Analyzing the Performance of Semantic Segmentation Techniques in Automatic Corrosion Detection

Among the competitors in semantic segmentation for image understanding, DeepLabv3 has dominated as a state-of-the-art model that is well-known to offer excellent pixel-level classification. Explaining the proposed corrosion detection techniques will be well explained in this paper through a comparison between DeepLabv3 and its improved version called DeepLabv3+ for semantic segmentation of corrosion formats. We also compare DeepLabv3 with other state-of-the-art models, namely SegFormer and SegNet, LinkNet…, to show why it is superior in identifying corroded regions correctly in images. We learn about the differences between them and then analyze how well they can detect if a certain area in an input image is corroded. This paper emphasizes the leaps made by DeepLabv3+ and serves to provide a better understanding of industrial concentration across recent corrosion detection automation advancement techniques.