Research On Intelligent Classification Of Domestic Waste Based On CNN

At present,China’s domestic waste classification is still in the stage of manual sorting.There are many problems in the process of waste classification,such as heavy workload,low classification accuracy,unable to feed back information in time and so on.Also,the current intelligent research of waste classification cannot be applied to the intelligent classification of residential domestic waste due to the lack of research categories or the poor real-time performance of waste identification.For the sake of the intellectualization and high efficiency of community garbage classification,a real-time garbage identification research method based on YOLOv5 s model is proposed.Yolov5 s model has the advantage of fast target detection speed,and can ensure the intelligent garbage classification system to detect the target object in real time.However,due to the fuzzy imaging of domestic garbage,mutual occlusion of target objects,complex scene and other shortcomings,the accuracy of some target objects detected by yolov5 s model is not high.So the research proposes an improved method based on YOLOv5 s model on the basis of basically maintaining the original detection speed.The improvement methods are as follows:1.In order to improve the problems caused by the shortcomings of blurred imaging and complex scenes,the research improves the image contrast through image enhancement,and uses image processing technologies such as image weighting to alleviate the imbalance of the number of each category’s labels in the image data.So as to improve the target detection accuracy of the model.2.For the gradient descent optimization algorithm of the model,through the comparative study of MBGD algorithm with momentum and Adam optimization algorithm,MBGD algorithm with momentum is determined as the optimal gradient descent optimization algorithm of yolov5 s model in this experimental environment.So as to ensure the better accuracy of target detection model.3.In order to alleviate the problem that YOLOv5 s model is difficult to distinguish mutually occluded target objects,the research introduces the attention mechanism into the backbone network of the model,and adds channel attention and spatial attention to the convolution layer of the backbone network to improve the feature extraction ability of the model,so as to further improve the target detection effect of the model.The experimental results show that under the same experimental environment and conditions,the improved yolov5 s model improves the real-time target detection accuracy of most garbage categories,and the target detection accuracy of garbage cans in a single garbage category increases the most,17.1% higher than the original model,and the value of m AP@.50 is 2.3% higher than the original model.