The invention discloses an end-to-end
impact crater detection and identification method based on a fully
convolutional neural network structure. According to the method, detection and identification of
impact craters in a celestial
remote sensing image are realized synchronously. A network established according to the method is named as CraterIDNet. The established network is composed of two
impact crater detection channels and an
impact crater identification channel. Network weight parameters are only composed of convolutional
layers and have no fully connected
layers. According to the method, for the
impact crater detection channels, a candidate box scale optimization and density adjustment mechanism is provided, the optimum candidate box selection is realized, the detection performancefor small
impact crater targets is greatly improved, moreover,
synchronous detection is carried out on multiscale impact crater targets through utilization of different receptive fields, and the network has the capability of detecting the large-scale range changing impact crater targets. For the impact crater identification channel,
grid pattern graphs with rotation and
scale invariance are generated through utilization of a
grid pattern layer, so the impact crater identification is realized without establishing a matching feature
database According to the method, the identification robustnessis improved.