Utomatic target detection from SAR images

Automatic target detection in SAR images is an application that has grown in popularity in recent years. The most important reason for its popularity is that SAR sensors can display in all weather conditions, day and night. This situation increases the importance of SAR images. The fact that SAR images are very difficult to interpret with the eye and are different from the images that the eye is accustomed to have increased the need for object detection in these images. For this reason, in this study, deep learning method was used to detect objects on SAR images quickly and with high accuracy. The importance of the concept of deep learning is increasing day by day, with the rapid progress of technology and with the advancement of computer hardware connected to these systems. In many sectors, autonomous systems have started to be used instead of humans. One of the areas in which deep learning algorithms are included and constantly improved is object detection. In this thesis, the Mask R-CNN algorithm was used for object detection in SAR images, unlike the studies in the literature, and its accuracy was compared with the Faster R-CNN algorithm. In the thesis study, the analyzes were made in Python language using Google Colab, Jupyter Notebook, and it was determined that the Mask R-CNN algorithm had a higher classification rate than the Faster R-CNN algorithm during the test phase. This estimation rate is considered to be a relatively high classification rate when considering military vehicle types