parameters of the meta-model based on the local underwater acoustic data to achieve a personalized model. Based on the improved RepViT and the improved ViG, a dual-branch sensing model of RepViG is designed. In the improved RepViT, we employ the Haar wavelet downsampling (HWD) module to retain the low-frequency and high-frequency detail features of the underwater acoustic signal through multi-resolution. And we employ Interactive Convolution Block (ICB) to capture the relationship between local features and global features through multi-scale dynamic convolution kernels. In the improved ViG, we adopt the lightweight Star-Blcok module to reduce feature redundancy and enhance sensing efficiency. Compared with other algorithms, the simulation results show that the detection probability is increased by 7.6%, and the false alarm probability is reduced by 7.5%.

Sketch-based face recognition is an interesting task in vision and multimedia research, yet it is quite challenging due to the great difference between face photos and sketches. In this paper, we propose a novel approach for photo-sketch generation, aiming to automatically transform face photos into detail-preserving personal sketches. Unlike the traditional models synthesizing sketches based on a dictionary of exemplars, we develop a fully convolutional network to learn the end-to-end photo-sketch mapping. Our approach takes whole face photos as inputs and directly generates the corresponding sketch images with efficient inference and learning, in which the architecture is stacked by only convolutional kernels of very small sizes. To well capture the person identity during the photo-sketch transformation, we define our optimization objective in the form of joint generative discriminative minimization. In particular, a discriminative regularization term is incorporated into the photo-sketch generation, enhancing the discriminability of the generated person sketches against other individuals. Extensive experiments on several standard benchmarks suggest that our approach outperforms other state-of-the-arts in both photo sketch generation and face sketch verification.