| With the advent of the era of large data and the era of artificial intelligence,Deep learning has gradually entered people’s horizons.At present,deep learning has achieved great sucess in the field of computer vision and speech recognition.It adopts multi-level network structure,It can better extract features,and almost no artificial intervention.it breaks the traditional way of relying on manual extraction of features.And the effect of the algorithm is greatly improved.Convolution neural network is one of the many network structures in deep learning.It can read two-dimensional images directly.And,It’s structure is more similar to the human visual nerve system.So the effect of the algorithm is far beyond the traditional algorithm.Convolution neural network is widely used in image recognition and target detection.Among them,face recognition,pedestrian detection and face feature detection are the most widely used tasks.The main work of this article includes the following three aspects:Firstly,Introducing the calculation method of single neuron、multilayer perceptron and the network structure and algorithm principle of BP neural network.Taking the classic handwritten digital network as an example,the basic structure and algorithm principle of convolution neural network are introduced in detail.Secondly,A DeepID face recognition algorithm based on convolution neural network is studied.Using DeepID algorithm for reference,we use multiple parallel convolutional neural networks,It effectively improves the expression ability of facial features extracted by the algorithm.A residual network with better performance than convolution neural network is introduced as the feature extraction network of the model.Thirdly,Research on SSD target detection algorithm.Modified the paving strategy of default Box and use a lower level feature map to detect smaller targets.Effectively improving the detection rate of pedestrian detection. |
