Objectives It is urgent to solve the problem of rapid construction and optimal deployment of emergency communication network when large-scale natural disaster occurs.

Methods First, the peak points feature dataset is built up by fusing the digital elevation model with contour, slope, and slope aspect to improve the feature information of the peak area. Then, the screening conditions of the peak area for emergency communication are given and an enhanced faster region-based convolutional neural network (Faster R-CNN) is proposed to adapted to the multi-scale positioning of the peak area. Finally, we split, detect, and then merge the large area to improve the recognition effect of the peak area, and use the local maximum value method to achieve the precise extraction of peak points for emergency communication.

Results Comparing the proposed improved algorithm with other algorithms, the mean average precision (mAP) reaches 94.92%, the accuracy of peak points extraction reaches 94.2%, the effective coverage rate of the communication node reaches 80.56%, and the visibility rate reaches 77.43%.

Conclusions The proposed method is effective in identifying small targets, more in line with the terrain characteristics than other methods, and meets the actual communication needs in terms of communication to achieve a larger coverage. The extraction results of the proposed method can be used as a reference for the deployment of emergency communication nodes to help complete the optimization of the deployment of communication nodes for emergency rescue, and to protect the regional communication smooth.