Garbage classification system based on improved ShuffleNet v2

Garbage classification technology is not only an important basis for the harmless treatment of waste and resource recovery, but also the inevitable trend of social development. The current garbage classification methods rely on manual classification in the garbage collection stage, and it is difficult to achieve satisfying results in consistency, stability, and sanitary conditions. For this reason, this study designs and develops a garbage classification system based on deep learning that can recognize and recycle domestic garbage. Focusing on the problems of low accuracy and poor real-time performance, a lightweight garbage classification model GCNet (Garbage Classification Network) is proposed. GCNet contains three improvements on ShuffleNet v2, including the design of parallel mixed attention mechanism (PMAM), the use of new activation functions, and transfer learning. The experimental results show that the average accuracy of GCNet on the self-built dataset is 97.9%, the amount of model parameters is only 1.3M, the single inference time on Raspberry Pi 4B is about 105ms, and the classification system needs only 0.88 seconds to complete the classification and collection of a single object. The method proposed in this paper is an effective attempt at machine vision in garbage classification and resource recovery. With the improvement of technology, it will effectively promote academic exploration and engineering application in the field of resources and environment.