Target Extraction Method of Remote Sensing Image Based on Scene Recognition Task

Abstract

A remote sensing image target extraction method based on a scene recognition task, comprising: S100: inputting an original remote sensing image; S200: extracting a target scene image from the original remote sensing image and obtaining a directory file of the target scene image; S300: according to The directory file of the target scene image, after obtaining different types of images corresponding to the same target scene from different types of image folders, input it into the improved segmentation network for target extraction; S400: output the extracted target result . The method firstly identifies the scene, and then segments the target in the scene where the target may exist, which solves the problem of extracting specific targets from large-scale high-resolution remote sensing images. Secondly, by extracting rich context information, not only the feature fusion is enhanced, the network can extract rich context information, but also the weighted probability fusion is performed at the end of the network, which can effectively suppress misclassification and improve segmentation performance while highlighting the target.

Description

technical field [0001] The disclosure belongs to the technical field of remote sensing image processing, and in particular relates to a remote sensing image object extraction method based on a scene recognition task. Background technique [0002] In recent years, with the rapid development of the aerospace field, my country's self-developed aerospace platforms, sensors, communication and information processing technologies have been rapidly improved, remote sensing earth observation technology has been able to provide high temporal resolution, high spatial resolution, large Remote sensing image observation data of scale range. In addition to the advantages of high spatial resolution, temporal resolution, and rich texture features, high-resolution remote sensing images also expose some problems. For example, the problem of different objects with the same spectrum due to the characteristics of the sensor and the large scale, the problem of ground object occlusion due to the sa...

AI Technical Summary

Problems solved by technology

The advantage of deep learning technology is that it can automatically extract more suitable features, and it does not completely rely on the features designed by humans. Generally, the automatically extracted features are more effective, but there is still room for improvement.

First of all, the existing deep learning technology cannot directly process large-scale images. It is necessary to cut the image into small pieces and input it to the network for segmentation and extraction. However, the cropped image of the full background area will cause misclassification because it does not contain targets. Therefore, the scene It is very necessary to identify

Secondly, for the deep convolutional neural network, the repeated use of the pooling operation reduces the feature resolution, and the prediction result through upsampling is relatively rough, and it is difficult to accurately retain the details such as the edge and position of the target area.

In addition, using a convolutional neural network with higher performance and deeper layers as the backbone network for segmentation is beneficial to feature extraction, but it will increase the number of parameters and require more labeled samples to train the network, and it is difficult to provide sufficient training for most practical applications. sample

Images