Research On Automatic Identification Of Lunar Crater Gravity Anomaly Based On Convolutional Neural Network

Craters are the most common units on the lunar surface and contain important information about the geological evolution of the moon.By studying its size,distribution,age and other characteristics,researchers can infer the sequence of crater formation,impact strength,and lunar geological information.Therefore,crater has become a key point for scholars to study the lunar surface and the shallow lunar surface.Since the 1960 s,major space countries(organizations)have successively carried out lunar exploration activities and obtained multidisciplinary data.In recent years,with the advancement of technology and the capacity increasing of the probe,it has become a reality to research the lunar surface and internal structure by gravity methods.To research on craters through gravity data,the first step is to delineate anomalies.Similar to the outline of crater,delineating anomalies outlines also has the problem of relying on the subjective experience of researchers and low work efficiency.Predecessors have done a lot of research on recognition methods,but due to the limitations of the method itself and hardware,good recognition results can only be obtained in a small range or under ideal conditions.To solve this problem,this research combines lunar gravity and convolutional neural networks to delineate the contours of gravity anomalies caused by craters in some area as a data set,and divide the data set into training sets,test set and validation set according to training requirements,then input the data set into the classic convolutional neural network model U-Net for model training and update the model parameters,so that the model can learn anomalies features at different scales.The trained model can accurately extract the features of the target anomaly according to the change trend of value in the anomaly map and automatically delineate its contour and position.In the training process,this research conducted training more than ten times with different hyperparameters values.The network model that finally obtained can more accurately delineate the abnormal location,which proves that the use of convolutional neural network to identify gravity anomalies caused by craters is feasible,the trained model can complete the model recognition in a short time,and the accuracy can reach more than 85%.The various indicators for evaluating the recognition quality have advantages obviously which compared with the previous research methods.In addition,we also used the network model to automatically generate anomalies and identify the gravity anomalies generated by the research hotspot craters for model evaluation.The results obtained show that the method is feasible.