Efficient object detection of non-perishable in smart garbage sorting booths using the Efficient-Net model
Zhang Wei, Shi Zhebin, Huang Qiucheng, Gong Baojun · 2022
For object detection in garbage sorting applications, the YOLOv4 model have been employed in the detection of abnormal objects during waste sorting. YOLO models have been proven as an effective solution with stable performance and accuracy. In the deployment, YOLO models usually consume considerable computational resources of GPU, which limit wide application of YOLO-based monitoring system. In addition, mobile terminals such as smart cameras have limited computing resources, and conventional AI models face difficulty in deployment on mobile terminals. In this case, it is of high practical significance and application value to carry out the research on the lightweight of the recognition model. The key point to realize the lightweight of the object recognition model without loss of recognition accuracy. This study designs and implements a lightweight object detection model based on the EfficientNet architecture, which can identify non-organic wastes in the perishable buckets of intelligent garbage sorting booths, thus performing monitoring of residents’ inappropriate garbage delivery behavior.