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

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