关键词: 关键词: 深度压缩感知, 注意力机制, 多尺度, 多腔体混沌系统, 混沌加密

Abstract: To address the resource demands and privacy leakage risks associated with the transmission and storage of a large number of redundant images in the Internet of Things (IoT), a deep compressed sensing network and its chaotic encryption protection method for IoT images are proposed. First, an improved deep compressed sensing network is introduced for high-quality image compression and reconstruction. Specifically, this network enhances residual blocks in traditional deep reconstruction networks using a channel attention mechanism and integrates multi-scale branches and fusion modules through parallel fusion design. These enhancements significantly improve reconstruction performance compared to conventional deep reconstruction networks based on convolutional residual stacking layers. Experimental results demonstrate that the proposed network achieves an average improvement of 0.606 dB in Peak Signal-to-Noise Ratio (PSNR) and an average increase of 0.0111 in Structural Similarity Index Measure (SSIM) over the classical deep compressed sensing method CSNet+. Subsequently, a novel multi-cavity chaotic system is proposed, employing spherical coordinate transformations and two sets of parity-controlled step functions to facilitate cavity expansions in arbitrary single or double directions along the x, y, and z axes. This system exhibits strong chaotic characteristics and randomness, making it highly effective for image encryption. Finally, integrating the proposed deep compressed sensing network and multi-cavity chaotic system, an encryption and decryption scheme based on chaotic index scrambling and diffusion for deep compressed sensing measurements is designed, accompanied by detailed security analyses, thus ensuring secure image transmission.

Key words: Keywords: deep compressed sensing, attention mechanism, multi-scale, multi-cavity chaotic system, chaotic encryption