| It is necessary to carry out waste reduction,resource utilization,innocuity and ecological civilization.At present,people mainly classify waste by manual judgment,but the ability of public waste classification is limited,so it is difficult to distinguish all kinds of waste with naked eye in a short time.With the popularization of smart devices such as mobile phones,it is possible to quickly detect waste classification based on smart terminals.In this paper,a waste classification depth learning model based on image recognition is proposed,and a model and method for waste classification based on computer vision are provided by using the existing waste image data and the data set created in this paper.Specific studies include:Firstly,this paper studies the characteristics of deep learning technology and the waste classification image data set is constructed.Under the trend of promoting waste classification and resource recycling,how to carry out waste classification quickly,efficiently and automatically has become a research trend and hot spot in the construction of waste classification model based on deep learning.At the same time,there is a great demand for the standard waste image classification data set.Therefore,this paper collates the existing public waste image data sets,and finds that there are some limitations in this kind of waste data sets,such as the small number of data sets,the single waste background and the obvious features.It is not representative with the waste form in daily life and the sample distribution is uneven.Aiming at the above problems,this paper constructs an efficient and complete single target waste classification image data set based on image recognition,which can recycle waste image data set(NWNU-WASTE),which contains 21600 recyclable waste images.The data preprocessing is carried out to provide the data basis for the training of the following classification model.Secondly,a single target waste image classification model based on attention mechanism is proposed.To solve the problem of imperfect waste classification model based on deep learning,this paper adds self-supervision module on the basis of residual network to construct an efficient and accurate waste classification model.By using the open waste classification data set(Trash Net)and the NWNU-WASTE data set created in this paper,the confusion matrix,the receiver operation characteristic curve and the area below the curve,the change of loss value during model training,and the classification accuracy are used to evaluate the model.The model proposed in this paper is classified on the Trash Net data set,the accuracy is 95.87,the accuracy is higher than that of the waste classification model proposed by other scholars,and the accuracy is85.98 on the NWNU-WASTE data set.The model proposed in this paper is Alexnet、Googlenet、VGG16 more accurate in different categories of waste classification.Finally,a YOLO-WASTE multi-objective waste detection model based on transfer learning is proposed.For the same picture of multiple waste identification problems,Create multi-target waste detection image data set,the dataset contains a total of 1547 images.Based on the You Only Look Once model of transfer learning and single-stage target detection methods,Using the multi-target waste detection image data set constructed in this paper,A YOLO-WASTE multi-objective waste detection model is constructed.And compared with the Single Shot Multi Box Detector model,The experimental results show that the m AP value of waste detection model based on YOLOv4 is 90.58,The m AP value of waste detection based on SSD is 63.55.Through the test results of the model,the change of the loss value in the process of model learning,the detection accuracy,the AP value,the MAP value and the detection time,A YOLO-WASTE multi-objective waste detection model proposed in this paper has good detection effect.this study applies a single-stage target detection algorithm to waste classification. |
