Segment anything model for crater detection on the lunar surface

In this research we will discover how to implement Segment Anything Model for crater detection, Crater Detection is an essential part of planetary geology and scientific research, for mission planning and future lunar landing missions, this thesis will present a semi-automated crater detection pipeline, which will combine Segment Anything Model, which is a new state of the Art Deep learning model developed by Meta AI, we will implement classical geometric analysis, leveraging' SAM's promotable vision transformer, to generate initial crater mask candidates from 1024x1024 lunar surface images, due to the lack of datasets of the lunar surface these images will be generated using Three.js a program that uses WebGL to render 3D objects in the web browser, which then will make use of NASA's MoonKit to create an actual accurate sphere of the lunar surface with proper topographies and surface details and diverse lighting conditions, once our image is processed by SAM we will run a custom ellipse fitting algorithm, which then applies geometric filters, like area ratio, elongation and brightness to isolate crater like structures, it will use Least Squares Method for such operation, later we export the results in .PNG format showing our segmented image and Ellipse fitted image, in this research we will also discuss the difficulties and challenges of Lunar Crater Detection, shedding light on the different conditions of lunar surface, from lighting to shading, and it's challenges on image segmentation models like SAM, we will discuss the limitations of SAM in such regards for crater detection. Keywords: Segment Anything Model, SAM, Image Segmentation, Crater Detection, Computer Vision, OpenCV, Ellipse Fitter, Least Squares Method, Deep Learning, Three.js