Abstract:This paper aims to solve the problem of difficulty in feature selection of mountain peak, considering the transformation of mountain peak recognition in DEM data into a target detection task in digital images. With the help of deep learning technology, a mountain peak extraction method based on improved Faster R-CNN is designed. First, the DEM data is superimposed on the contour map and the grayscale map, Used the Faster R-CNN-based target recognition framework, replaced the original VGG16 with ResNet-101 as the feature extraction network of the mountain peak recognition model, and introduced the K-Means clustering algorithm in the RPN anchor frame size setting to realize the application of self-built mountain peaks the anchor frame parameter setting of the sample set PEAK-100. The improved Faster R-CNN is used to automatically extract the depth features of the mountain top,generate high-quality mountain top area, and combine the elevation to identify the final mountain peak coordinates.The experimental results show that the accuracy of the new method of mountain apex extraction is 94.82%, Compared with the traditional method, the omission rate is reduced by about 60%. To a certain extent, it reduces the mountain peak extraction effect that is susceptible to manual selection of features, and provide a new point element recognition in DEM technical approach.