New Methods for Lunar Impact Crater Detection Based on YOLO v7 with Deformable ConvNets

The Lunar impact crater is one of the most remarkable geographic features on the surface of the Lunar. The size and shape of the Lunar impact crater are important for future Lunar landings and resource development on the Lunar surface. Hence, detecting impact craters is an essential research orientation. In this paper, we introduced two novel convolutional neural networks that combine YOLO v7 and deformable convolution of Deformable Convolutional Networks, called YOLOv7_DCN and YOLOv7_DCNv2. With the global Lunar CCD data provided by NASA, the experiment has been conducted under the same settings. The result of the experiment showed that the Precision of YOLOv7_DCN is 77.36%, and the Precision of YOLOv7_DCNv2 is 77.19%. Both values exceed the Precision of YOLO v7, which is 75.86%. Improvements have been found in several other key metrics. Demonstrating that YOLOv7_DCN and YOLOv7_DCNv2 have better performance than YOLO V7 in detecting Lunar impact crater.