Coastal wetland high-definition remote sensing image ground object identification method based on double-flow coding and decoding

Abstract

The invention discloses a coastal wetland high-definition remote sensing image surface feature recognition method based on double-flow coding and decoding, and the method comprises the steps: combining the feature extraction processes of a densely connected cavity convolutional network and a graph convolutional neural network to form a double-flow hybrid convolutional neural network coding model, so as to extract the characterization feature information of a coastal wetland remote sensing image surface feature; and a decoding network is further used to decode and predict the accurate and fine coastal wetland image ground object identification type according to the feature information extracted in the double-flow manner. The method comprises the following steps: firstly, carrying out preprocessing operations such as normalization, data enhancement and cutting on coastal wetland remote sensing image data; inputting the preprocessed data into a double-flow hybrid convolutional neural network coding model for feature extraction and model training; and finally, inputting a to-be-segmented test sample in the data set into the trained model, and adopting the model to achieve the purpose of accurate ground feature recognition on the coastal wetland remote sensing image.

Description

Technical field [0001] The present invention relates to the field of feature type recognition, specifically based on codec dual-stream hybrid convolutional neural network coastal wetland remote sensing image feature recognition method. Background [0002] As one of the most important ecosystems on the earth's surface, wetlands are known as the "kidneys of the earth". Coastal wetland area is wide, the internal is unreachable, the traditional artificial ground survey is time-consuming and labor-intensive, it is difficult to meet the needs of timely access to wetland cover information, and remote sensing technology has the advantages of large observation area, short cycle and low cost, which provides help for dynamic extraction of wetland information. At present, the identification of features in remote sensing images of coastal wetlands is a hot issue in the field of remote sensing. How to quickly classify massive remote sensing images is the key to accelerating the utilization rat...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to provide a dual-stream encoding and decoding method for the recognition of features in high-definition remote sensing images of coastal wetlands, and learn the characteristics of features contained in remote sensing images of coastal wetlands and the association between features of arbitrary features according to the advantages of different deep neural network streams Information, while introducing encoding and decoding methods to model and restore feature maps of feature representation features extracted by different convolutional neural networks. The problem of low precision

Images