Mobile Garbage Classification Auxiliary Identification System Based On TVM

The classification of domestic waste can greatly improve the utilization rate of resources and the living environment of human beings.Therefore,it is very necessary to carry out garbage classification.However,there are many kinds of garbage,and many residents cannot accurately classify the garbage.In recent years,with the continuous development and progress of artificial intelligence technology,great success has been achieved in the field of computer vision.On the Image Net dataset,using convolution neural network,the top5 accuracy of image classification reaches 97.1%,which exceeds the accuracy of the human eye.Convolutional neural networks have achieved good results on image classification.Therefore,in-depth learning technology can be used to help residents classify garbage.A mobile garbage classification system is designed and implemented in this paper.However,most of the in-depth learning frameworks do not provide optimization on mobile hardware,which makes it more difficult to deploy indepth learning networks on mobile devices that have limited computing power.Therefore,based on the TVM open source framework,the convolutional neural network is optimized at the graph level and operator level on the mobile hardware.The work includes: training a deep learning model for garbage classification based on efficientnet algorithm,deploying the deep learning model to mobile devices based on TVM,and accelerating the calculation process on mobile devices based on autotune technology.The accuracy of the model on the test set is 91.2%.In the test environment of the Android operating system with the hardware of the g90 t of Media Tek,it takes 143 ms to perform a garbage classification.The results show that the system has a good performance of garbage classification assistant recognition and meets the required delay.