Optical Remote Sensing Image Retrieval Method Based on Deep Convolutional Semantic Web

Abstract

The invention discloses an optical remote sensing image retrieval method based on a deep convolutional semantic network, which mainly solves the problem of low optical remote sensing image retrieval accuracy in the prior art. The specific steps of the present invention are as follows: (1) construct a deep convolutional semantic network; (2) construct a training set; (3) adjust network parameters; (4) construct an optical remote sensing image retrieval database; (5) extract remote sensing image feature vectors; ( 6) Fusion of feature vectors of remote sensing images; (7) Establishment of feature database; (8) Retrieval of optical remote sensing images in the optical remote sensing image retrieval database. The invention builds a deep convolutional semantic network, extracts shallow features, transition features and deep features of graphics, weights and fuses the three features, realizes the complementary advantages of different levels of features, improves the expressive ability of image features, and further improves the accuracy of optical remote sensing images. Search precision.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a remote sensing image retrieval method based on a deep convolutional semantic network in the technical field of optical remote sensing image retrieval and sorting. The invention can quickly and accurately query interested scene images from a massive remote sensing image database. Background technique [0002] With the continuous development of satellite remote sensing technology, remote sensing image data has been increasingly widely used in urban planning, environmental protection, geological exploration, disaster management, military investigation and strike and other fields. However, with the rapid growth of acquired data, how to quickly and accurately retrieve relevant scene images from massive remote sensing image databases has become one of the problems that need to be solved urgently. The feature extraction method of remote sensing image is the key link of...

AI Technical Summary

Problems solved by technology

However, there are still shortcomings in this method: since the convolutional network adopts the cascaded form of convolutional layer and pooling layer, the network's semantic information extraction function for images is weak.

However, the shortcomings of this method are: only extracting a single feature of the last fully connected layer of the deep convolutional network, the ability to retrieve images with high information complexity is weak, and the hash layer binarized The quantization error caused by encoding has not been reasonably resolved, thus resulting in low retrieval accuracy

Images