Remote sensing image semantic segmentation method based on multi-modal attention and adaptive fusion

Abstract

The invention discloses a remote sensing image semantic segmentation method based on multi-modal attention and adaptive fusion, and belongs to the field of computer vision. The remote sensing image semantic segmentation method specifically comprises the following steps: 1) constructing a double-flow semantic segmentation network by using a remote sensing image multi-modal data set including a remote sensing image after data processing and a corresponding depth map; 2) respectively extracting features of different scales from the input image, and carrying out multi-layer adaptive feature fusionon the obtained features; and 3) extracting rich semantic information from the input features of the network decoder part and the encoder features by using a multi-modal attention mechanism, and paying attention to similar pixel points. The remote sensing image semantic segmentation method utilizes the multi-modal remote sensing data set to process image data, combines a double-flow network structure, fuses extracted features in a self-adaptive mode, and uses a multi-modal attention mechanism to pay attention to fused features and coded features, so as to optimize model performance.

Description

technical field [0001] The invention relates to remote sensing image processing technology, in particular to a remote sensing image semantic segmentation method based on multimodal attention and adaptive fusion. Background technique [0002] Remote sensing is a non-contact, long-distance detection technology. Generally speaking, it is used to detect and identify electromagnetic waves, infrared rays and visible light emitted or reflected by the target object itself through the sensor. With the rapid development of remote sensing technology, especially the emergence of high-resolution remote sensing images in recent years, this technology has become an important means for timely global or regional earth observation. The scale of remote sensing images is also gradually expanding, and the information provided by image content is becoming more and more abundant. [0003] The goal of image semantic segmentation is to label each pixel in an image with its corresponding class. It...

AI Technical Summary

Problems solved by technology

Learning robust feature representations from existing deep learning models poses new challenges, which is the key to improving the accuracy of semantic segmentation of remote sensing images

[0010] (2) With the development of sensor technology, other data collection of remote sensing images has not been fully utilized in semantic segmentation, and the collection process of remote sensing image datasets has noise influence

[0011] (3) The existing semantic segmentation methods have some shortcomings in the feature extraction of small targets in large-scale remote sensing images

Images