Research On Demodulation Algorithm Of OCC System Based On Deep Learning

The Intelligent Transportation System（ITS）is an efficient transportation management system that can alleviate traffic congestion and improve road safety.As one of the research hotspots of ITS,the V2V（Vehicle to Vehicle）communication can reduce traffic accidents by exchanging driving data information between vehicles,which plays an important role in ITS.The current V2 V system generally uses Radio Frequency（RF）communication technology.Although the technology has matured,there are still problems such as shortage of spectrum resources and susceptibility to electromagnetic radiation interference.Optical Camera Communication（OCC）has the advantages of wide spectrum range,no radio frequency interference and better communication security.The OCC system uses the Light Emitting Diode（LED）as the transmitter and the camera as the receiver,which can make full use of the light intensity,color and spatial domain of the LED to carry information.The OCC can make up for the defects of RF communication technology within a certain extent,and is expected to become a complementary scheme of RF communication technology.At present,some researchers have applied OCC to the V2 V scenes and achieved certain results.Although the OCC has achieved certain research results in the V2 V scenes,the existing demodulation algorithms still have the limitations of high bit error rate,poor anti-interference ability and slow speed of obtaining demodulated data,which are difficult to meet the requirements of V2 V for communication reliability and real-time communication.In view of the above problems,this thesis carries out the following research work:This thesis takes V2 V as the application scene,and proposes an OCC system demodulation algorithm based on deep learning.The algorithm consists of a multispectral fast recognition algorithm for LED state identification and an online deep extreme learning machine algorithm for time series prediction.Firstly,in view of the high bit error rate and poor anti-interference ability in the existing OCC system demodulation algorithm,the concept of n-ary image and the Multi-spectrum Fast Recognition algorithm（MFR）are proposed.This thesis converts the input image into the n-ary image to quickly detect the LED position,and uses the n-ary image as the input to the MFR algorithm to accurately identify the LED state.The n-ary image contains the binary image information of different thresholds in the LED area of the input image.The MFR algorithm takes n-ary image as input,which can construct the optimal state search space for each LED in the image.The algorithm pays more attention to the important features in the search space by combining the multispectral channel attention mechanism,which can effectively alleviate the influence of artifacts,light bursts,and blur on the LED recognition accuracy,and reduce the system bit error rate.Secondly,in view of the slow speed of obtaining the demodulated data in the existing OCC system demodulation algorithm,the Online Deep Extreme Learning Machine algorithm（OD-ELM）is proposed.The OD-ELM time series prediction algorithm is introduced into the OCC demodulation algorithm,which can effectively use the past frame data output by the MFR algorithm to obtain the current frame data in advance.The OD-ELM algorithm improves the network layer number and random initialization method of the online sequential extreme learning machine algorithm,which improves the accuracy of time series prediction.The algorithm retains the dynamic adjustment ability of the online sequential extreme learning machine.It can learn a certain experience from offline data in the way of offline learning,and continuously update the algorithm parameters in the way of online learning.Finally,the proposed algorithms are verified on the OCC system platform.The experimental results show that: The OCC system demodulation algorithm composed of the MFR algorithm and the OD-ELM algorithm reduces the system bit error rate,enhances the system robustness,and improves the speed of obtaining demodulated data.The research work in this thesis has important practical significance for promoting the maturity of OCC technology.