Underwater visual object tracking is crucial for marine resource exploration and military security. However, due to the effect of insufficient light and turbid background in underwater scenes, efficient and accurate target tracking cannot be realized on underwater edge devices with limited computing resources. To address this problem, we design an underwater object tracking network, namely DBSF, for edge computing devices based on sparse confidence feature learning guided by differential boundary attention. Specifically, we propose a differential boundary attention distribution model to compute the object edge distribution state to enhance the accurate perception of the underwater object edge structure. Then, the differential boundary attention-guided object tracking network learns to perceive the highly discriminative sparse features on the object structure, and computes the object sparse confidence matrix, which reduces the constraints of the edge devices with limited computational resources and ensures the tracking performance. Extensive experiments demonstrate that the DBSF network achieves accurate underwater target recognition and outperforms related advanced methods.
Deep-learning technologies have shown impressive performance on many tasks in recent years. However, there are multiple serious security risks when using deep-learning technologies. For examples, state-of-the-art deep-learning technologies are vulnerable to adversarial examples that make the model’s predictions wrong due to some specific subtle perturbation, and these technologies can be abused for the tampering with and forgery of multimedia, i.e., deep forgery. In this paper, we propose a universal detection framework for adversarial examples and fake images. We observe some differences in the distribution of model outputs for normal and adversarial examples (fake images) and train the detector to learn the differences. We perform extensive experiments on the CIFAR10 and CIFAR100 datasets. Experimental results show that the proposed framework has good feasibility and effectiveness in detecting adversarial examples or fake images. Moreover, the proposed framework has good generalizability for the different datasets and model structures.