YOLOv4-based semantic information extraction for indoor scene targets fetching algorithm research

Abstract

This paper investigates the semantic information extraction algorithm for indoor scene targets. The YOLOv4 algorithm is preferred, and the Leaky ReLU function is preferred as the new activation function scheme through the typical activation function comparison experiments to address the problems of YOLOv4 activation function preference and poor multi-scale representation of indoor targets; the attention fusion mechanism is introduced to improve the classification accuracy of the network. Experiments on the homemade Indoor-COCO indoor scene dataset show that the detection accuracy reaches 42.09%, which improves the accuracy of semantic information.