Remote sensing image classification method, device and equipment of self-adaptive recursive incremental model

Abstract

The invention discloses a remote sensing image classification method, device and equipment of a self-adaptive recursive increment model, and the method comprises the steps: obtaining a remote sensing image of a task region, and determining a difference value between the spatial heterogeneity level of the task region and the spatial heterogeneity level corresponding to the self-adaptive recursive increment model; the adaptive recursive incremental model comprises at least one neural network hidden layer; if the difference value is greater than a preset heterogeneity threshold value, adding at least one neural network hidden layer to the adaptive recursive increment model to obtain an updated adaptive recursive increment model; according to a training sample corresponding to the task area remote sensing image, training the updated adaptive recursive increment model; and classifying the task area remote sensing image to obtain a classification result of the task area remote sensing image. According to the embodiment of the invention, the number of layers of the deep learning model can be adaptively adjusted according to different heterogeneity degrees, automatic classification of remote sensing images is realized, and disastrous forgetting and waste of computing resources are avoided.

Description

technical field [0001] The invention relates to the cross technical field of data mining and remote sensing image processing, in particular, to a remote sensing image classification method, device and equipment using an adaptive recursive incremental model. Background technique [0002] With its powerful feature self-learning and application generalization capabilities, deep learning has received a lot of research in the field of remote sensing image classification. Many intelligent interpretation models of remote sensing data based on deep neural network models have been proposed and promoted. The current deep neural network models are generally static models and require a large amount of data for long-term training. However, the models can only be applied to the current task and cannot be adapted or expanded with the complexity of the surface heterogeneity. For example, when faced with two different heterogeneous regions for remote sensing image classification, the general...

AI Technical Summary

Problems solved by technology

The current deep neural network models are generally static models and require a large amount of data for long-term training. However, the model can only be applied to the current task and cannot be adapted or expanded with the complexity of surface heterogeneity.

For example, when facing two regions with different heterogeneity for remote sensing image classification, the common method is to train separate models for different regions. This method not only wastes computing resources, but also if the trained classification model is directly applied The heterogeneous area of ​​knowledge will lead to catastrophic forgetting of knowledge, which greatly limits the application range of deep neural network models.

Images