Edge Computing-based Intelligent Garbage Classification and Recognition Application

As the standard of living improves, there is a corresponding increase in the variety of materials, leading to a rise in household garbage. Dealing with the challenge of classifying numerous types of garbage can be a daunting task. The accuracy of waste separation cannot be effectively improved solely by relying on individuals' common sense perception, knowledge propaganda, and training. A more comprehensive and efficient approach is needed. This paper proposes an intelligent waste classification and recognition application based on edge computing to address the aforementioned issues. The edge mobile intelligent waste classification and recognition application utilizes a small-sample lightweight machine learning model for training compressed quantization on the Edge Impulse Platform (EI). The application is built using the MobileNetV2 model and migration learning theory. When a user scans a trash image by running the application through a mobile edge device, the tiny machine learning model is invoked to perform edge computation and identify the class of trash. System testing on the EI platform and field applications demonstrates that the average response time for intelligent waste classification and recognition is 0.66 seconds. Additionally, the average inference time is less than 5 milliseconds, and the application achieves an average recognition accuracy of 97% or higher. This intelligent waste classification and recognition system exhibits excellent overall performance and offers great convenience for individuals in their daily waste classification tasks.