Modified Algorithm Of Capsule Network For Classifying Waste Image

With the continuous development of social and economic and the rising awareness of environmental protection among the people,waste recycling and automatic sorting have become an effective way to solve the contradiction of increasing urban waste and have attracted much attention from academia and industry.It would be inefficient to rely solely on manual sorting.However,relying on traditional image classification methods tends to ignore the spatial relationships between image features and can easily lead to misclassification of the same object Therefore,there are still more significant problems in the development of waste image classification.For the problem of domestic waste image classification,this paper proposes an improved capsule network model(ResMsCapsule),which preserves the spatial relationship and location information of features to a certain extent and weakens the impact of information loss on network recognition performance.The model improves the performance of the initial capsule network by fusing the residual module and the multiscale capsule module,strengthens the adaptability of the network to multiscale features,and can obtain richer information on the features of household waste images.In addition,by comparing and analyzing the effects of different dimensional combinations of primary capsules on image classification performance during feature fusion and verifying the necessity of reconfiguring the sub-network,the preferred parameters of the model in the experimental scenario of this paper are derived.Finally,the ResMsCapsule model is validated on the household waste image dataset.The experimental results show that the model classification accuracy is 91.41%,which is about 8%,9%,and 5%higher than the average accuracy of classical convolutional neural networks AlexNet,VGG16,and ResNet18,respectively.Compared with the existing classification methods,the ResMsCapsule model proposed in this paper has high accuracy,few parameters,and strong generalization ability,which can provide better classification performance for waste images.