A method of image geographic annotation based on spatial cognition learning

Abstract

The present invention requests protection of an image geographic labeling method based on spatial cognitive learning, which relates to the fields of deep learning and image classification and labeling. The method includes: preprocessing the image to make the training set and verification set conform to the input format of the deep learning tool; using the deep learning tool to construct a convolutional neural network, using the global average pooling layer to learn the spatial distribution of image features; using the training set Carry out end-to-end training to enable the network to automatically learn the geographical features of the image, and modify the parameters of the network during the training process to make the global average pooling layer highly sensitive to geographical features; save the training model and use the verification set to verify the model; use The saved training model is calculated on the new test object to obtain the geo-labeling of the image. The invention automatically learns geographical features in images from input images, avoids the interference of manual feature selection, and realizes automatic and efficient feature learning in large-scale image geographical feature learning tasks.

Description

technical field [0001] The invention belongs to the field of deep learning and image classification and labeling, and in particular relates to an image geographic labeling method based on spatial cognition learning. Background technique [0002] Image annotation, that is, to complete the automatic annotation of image content according to the image features contained in the image, is an important implementation method of image retrieval. Compared with the indexing and retrieval of low-level image content such as color, shape and texture, image annotation can realize the indexing of image content, crossing the semantic gap between the underlying features that computers rely on and human understanding of image semantics, enabling images to It is indexed and retrieved like text, and has practical application requirements in fields such as medicine, remote sensing, architecture, and shopping. [0003] With the development of Internet technology in recent years, the emergence of ...

AI Technical Summary

Problems solved by technology

However, few studies have focused on the representation and learning of geographic features in the content of geographic images themselves.

Images