Research Of Indoor Scene Recognition Method Based On Convolutional Neural Network

Scene recognition is an important subject of computer vision,and it is a major part of mobile robot system.In recent years,the convolutional neural network has made a lot of achievements in the field of image understanding and recognition.Therefore,an indoor scene recognition method based on convolutional neural network is proposed in this paper.Feature extraction which is based on convolutional neural network is combined with classifier to recognize the indoor scene.The main contents are as follows:Firstly,the existing methods of scene recognition are summarized by looking up a large number of related documents.And these methods can be divided into two types,one is based on semantic feature and the other is based on convolutional neural network.Because of the complexity and randomness of indoor scene,a convolutional neural network model is proposed to solve the problem.Secondly,a method of indoor scene recognition is proposed.And a new structure named Inception-residual is designed based on Inception structure and residual structure.Consequently,an indoor scene recognition model which is based on GoogLeNet and Inception-residual structure is proposed.What's more,dropout and Batch Normalization are introduced to improve performance of the model.The indoor scene recognition process based on convolutional neural network is introduced in detail,including the way of model training and testing.Finally,the experiment is finished on the deep learning framework named TensorFlow,including data processing,model definition,and model training and testing.The model is trained and tested using MIT_Indoor dataset.And the dataset is processed by data augmentation and data pre-processing before model training.After model training,the result is analyzed concretely.Besides,the mechanism of convolutional neural network is studied by feature visualization.The results are analyzed and compared with other scene recognition results.The comparison result shows the effectiveness of convolutional neural network in indoor scene recognition.