Deep Learning Segmentation Based Attention Mechanism Module for Navigation System

This study implements state-of-the-art segmentation models on the unique conflate Area-of-Focus dataset. The goal is to accurately identify and segment a specific regions in images that contain a large number of objects and obstacles for navigation system. The effectiveness of these models will be evaluated using metrics such as dice loss score and IoU metrics. The study aims to demonstrate the potential of deep learningbased segmentation in removing insignificant portion of the image leaving only the Area-of-Focus. Out of the four model selected, SegNet achieves a dice accuracy rate of 98.98%, dice loss of 1.62% and an IoU of 95.2%.