Research On Domestic Garbage Identification Based On YOLOv5 Improved Algorithm

In recent years,domestic waste treatment,as a livelihood project to improve the quality of life and protect people’s health,has attracted more and more attention from governments around the world.However,in the face of a huge amount of domestic waste,the traditional domestic waste treatment plants are still struggling with manual sorting.In addition,the manual garbage sorting work is also accompanied by the existence of low work efficiency,harsh environment,labor instability and high labor costs.In the face of various disadvantages of manual labor in domestic garbage disposal,automated garbage sorting factories came into being.As its technical support,the target detection algorithm has received extensive attention in recent years and has been favored by many researchers.The target detection algorithm has been iteratively updated for a long time,and it has faster processing speed and higher recognition accuracy when processing images.Some scenes have incomparable advantages of artificial intelligence,making this technology more and more widely used in many fields,in various industries.shine.Therefore,the target detection technology using deep learning has become an important means of automatic garbage identification and classification,and it is of great significance to the identification of domestic garbage.This paper proposes a domestic garbage identification algorithm based on YOLOv5 and hierarchical classification.First,the data set is preprocessed,the category distribution of the data set is analyzed,and some data are manually labeled and supplemented to obtain a data set that can be used for garbage identification.An improved scheme of multi-level classification is proposed to improve the YOLOv5 algorithm.By analyzing the predictions of each garbage data in the classification confusion matrix diagram,combining multiple error-prone data for separate classification training,the corresponding secondary classifier is obtained,the role of the secondary classifier is to re-correct some of the predicted categories of the YOLOv5 algorithm,and return to the original algorithm for Loss calculation again.Comparing multiple models,the lightweight Mobile Netv3 model is finally selected as the secondary classification network of the improved scheme.Through experiments,it is proved that the improved algorithm is effective in the identification of domestic garbage,the m AP of the original algorithm is increased to78.8%,and the recognition accuracy is significantly improved.It has a higher detection speed and ensures the lightweight of the network model.Considering that the target output position of the improved algorithm is not accurate enough,it is proposed to use semantic segmentation technology to locate the target prediction area more accurately.The pre-training model provided by U2 Net is fitted by manually calibrating part of the data.The experimental results show that the semantic segmentation algorithm can effectively extract the target edge information,which makes the target position output by the model more accurate and has a stronger visual effect.Through a large number of experiments compared with the relevant performance indicators of the original algorithm,it is found that the improved algorithm in this paper can effectively improve the accuracy of household garbage identification while ensuring the model is as simple as possible,and the actual position of the target in the image can be further accurate through semantic segmentation technology,the algorithm improvement is effective,and it has high practical value in the field of domestic waste identification.