Remote sensing image classification method suitable for multi-task iterative learning and memory

Abstract

The invention discloses a remote sensing image classification method suitable for multi-task iterative learning and memory. The method comprises the following steps of: initializing previous parameters, fixed weights and temporary weights of a model classifier, and predicting performance by utilizing the parameter weights; calculating the sensitivity of each parameter in the model by utilizing thetraining data of a first task; when the model learns a new task, training the model by using a modified loss function, and learning to obtain previous parameters of the classifier; expanding new neurons in the classifier to learn new classes in the new task, reinitializing the temporary weight of the classifier, learning to obtain the temporary weight of the classifier, and predicting the performance by utilizing the parameter weights; calculating a sensitivity matrix of each parameter in the model by utilizing the training data of the new task, and calculating a sensitivity matrix of the parameters added in the loss function trained as the next task; repeating the steps whenever a new task comes and is trained; and using the trained model classifier to classify a remote sensing image.

Description

technical field [0001] The invention relates to the technical field of remote sensing image processing and recognition, in particular to a remote sensing image classification method suitable for multi-task iterative learning and memory. Background technique [0002] In the current era of rapid development of big data and artificial intelligence, in the face of massive data that is constantly updated and iterated, the deep learning model also needs to be continuously learned and updated, and the model must be constantly adjusted to meet the current needs of people. However, once the existing deep learning model is trained on a specific task, the model can only be used for the prediction of the task. Once the model continues to learn new tasks, it will appear on the previously learned tasks. The phenomenon of catastrophic forgetting, where the model fails to maintain performance on old tasks. The method of mixing all the data together and retraining the model every time not o...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the object of the present invention is to provide a remote sensing image classification method suitable for multi-task iterative learning and memory. The method is based on the combination of parameter sensitivity and structure, and can effectively overcome the catastrophic problems existing in the deep learning model. The problem of forgetting is suitable for solving the remote sensing image classification problem of multi-task iterative learning and memory

Images