A YOLOV3 System for Garbage Detection Based on MobileNetV3_Lite as Backbone

Abstract

With the development of science and technology, China has basically realized the goal of building a moderately well-off society in an all-round way. As people’s income and quality of life are constantly improving and their consumption level is gradually increasing, the quantity of household garbage is also growing rapidly. It is of great urgency to implement the classification and management of household garbage to make full use of garbage resources and reduce environmental pollution as well as land resource occupation. In order to conduct the classification and management work of household garbage more efficiently, this paper develops a garbage detecting system using a lightweight model MobileNetV3_Lite, which is promoted from MobileNetV3, as the backbone network to reduce computation and building up a YOLOV3-MobileNetV3_Lite detection model. The model successfully detects four categories of garbage, effectively and accurately determining the category and quantity of the waste. After the model is exported, deployed, and tested on NVIDIA Jetson Nano, the frame rate is up to 25. In the actual test, the accuracy for judging the category of garbage reached 94.56% and 90.91% for detecting the quantity of each category of garbage.