The Design And Implementation Of Billboard Recognition System Based On Deep Learning Neural Network

As a traditional and classical advertising method,billboards have the characteristics of facing wide range of viewers,convenience of delivery,high mobility of audiences and strongly aiming to specific crowds,thus have been an important part of advertising industry.How to keep the billboards,which is a classical way of advertising,live and productive has been drawing attention of advertisers and regulators as the wave of code-sweep payments grows.Based on convolutional neural network technology,and still with the billboard recognition demand of a certain enterprise as the research background,this thesis introduces the convolutional neural network to challenge the previous work of our group which uses traditional algorithm as the core of it.Firstly,this thesis studies the technology and basic steps of the image recognition system;secondly,it analyzes the specific requirements of billboard recognition on mobile platform,both with the requirements of adapting to complex applicating environment and real-time high accuracy;thirdly,it puts forward the above-mentioned research and analysis on the basis of the above,with the scheme of splitting and distributing the deep learning neural network model to client and server proposed.Fourthly,the scheme is designed in detail,with the real-time billboard recognition system based on C/S architecture realized.Finally,the related test of the system is carried out,and the test results show that the system avoids large-scale data upload is feasible.By reforming the split tiny-yolo model and convolution operation in a certain number of layers on the client side,the system achieves improvement in the efficiency of execution,and the requirement of real-time is achieved,compared with the traditional algorithm and the non-segmented neural network algorithm.It can also deal with complex environment of photos.