Multi-task processing system and method for two-phase remote sensing images

Abstract

A multi-task processing system and method for two-phase remote sensing images, including: a two-way branch feature extraction module, a pyramid fusion module, a semantic guidance module, a feature aggregation module, a two-way attention mechanism module, a boundary promotion module and an upsampling module , the present invention extracts the features of the two phases of remote sensing images by constructing a two-way network, designs the information flow module to guide the high-level semantic information to guide the underlying spatial information to perform feature learning, and effectively fuses the feature maps of the two phases of images through the feature aggregation module to fully capture the two phases Relationships and differences between images. Subsequently, the expression ability of feature information is enhanced through the attention mechanism module, so as to realize multi-task intelligent processing.

Description

technical field [0001] The invention relates to a technology in the field of image processing, in particular to a multi-task processing system and method for two-phase remote sensing images. Background technique [0002] With the successful application of deep learning in remote sensing image processing, multi-task intelligent processing of remote sensing images has become a research hotspot. Classification, segmentation, change detection, etc. Existing deep networks are introduced into two-phase remote sensing image change area change category information extraction, mainly including Fully Convolutional-Earlyfusion (FC-EF) based on semantic segmentation, deepSiamese multi-scale convolutional network (DSMS-CN), etc., based on target detection. Faster R-CNN, and AggregationNet based on two-way network and tile-level detection methods. However, in addition to the changing area of ​​the two-phase remote sensing images, the invariant area also contains the core elements of int...

AI Technical Summary

Problems solved by technology

[0003] The defect of the existing technology is that when completing multi-tasks, it is necessary to send two phases of images to the model for classification, and compare the classification results, which greatly increases the time of the testing phase

On the other hand, this method analyzes the two-phase images as two separate objects, and cannot take advantage of the differences and correlations between the two-phase image features, so the implementation effect is poor.

Images