A method of realizing remote sensing image object detection based on mask R-CNN as geographic WPS service

Abstract

The present invention provides a method for realizing the remote sensing image feature detection based on Mask R-CNN as geographic WPS service, which can realize the remote online and multi-person sharing application of the Mask R-CNN model remote sensing image feature detection. According to the application requirements, the present invention realizes the automatic feature detection and identification of remote sensing image data by calling the Mask R-CNN model, and the automatic conversion of the space vector polygon and the storage of the spatial database for the result of the feature detection, and uses the standard geocoding format as The complete application service process of network transmission of ground object detection results to the client. The present invention greatly reduces the manual workload in the process of ground object detection in remote sensing images, which has a very good role in promoting the in-depth application of the Mask R-CNN model in ground object detection.

Description

technical field [0001] The method relates to the technical fields of geographic information systems, computer networks, and deep learning, and is specifically a method for realizing remote sensing image object detection based on Mask R-CNN as a geographic WPS service. Background technique [0002] Mask R-CNN is a deep learning model for image recognition and detection. Because of its excellent target detection capabilities, it has also been extended to remote sensing images for the detection and recognition of specific objects. Compared with traditional remote sensing image object recognition algorithms , and achieved good application results. [0003] Although using the Mask R-CNN model to detect ground objects in remote sensing images has great application advantages, in the current specific practice, there are the following difficulties: [0004] (1) Using the Mask R-CNN model to perform faster and better detection operations requires a hardware platform equipped with hi...

AI Technical Summary

Problems solved by technology

[0004] (1) Using the Mask R-CNN model to perform faster and better detection operations requires a hardware platform equipped with high-performance CPU and GPU, which is often not available to ordinary application personnel;

[0005] (2) The size and file size of the remote sensing image far exceeds the image usually detected by the Mask R-CNN model, so the common method now is to crop the remote sensing image into a small image that meets the requirements, and then give it to the Mask R-CNN model separately. Detection, which increases a lot of additional data preprocessing workload;

[0006] (3) The detection result of the Mask R-CNN model is a pixel image without geographic coordinates and has no geographic spatial meaning. It needs to be manually converted into geographic data such as space vectors to have practical application value;

[0007] (4) The ground object detection of the Mask R-CNN model is difficult to integrate with other related geographic processing operations to form a unified remote sensing image ground object detection process

Images