Application Of A Novel Deep Convolutional Neural Network In Campus Object Detection

As China's modern cloud computing,Internet of Things,big data and other emerging era of the Internet continue to deepen and accelerate,intelligent emerging industries will continue to step forward into the society,stepping into a new industrial era and development climax.Wisdom of the classroom building fully embodies the wisdom of science and technology in our country's higher education basis of modern information technology is increasingly mature,wisdom industry development pace of reform will be further applied in more fields,this is also makes "intelligent campus" construction of higher education in China in the field of science and technology a new research hotspot.At present,the traditional attendance mode has not yet met students' daily needs of punching in and attendance management.Therefore,the school urgently needs a more perfect and intelligent free classroom attendance management model,so as to effectively ensure the school's effect in teaching and attendance management.Attendance model is a very important part of the management system,and a sound attendance management system can guarantee the smooth progress of classroom teaching.Full use of the power of the Internet of Things technology,to ensure the correct and orderly implementation of the attendance model,accelerate the maturity and implementation of the future period of wisdom classroom is coming.The main work of this paper is as follows:(1)Related theories of convolutional neural network are introduced.Firstly,the convolutional neural network's convolutional layer,pooling layer,full connection layer,Dropout,supervised and unsupervised deep learning,as well as the common deep learning model and the application of convolutional neural network in other fields are briefly described.Secondly,activation function and loss function are analyzed.Finally,the gradient descent algorithm,forward propagation and back propagation required for neural network training are introduced.(2)Based on the three classical convolution structure and the analysis of the characteristics of the neural network model,an improved convolution neural network model structure,will be the first convolution Alex Net network layer split into three convolution kernels instead of,at the same time,the second use two convolution kernels instead of convolution layer,finally to delete a full connection layer,not only reduced the parameters,can also have more nonlinear transformation,increase the characteristics of network structure on the ability to learn.(3)Use public data sets(RMFRD and SMFRD)and private campus object data sets(OCD)through the Open CV face recognition algorithm simulation results,through the test can be found that the proposed network model framework compared with other advanced neural network model of lower loss value,at the same time in terms of recognition accuracy of the improved model can more effectively extract local features and detection of human face.