Garbage Detection And Identification Based On Deep Convolution Neural Network

The continuous rise of social material production level and population leads to the expansion of the total output of waste.The detection,identification and classification of waste to improve the automation level of treatment has gradually become the focus of the times under the above background,which has great social and economic value.Therefore,this paper proposes a garbage detection and recognition algorithm,improves the speed and accuracy of terminal deployment,and then constructs a complete micro supervision system including garbage person face identification.It mainly includes the following aspects:(1)Aiming at the problem that the current garbage detection data sets are small and mostly classified data sets,this study can not be used directly.Firstly,the initial data set is constructed based on Huawei garbage data set? Secondly,an automatic script is written to fully analyze the data set and visualize the results? Finally,a data expansion strategy based on crawler technology and data semi intelligent annotation method are proposed for data expansion.Finally,a garbage detection data set including 44 categories,34395 targets and labels is constructed,which provides a good data basis for follow-up research.(2)According to the hardware constraints and real-time constraints of the algorithm in the terminal deployment in this research scenario,and weigh the accuracy requirements.Firstly,yolov5 is selected as the baseline network and ghost module is introduced to improve the structure of bottleneckcsp,the main unit of the network? Secondly,the original detection head branch structure of the algorithm is redesigned,and the part with less revenue contribution is replaced by lightweight convolution module,and the detection head anchor frame is redesigned? Finally,aiming at the side effects of some lightweight improvements,an attention mechanism is integrated into each stage of the size change of the backbone network feature map to improve the network accuracy.Finally,the yolov5-4dwga algorithm is proposed.Compared with the original yolov5,the map is improved by 2.95 % and the detection speed is improved by 28.15 %.(3)A garbage detection and identification system is designed and implemented.The system includes garbage identification and detection module,authentication module and interface display module.Firstly,the overall architecture design of the system is completed,and the yolov5-4dwga garbage detection and recognition algorithm is embedded? Secondly,face recognition is realized by using face detection based on RESNET and key point extraction based on landmarks,and the centroid tracking algorithm is introduced to improve it,and the authentication module is completed? Finally,the interface display module is realized by pyqt5.Although the interface is relatively simple,it still provides basic module components for subsequent system updates.Finally,the development of the whole micro system is completed,which provides support for the implementation of the algorithm.