Akademik Veri Yönetim Sistemi
EUROPEAN PHYSICAL JOURNAL PLUS, cilt.141, sa.3, 2026 (SCI-Expanded, Scopus)
With the extensive application of images in scenarios such as communication, surveillance, and intelligent recognition, the problem of their secure transmission in complex network environments is rendered increasingly prominent. Existing chaotic image encryption methods are primarily dependent on pixel-level perturbation, and the lack of adaptive processing capabilities for target semantic regions results in deficiencies in both overall security and the protection of key regions. A semantic-aware encryption scheme integrating YOLOv13 with chaotic systems is proposed in this paper. The bounding boxes output by YOLOv13 are combined with chaotic systems, and the rapid object detection capability of YOLOv13 is leveraged to generate plaintext-dependent chaotic sequences. Furthermore, parameter-dynamic Arnold scrambling is adopted to effectively rearrange image pixels spatially. Finally, a dual-diffusion approach combining modular arithmetic and cyclic shifting is employed. Through cipher feedback and bit-level shifting, the global diffusion of minute pixel variations is realized by this scheme. Simulation results demonstrate that the proposed scheme is capable of effectively performing image encryption and decryption. These results indicate that the proposed scheme achieves superior security performance, providing a useful reference for integrating deep learning with chaotic systems in image encryption.