Research On Waste Classification And Identification Method Based On Improved ResNet Network Model

Unlike in the past,when people could rely on nature to decompose and handle waste,the amount of waste today far exceeds the natural limit of capacity.This not only poses a serious threat to human life but also causes significant damage to the environment.Therefore,the problem of efficient waste disposal has become a very real and important issue in today’s society,which needs to be solved by the joint efforts of all parties.Sorting waste can turn a lot of it into something useful,but in reality,many people do not know much about each type of waste,resulting in a lot of waste being thrown on the ground randomly,which leads to many problems,such as:higher costs for manual sorting and waste of resources."To do a good job,a good tool must first be used." In order to better promote waste separation,it is necessary to make full use of the power of technology.With the development of technology,it has gradually become possible to use artificial intelligence technology to achieve automatic classification of waste,and it has a wide range of application prospects and social benefits.The use of AI technology can make waste sorting more efficient and accurate,and can improve the accuracy of sorting by processing large amounts of data.In addition,the use of AI for waste sorting can reduce error rates,improve sorting efficiency,and greatly ease the burden of manual labor compared to traditional manual sorting.In addition,the use of AI technology can provide more data support and innovative ideas for waste classification,as well as reduce problems such as environmental pollution and waste of resources.Based on the above background,the main work of this thesis includes the following:(1)A method for waste classification and recognition based on an improved residual network model is proposed in this thesis.The study uses publicly available household waste image data from the Huawei competition as the research object.The ResNet network is used as the basis,and the model’s convolutional layers are modified,the activation function is replaced,and the residual is improved to obtain more rich features of household waste images,thereby improving the recognition and classification accuracy of the model established in the thesis.The fully connected layer is removed,the dropout layer and dense layer are added,and the optimizer is replaced to enhance the model’s generalization ability.The experiment shows that this algorithm achieves up to 92.35%classification accuracy on public image datasets,which is better than any other algorithm,so it has some research and application value.(2)A method for waste classification and recognition based on an improved ResNet network and an introduced attention mechanism is proposed in this thesis.The main purpose of this study is to use a ResNet-based classification network and data augmentation methods to expand the Huawei waste image dataset.The method improves the performance of the neural network by modifying the convolution structure in the residual block,adding attention mechanisms,and replacing the loss function to enhance the importance of the upper-layer signal.This improves the classification effect of the neural network when processing waste information.Experimental results show that the proposed method is effective and performs better than other methods in various indicators,thus demonstrating its feasibility.(3)A visualization test was carried out using the Gradio framework in a waste sorting and recognition platform and combined with the model experimental results parameters.The platform essentially allows users to upload images of waste and send back the results of their classification and identification in order to spread knowledge about waste.This allows users to quickly identify the type of waste.