A Novel General Semisupervised Deep Learning Framework for Classification and Regression with Remote Sensing Images

Remote sensing image analysis often involves image-level classification and/ or regression. One major problem with remote sensing data is that it is difficult to obtain abundant precise manual annotations to train fully supervised deep networks that have achieved great success in computer vision. Th...

Full description

Saved in:
Bibliographic Details
Main Authors: Chen, Zhao, Chen, Guangchen, Zhou, Feng, Yang, Bin, Wang, Lili, Liu, Qiong, Chen, Yonghang
Format: Conference Proceeding
Language:English
Subjects:
Artificial neural networks
classification
Cyclones
Estimation
infrared images
multispectral images
regression
Remote sensing
Semisupervised
Task analysis
Training
Wind speed
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Remote sensing image analysis often involves image-level classification and/ or regression. One major problem with remote sensing data is that it is difficult to obtain abundant precise manual annotations to train fully supervised deep networks that have achieved great success in computer vision. Therefore, this paper proposes a novel general semisupervised framework (GSF) which only requires a small amount of annotated samples for training. It employs a new hybrid (dis) similarity to characterize different aspects of the images and realizes label propagation while fine- tuning a deep neural network (NN). As shown by the experimental results, GSF outperforms several supervised baselines and state-of-the-art semisupervised models in classification and regression.
ISSN:2153-7003
DOI:10.1109/IGARSS39084.2020.9323932