Visible light MIMO communication system color detection method based on auto-associative neural network

Abstract

The invention discloses a visible light MIMO communication system color detection method based on an auto-associative neural network. The method comprises the following steps of: collecting an LED image data set of a target region, and dividing the LED image data set into a training set and a prediction set; adjusting the sizes of LED images based on a bicubic interpolation method, and adjusting the resolutions of all the LED images to the same size; adding a bottleneck layer in the middle of the auto-associative neural network, and combining the auto-associative neural network with principalcomponent analysis (PCA) to construct an auto-associative neural network for extracting principal components; decomposing the mapping layer, bottleneck layer and inverse mapping layer of the auto-associative neural network into a plurality of sub-networks which are connected in parallel, completing the construction of a parallel auto-associative neural network PANN, and achieving the nonlinear principal component analysis of the parallel auto-associative neural network; searching and determining the coefficient of the neural network, inputting the training set of the LED images into the PANN obtained in the step S4, and performing updating iterative training on the PANN by using a back propagation algorithm; and inputting an LED image test set into the PAANN obtained in the step S5 to complete color detection.

Description

technical field [0001] The present invention relates to the technical field of color detection of optical communication systems, and more specifically, to a color detection method of a visible light MIMO communication system based on an auto-associative neural network. Background technique [0002] In a system that uses a camera to collect visible light signals, the signal is generally received and restored according to the light intensity. First, each LED is modulated independently, and a sequence of images of all LEDs is captured by a camera, with each light source at a different position in the image. Then, the brightness of the detected image is calculated, and the signal flow of each light source is recovered using image processing technology. The existing multiple-input multiple-output (MIMO) visible light wireless communication system with multiple light sources successfully realizes the camera's collection of visible light signals, but there are still some shortcomi...

AI Technical Summary

Problems solved by technology

[0007] 1. Due to the interference between multiple light sources and the interference of ambient light, the light source color detection method based on machine learning can only achieve the effect of a weak learning machine, so it cannot be applied to the actual VLC demodulation program

[0008] 2. In the field of machine learning, the Adaboost algorithm is a common method to upgrade a weak learning machine to a strong learning machine. However, due to the weak nonlinear mapping ability of the weak learning machine itself, and the high training complexity of multiple weak learning machines, Therefore, the color detection accuracy and time efficiency of the multi-light source visible light communication system cannot be guaranteed

[0009] 3. The deep neural network has a strong nonlinear mapping ability, but there are problems such as difficulty in determining hyperparameters and long training time

Images