Garbage classification recognition method based on YOLOv5-BC

Abstract

The application discloses a garbage classification and recognition method based on YOLOv5-BC, which comprises the following steps: acquiring a three-channel garbage image, and generating an input tensor after pretreatment; inputting the input tensor into a garbage classification and recognition model, using a feature extraction module to extract a detail texture feature map, a shape structure feature map and a semantic background feature map, and performing mixed attention enhancement and multi-scale pooling processing on the semantic background feature map to generate a deep semantic encoding feature map; using a multi-scale bidirectional feature fusion module, sequentially fusing the deep semantic encoding feature map with the shape structure feature map and the detail texture feature map through a top-down path, and performing secondary fusion through a bottom-up path to output a multi-scale fusion feature map; and inputting the fusion feature map into a detection head to predict a bounding box and a class probability, and outputting a result after non-maximum suppression. The application effectively suppresses complex background interference and improves the detection accuracy of multi-scale garbage targets.

Images