In recent years, with the rapid advancement of artificial intelligence, object detection algorithms have made significant strides in accuracy and computational efficiency. Notably, research and applications of Anchor-Free models have opened new avenues for real-time target detection in optical remote sensing images (ORSIs). However, in the realm of adversarial attacks, developing adversarial techniques tailored to Anchor-Free models remains challenging. Adversarial examples generated based on Anchor-Based models often exhibit poor transferability to these new model architectures. Furthermore, the growing diversity of Anchor-Free models poses additional hurdles to achieving robust transferability of adversarial attacks. This study presents an improved cross-conv-block feature fusion You Only Look Once (YOLO) architecture, meticulously engineered to facilitate the extraction of more comprehensive semantic features during the backpropagation process. To address the asymmetry between densely distributed objects in ORSIs and the corresponding detector outputs, a novel dense bounding box attack strategy is proposed. This approach leverages dense target bounding boxes loss in the calculation of adversarial loss functions. Furthermore, by integrating translation-invariant (TI) and momentum-iteration (MI) adversarial methodologies, the proposed framework significantly improves the transferability of adversarial attacks. Experimental results demonstrate that our method achieves superior adversarial attack performance, with adversarial transferability rates (ATR) of 67.53% on the NWPU VHR-10 dataset and 90.71% on the HRSC2016 dataset. Compared to ensemble adversarial attack and cascaded adversarial attack approaches, our method generates adversarial examples in an average of 0.64 s, representing an approximately 14.5% improvement in efficiency under equivalent conditions.

The information infrastructures of countries face serious threats from network attacks. Given that complex networks like the Internet are inherently intricate and multifaceted, it is crucial to research defense decision-making methods that are adapted to the structural nuances of such networks. Choosing the optimal defense timing to implement targeted measures is an effective way to enhance defense capability. However, most complex network defense methods adopt the fixed-period defense time strategy, ignoring network attack-defense behavior’s key characteristics of confrontation, dependence, and dynamic changes, which seriously weakening defense effectiveness. On the other hand, defense decision-making methods based on game theory generally use random network models to analyze real networks, which contradicts actual complex networks. To effectively fit the real complex network environment and enhance the accuracy of network security decision-making, we model and analyze attack-defense behaviors of complex networks, and propose a complex network defense timing decision method based on FlipIt game model. Firstly, on the basis of improving the propagation dynamics susceptible infected recovered (SIR) model, we analyze the evolution process of complex network security states. Secondly, our study construct an attack-defense FlipIt game model and design the attack and defense benefit function. Thirdly, we provide the calculation method of game equilibrium strategies and design a decision algorithm for the optimal defense time strategy of complex networks. Finally, targeting two typical complex network environments, small world networks and scale-free networks, we analyze the impact of different attack and defense time strategies on network node status and security defense effectiveness through simulation experiments. Compared with the classic defense time strategy, our method can effectively improve network defense performance by dynamically selecting and adjusting the optimal defense time strategy.