Research On Vision Of Intelligent Car Based On Broad Learning System

Deep learning neural network has been restricted by many factors,such as the complex process of network updating,the long time of parameter adjustment and the difficulty of experiment analysis,which hinder its development to some extent.According to the theoretical basis of single hidden layer feedforward neural network,another dimension solution,width learning system,is proposed.Because of its simple structure,rapid network update and easy implementation,Broad Learning System can be applied to various intelligent fields.However,there are some problems in the system,such as redundancy,instability of neural nodes and adaptability of low dimension vectors.In the vision direction of intelligent robot,the target recognition system based on neural network is one of the key research contents.It has great research significance in intelligent driving,UAV fault detection,construction robot and other different working environments.Compared with the traditional visual detection algorithm,it can extract multi-dimensional recognition features in different environments,and further improve the detection accuracy and robustness of the algorithm.Therefore,in order to improve the detection accuracy,this paper studies Broad Learning System of intelligent vehicle in different target environments.Firstly,this paper obtains the target recognition image data to be trained and detected through the AGV mobile platform,grabs the recognition image of different angles and backgrounds,marks the image,increases the data set according to the data normalization and data enhancement processing steps,so as to avoid the data generalization phenomenon;secondly,the data is input into the shared convolution layer to extract the feature image and maintain the image The parameters of image height,width and channel number are invariable,and the new feature image is obtained by further extraction.Furthermore,the RPN pre-filtering algorithm based on hierarchical clustering is used to filter the objects in the candidate box in advance,so as to reduce the calculation and operation time of the system,so as to determine the region image corresponding to the feature image;then,the feature images of different sizes are input into ROI(region of interest Pooling)is used to keep the size of the image in the region of interest consistent and facilitate the processing oftarget recognition and detection.Finally,the normalized image is input into the classifier module to get the category of the target recognition image to be detected.The box regression module is used to process the normalized feature map and the non maximum algorithm is used Improve the candidate box,input the acquired coordinate data to the control module of the intelligent small test,achieve the coordinate conversion of the final coordinate of the object to realize the target recognition of the intelligent car.Through the simulation experiments of different groups,it can be seen that the target recognition system proposed in this design can not only detect the objects accurately,but also recognize the objects stably in different environments.