Object Detection And Recognition Based On End To End Convolutional Neural Network

Since 2012,deep learning models have been successfully applied to many areas such as speech recognition,classification and object detection of the images,handwriting recognition,and neural language processing and video tracking and achieve promising results.The business of Deep Learning brings much convenience to people.Nowadays,the structure of deep learning becoming more and more complex while the amount of data used to turn them is becoming large and large,under this circumstance,structing the efficient and robust convolutional neural network got more and more attention form the scholar and the companies of technology.Firstly,we present the function of the basic unit in deep classification architecture.In order to avoid the risk of the overfitting for the large amount of calculate by the complex convolutional neural network.We present the four design patterns for the network for the classification.Appropriate learning strategy,appropriate networks' path,multi-branch for network and multi-cluster for convolutional neural network.Secondly,in this part,we use the two full connected layers as the backbone on which we adding the basic unit respectively.We discuss the function of the basic unit used in the highly modularized network architecture for the task of image classification.We could use the simple basic unit with more complex structures to abstract the data with the receptive field.We instantiate the basic unit architecture with a multi-layer feature perception and that is a potent functional approximation.Lastly,following the design of the FPN,we further add an extra 1*1 convolution layer to benefit feature extraction,via the batch normalization.In addition,the designed network architecture for feature extraction combines low-resolution and high-resolution feature layers to predict the category of the object in images.The proposed architecture shows competitive results compared to some state-of-the-art algorithms both in accuracy and in speed on some datasets.