Skip-YOLO: Domestic Garbage Detection Using Deep Learning Method in Complex Multi-scenes

Abstract

It is of great significance to identify all types of domestic garbage quickly and intelligently to improve people's quality of life. Based on the visual analysis of feature map changes in different neural networks, a Skip-YOLO model is proposed for real-life garbage detection, targeting the problem of recognizing garbage with similar features. First, the receptive field of the model is enlarged through the large-size convolution kernel which enhanced the shallow information of images. Second, the high-dimensional features of the garbage maps are extracted by dense convolutional blocks. The sensitivity of similar features in the same type of garbage increases by strengthening the sharing of shallow low semantics and deep high semantics information. Finally, multiscale high-dimensional feature maps are integrated and routed to the YOLO layer for predicting garbage type and location. The overall detection accuracy is increased by 22.5% and the average recall rate is increased by 18.6% comparing the experimental results with the YOLOv3 analysis. In qualitative comparison, it successfully detects domestic garbage in complex multi-scenes. In addition, this approach alleviates the overfitting problem of deep residual blocks. The application case of waste sorting production line is used to further highlight the model generalization performance of the method.