Research on IC-LOGO Extraction Algorithm Based on Deep Semantic Segmentation Network

Considering the characteristics of the logo and character areas being intertwined, along with the variability in size and irregularity in shape of logos within integrated circuit (IC) identification images, this paper propose a logo extraction algorithm based on a deep semantic segmentation network. Initially, preprocessing and data augmentation techniques are employed to facilitate the creation of an IC identification dataset, alongside the establishment of a semantic segmentation dataset specifically for the LOGO area. Subsequently, network models including DeepLabv3+, U-Net, PSPNet, and HRNet are trained and validated to identify the optimal model for IC logo extraction. The experimental results indicate that the deep semantic segmentation network demonstrates effective segmentation performance in the extraction of integrated circuit logos, with the HRNet model achieving high segmentation accuracy and an average intersection-over-union ratio of 80%.