Segmentation is the process of partition digital images into different parts depending on texture, color, or intensity, and can be used in different fields in order to segment and isolate the area to be partitioned. In this work images of the Moon obtained through observations in Astronomy and space dep. College of science university of Baghdad by ( Toward space telescopes and widespread used of a CCD camera) . Different segmentation methods were used to segment lunar craters. Different celestial objects cause craters when they crash into the surface of the Moon like asteroids and meteorites. Thousands of craters appears on the Moon's surface with ranges in size from meter to many kilometers, it provide insights into the age and geology of a Moon's surface. Therefore, it is important to study them and determine their characteristics. So, several segmentations methods were used in this study these are: K-Means, Single Feed Forward Neural Network (SFFNN), and hybrid segmentation methods. K-Means method applied with different number of clusters (k), that were used to segment Moon images and isolate lunar craters, where k=1,2,3, and 4 were used. But, all of them did not identify the boundary of craters, only K=3 gave useful results. SFFNN was also used in this work, it trained by a novel method, where weights have been replaced by masks, that create depending on the images features and targets. Thirteen lunar craters were used, ten of them utilized in training process and the last three images were used to test the performance of network. But also this method did not segment lunar images and identify the boundaries of lunar craters clearly. So, in attempt to overcome this problem, the new hybrid method was proposed, that combine the concepts of KMeans and SFFNN methods. The main advantages of the proposed hybrid method is that it does not require much data in the training process as it is known in other networks, where the K-Means cluster segmentation method gave a shortcut to correlating masks with images, which led to giving perfect results in a short time. Then, results show the proposed hybrid segmentation method was succeed to segment lunar crater and identify the craters boundaries clearly.
