| The harmonious integration of human and environment is an important guarantee for human survival and social sustainable development.Garbage classification plays an important role in environmental protection and resource recycle.At present,the garbage classification is mainly completed by artificial assistant method,which is inefficient and wastes human resources.Therefore,this thsis proposes a kind of common recyclable bottle garbage detection algorithm based on convolution neural network,which realizes the intelligent detection and classification of bottle garbage.The main contents of this thsis are as follows:Firstly,the basic knowledge of convolutional neural network and the target detection algorithm of common convolutional neural network are studied.On the basis of analyzing the performance of different detection algorithms,Faster RCNN(Faster Region Based Convolutional Neural Networks)is adopted as a target detection algorithm,In view of the different sizes of the bottle target to be detected in the garbage image,the feature extraction module and candidate box generation module in the algorithm are improved.Secondly,in view of the fact that there are few garbage image public data sets,5000 common recyclable bottle garbage images are collected,and the number of images is expanded to 20000 through data enhancement technology,so as to build a data set of common recyclable bottle garbage images.Finally,the improved VGG16 model is used to extract the features of the bottle garbage image after scaling by bilinear interpolation.In the improved VGG16 model,the convolution kernel size and network structure are mainly improved,and the random regularization function is introduced to reduce the feature dimension and weaken the over fitting phenomenon.On this basis,optimize the anchor mechanism parameters in the regional suggestion network,adjust the classification function of the regional suggestion network and the back-end of Faster RCNN algorithm,and achieve the positioning and classification of bottle targets in the garbage image.The algorithm in this thsis is simulated in the deep learning framework to verify the effectiveness of Faster RCNN target detection algorithm based on the improved VGG16 model. |
