Image geography marking method based on spatial cognition learning

Abstract

The invention provides an image geography marking method based on spatial cognition learning and relates to the field of deep learning and image classifying and marking. The image geography marking method comprises the steps that images are preprocessed to make a training set and a verification set meet an input format of a deep learning tool; the deep learning tool is utilized to establish a convolutional neural network, and spatial distribution of image features is learned by using a global average pooling layer; the training set is used for performing end-to-end training, so that a network automatically learns the geographic features of images, and parameters of the network are modified in the training process, so that the global average pooling layer has the high sensitivity to the geographic features; a training model is saved, and the model is verified by using a verification set; the saved training model is used for calculating a new testing object to obtain geographic marks of the images. The geographic features in the images are automatically learned from the input images, the interference to manual feature selection is avoided, and automatic and efficient feature learning is achieved in large-scale image geographic 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