Research On Furniture Image Classification Based On Feature Fusion Of Convolutional Neural Networks

With the rapid development of artificial intelligence,deep learning methods combined with image processing technology have been widely applied to all aspects of daily life.Aiming at the diversity and complexity of furniture image data,this thesis focuses on the research of furniture image preprocessing,furniture local feature extraction,feature fusion based on convolution neural network,fine-grained furniture image classification and so on.The main work of the thesis includes:1.In the aspect of furniture image preprocessing,the thesis analyzes the classification performance of the convolutional neural network for the collected furniture image.The GrabCut algorithm is used to segment the background of the furniture image.Then,in order to automatically generate dimension normalized data suitable for convolution neural network,the geometric relationship between the detected width and height of furniture contour and the width and height of the image to be generated is used to scale up to normalize the scale,and then write it to the specified file directory according to the rule name.This preprocessing automation process effectively reduces the amount of manual processing of furniture images.2.For furniture image local feature extraction and feature fusion based on convolution neural network,a feature fusion algorithm based on AlexNet convolution neural network model is proposed to fuse LBP and HOG local features in the full connection layer with global features,and the learning algorithm of fusion weight is studied.The simulation results show that the feature fusion algorithm can better describe the basic features of the image and improve the accuracy of furniture image classification.3.A fine-grained Bilinear-CNN model is proposed for the problem of low classification accuracy between similar furniture.The convolutional neural network and LBP are merged into bilinear features by bilinear functional relationship to realize the interaction and fusion of global features and LBP features in spatial position.Simulation results show that the improved Bilinear-CNN model can further improve the classification accuracy between similar furniture images.