Supervised deep learning applied for change detection using multispectral remote sensing images

The temporal resolution of the remotely sensed images is improving. Therefore, more attention is sought for the change detection applications. The approach of deep learning is being tried on remote sensing datasets. 

Here are a few supervised deep learning approaches that are applied for change detection applications using multispectral remote sensing images:

Siamese convolutional neural network

– Post-processing is not required

– Skip connections can be used to increase the spatial accuracy

– 500 times faster architecture than earlier approaches

Recurrent convolutional neural network 

– A unique network design combining both CNN and RNN

– Have the capability to extract information from all the three aspects – spectral, spatial and temporal

End-to-end spectral-spatial joint learning network (SSJLN) 

– Made up of three components: spectral-spatial joint representation, feature fusion, and discrimination learning

Bilateral convolutional neural network

– Training of model using two symmetric CNNs 

– Combined bilinear features were obtained from output feature maps

– Softmax classifier to produce change detection results

Reference:

Shafique, A., Cao, G., Khan, Z., Asad, M., & Aslam, M. (2022). Deep Learning-Based Change Detection in Remote Sensing Images: A Review. Remote Sensing, 14(4), 871.