Vehicle Type Recognition Based On Multi-Scale Convolution Neural Network

Convolution neural network for image recognition classification is an important application of image processing in depth learning.One of the more significant advantages of convolution neural network is that it can use the image for convolution operation and extract the image from the pixels of the image.And the weight sharing and pooling of the convolution neural network greatly reduces the parameters that the network needs to be trained,simplifies the network model and improves the training efficiency of the network.However,the traditional convolution neural network can only learn the training of one of the gray-scale images and the color image channel,and can only extract the local features of the image so that some important information of the image is inevitably lost,but the image Local features and global features also play an important role in the classification and recognition of images,and depth learning is an important part of the development of artificial intelligence,and depth learning is obtained in terms of speech recognition,image classification and recognition,and natural language processing.Great achievements,with the traditional convolution neural network is not easy to solve the problem has also made a great breakthrough,including the retrieval of commodity images,handwriting recognition and license plate automatic identification.At present,with the research and optimization of the depth learning algorithm,the traditional deep learning programming algorithm has been far from meeting the needs of the programming staff,because the traditional basic algorithm implementation requires the researchers to spend a lot of time and energy.At the same time,some researchers have begun to pursue fast and efficient depth learning algorithms,which are based on a variety of deep learning frameworks in the Caffe depth learning framework,which not only provide researchers with a fast and efficient development model,but also provides a number of convolution neural network development model to enable researchers in the more advanced and perfect model to improve and research.Based on the deep study of the AlexNet model in the Caffe framework,this paper proposes a model based on multi-scale convolutional neural network(MSCNN).MSCNN model first samples the original image to obtain multiple images,and as a training sample,the training samples for a number of path training and learning,each channel and its corresponding filter convolution operation,the characteristics of the image.And then the feature dimension of each path is merged through a fully connected layer.The final feature is used for image recognition classification,and the resulting final feature is input as input data into classifier to complete the recognition of object.Experiments show that the MSCNN structure model proposed in this paper has a higher recognition rate for image recognition.In this paper,we study the network structure and the optimization of network parameters by using different data sets.By analyzing and summarizing the rules of depth learning for image recognition classification,this paper adjusts the network structure Parameters of good or bad have a direct impact on its performance,so to solve practical problems have a good role in guiding.