Waste recycling has always been an important issue for environmental protection, resource recovery and social livelihood. In order to improve the resource utilization rate of recyclable materials, this work designs an intelligent garbage classification system based on deep learning, which mainly includes two parts: user software and sorting equipment. Among them, user software is used for billing of recyclable materials and visualization of classification information, and sorting equipment classifies common recyclable materials into boxes.

The user software uses PyQt technology to develop the interface to realize the information visualization of the delivery process, and uses the Internet of Things technology to realize the billing management of the user's IC card, and gives the user economic return according to the garbage put by the user. The sorting equipment uses strawberry pie 4B + board as the central controller, based on the Tensorflow framework, uses the ResNet-50 model to extract and classify image features, and uses cosine annealing and migration learning strategies to optimize and improve the model to realize high-precision recyclable identification. Considering the calculation and storage requirements, cloud deployment is adopted, and Flask technology is used to directly package the model prediction into http interface, through TCP network communication technology to achieve the interaction of strawberry pie and cloud server data, to achieve rapid identification, the use of programmable pins to transmit pulse signals, with the driver as the medium, through the programming of 86 motor speed control and positioning, and then control the square skateboard angle to achieve classification collection;

after testing, the user software has good human-computer interaction effect, the income calculation is accurate, and the background data update speed is fast. The identification speed of the sorting device using cloud deployment is 0.114s/time, and the identification accuracy is 97.42. The system is stable, the garbage classification efficiency is high, and the hardware cost is low, which has a good market prospect.