Scholarlink

<p>Nonlinear activation functions are central to the rich dynamics of Hopfield neural networks (HNNs), yet the impact of heterogeneous nonlinearities remains underexplored. This paper proposes a three-dimensional discrete heterogeneous Hopfield neural network (3D-DHNN) with sine, hyperbolic tangent and sigmoid-type activation functions. We show that this simple, sparsely coupled architecture generates coexisting multifold hyperchaotic attractors with high entropy and strong randomness, and we validate its behavior through FPGA implementation. Building on this chaotic source, we design a password-driven image encryption scheme that combines CML-Hilbert scrambling with a strictly invertible bidirectional plaintext-feedback diffusion, where all parameters are derived from a joint key/plaintext key-derivation function. Experiments and hardware tests on a Zynq-7020 device demonstrate near-ideal statistical security, negligible pixel correlation, perfect decryption, and a throughput of 76.8 Mbps, confirming the practicality of the proposed system for real-time secure image transmission.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Three-Dimensional Discrete Heterogeneous-Neuron HNN and Hardware Image Encryption

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Yi Yuan,
  • Dengwei Yan,
  • Zhengran Cao,
  • Qiang Jiang,
  • Lidan Wang,
  • Shukai Duan

摘要

Nonlinear activation functions are central to the rich dynamics of Hopfield neural networks (HNNs), yet the impact of heterogeneous nonlinearities remains underexplored. This paper proposes a three-dimensional discrete heterogeneous Hopfield neural network (3D-DHNN) with sine, hyperbolic tangent and sigmoid-type activation functions. We show that this simple, sparsely coupled architecture generates coexisting multifold hyperchaotic attractors with high entropy and strong randomness, and we validate its behavior through FPGA implementation. Building on this chaotic source, we design a password-driven image encryption scheme that combines CML-Hilbert scrambling with a strictly invertible bidirectional plaintext-feedback diffusion, where all parameters are derived from a joint key/plaintext key-derivation function. Experiments and hardware tests on a Zynq-7020 device demonstrate near-ideal statistical security, negligible pixel correlation, perfect decryption, and a throughput of 76.8 Mbps, confirming the practicality of the proposed system for real-time secure image transmission.