Remote sensing image instance increment detection method based on sequence perception

Abstract

The invention discloses an order-perceived remote sensing image instance increment detection method. The method comprises the following steps: acquiring a remote sensing image instance; according to whether there is a real label, designing an order score calculation function to respectively evaluate uncertainty and inaccuracy of an inference result of the target detection model, and measuring an incremental learning value in a newly added instance and an incremental learning value in a historical instance; while the target detection model is finely tuned, training loss is weighted based on order scores to balance the contribution of the training data of the value difference. According to the method, streaming data with different values can be continuously learned, and data differences are comprehensively concerned; newly added instances and historical instances are sorted in a self-adaptive mode, and a unified sample weighting direction is provided for model training.

Description

technical field [0001] The invention belongs to the technical field of remote sensing images, in particular to a sequence-aware remote sensing image instance incremental detection method. Background technique [0002] The target detection task of remote sensing images, as one of the important tasks of remote sensing image interpretation, needs to detect various targets describing surface objects on remote sensing images, so as to achieve the purpose of class judgment and precise positioning of targets. The categories of interest in remote sensing image target detection tasks are mainly aircraft, buildings, vehicles, ships, etc., which play a key role in application fields such as object tracking and monitoring, activity detection, scene analysis, urban planning, and military research. [0003] The target detection algorithm based on deep learning automatically learns low-level features such as color, edge, texture, and contour of images and high-level features of macroscopic...

AI Technical Summary

Problems solved by technology

The current target detection method for incremental learning tasks lacks the scene of increasing examples in the field of remote sensing image target detection, and requires more in-depth research on the model's ability to learn new examples and remember historical examples.

Images