19mech01/corrosion-monitoring-using-UNET

In order to avoid regulatory violations, downtime, or fatal disasters, industrial assets must be maintained in a timely manner. One of the biggest factors for timely maintenance is corrosion of the assets. Corrosion causes generation of irregular surface that looks bad in appearance and can produce serious problems. To avoid corrosion problems, timely maintenance must be scheduled well in advance. For this, images-based corrosion detection can be a useful tool with respect to real time corrosion tests. In this work, images-based corrosion detection is performed using Deep-learning UNET-8layer architecture. More than 400 images of corrosion data with their binary ground truth or labels are trained to analyze and classify the corroded region in the images. After training, the model is tested, and it is found that the prediction over corroded part of original image data can be done with the model with binary accuracy of 95.28% and validation binary accuracy of 96.87%.

This repository is not active