A Secure Scheme for Network Coding with Deep Learning in Industrial Internet of Things

This paper proposes a network coding scheme that utilizes neural network coding and steganography technique for Industrial Internet of Things (IIoT). This proposal gives an efficient error-correcting and secure network coding strategy simultaneously for general non-multicast network for the first time. The trivial concatenation of the above two schemes will not work, and we adopt a smart skill to fuse the neural network coding scheme and the steganography scheme seamlessly. The whole neural network model includes the network coding part and the steganography part. The first skill is making the network coding part converge in advance, and then we make the steganography part converged. The second skill is delicately designing the neural network structure and parameters to make the whole model work efficiently. The new proposal can correct more errors and confront more eavesdroppers than state of the art method in network coding. Besides, the transmission information rate is three, which is bigger than one, which is the most information rate of traditional channel code based on Shannon theory. We use a steganography neural network model to encode three images into one image and at last, the three images can be recovered simultaneously. The new scheme will give a practical error-correcting and secure network coding strategy, and then decrease the energy consume in the communication, which will benefit IIoT.