Lightweight island multi-temporal remote sensing image change detection method based on multi-scale characteristic distillation
The invention discloses a lightweight island multi-temporal remote sensing image change detection method based on multi-scale characteristic distillation. The method comprises the following steps: 1) carrying out data processing on an island data set; 2) constructing a detection network and performi...
Gespeichert in:
Hauptverfasser: | , , , , , , |
---|---|
Format: | Patent |
Sprache: | chi ; eng |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | The invention discloses a lightweight island multi-temporal remote sensing image change detection method based on multi-scale characteristic distillation. The method comprises the following steps: 1) carrying out data processing on an island data set; 2) constructing a detection network and performing training; 3) inputting the processed island data set into a detection network; 4) the teacher network outputs a prediction result and takes the prediction result as a soft label, a real change binarization result in the island data is taken as a hard label, and the soft label and the hard label are taken together as a supervision signal to supervise the student network; 5) calculating a total loss function during supervision and carrying out gradient return; and 6) outputting a detection result. According to the lightweight island multi-temporal remote sensing image change detection method based on multi-scale characteristic distillation, a teacher and student network structure is adopted, a knowledge distillati |
---|