Semantic Segmentation and Edge Detection—Approach to Road Detection in Very High Resolution Satellite Images

Road detection technology plays an essential role in a variety of applications, such as urban planning, map updating, traffic monitoring and automatic vehicle navigation. Recently, there has been much development detecting roads high-resolution (HR) satellite images based on semantic segmentation. However, the objects being segmented are small size, not all information is equally important when making decision. This paper proposes novel approach to road segmentation edge detection. Our aims combine these two techniques improve detection, it produces sharp-pixel maps, using masks generate edges. In addition, some well-known architectures, SegNet, used multi-scale features without refinement; thus, attention blocks encoder predict fine resulted finer A combination weighted cross-entropy loss focal Tversky function also deal with highly imbalanced dataset. We conducted various experiments datasets describing real-world covering three largest regions Saudi Arabia Massachusetts. The results demonstrated that proposed method encoding HR feature maps effectively predicts sharp facilitate accurate even against harsh complicated background.