Abstract:
The development of a synthetic method for planetary object recognition based on the integration of two architectures, Mask R-CNN and the convolutional neural network (CNN) U-Net, is presented. The proposed method was verified on lunar craters of various categories selected from images obtained by modern satellite missions. Object recognition is performed using criteria such as the ratio of stratigraphic characteristics, morphological features, and optical structure.
Fulltext:
PDF file (1880 kB)