A Household Garbage Classification and Collection Device Based on Machine Vision and Deep Learning

Garbage classification is the key link of garbage recycling. Based on machine vision and lightweight convolutional neural network MobileNet V2, a garbage collection system for municipal solid waste classification and identification is developed in this paper. Firstly, the MobileNet V2 network model was reconstructed using Python language in the framework of TensorFlow2.0, and the transfer learning was carried out using the garbage sample made by our own. The results show that the validation accuracy of 99.24% on the validation dataset can be obtained when the input image size is 160×160. Secondly by using c # language a PC control software with functions of image acquisition, classification recognition and condition monitoring is developed, by which living garbage can be divided into recyclable waste, kitchen waste, hazardous waste and other waste four categories quickly and intelligently without or reduce human participation. The delivery accuracy of the device can be high as 98.9%, and the pitch time is 3-6 seconds. The system can be used as an intelligent home kit to achieve the purpose of garbage classification from the source.