Siamese graph convolutional network for content based remote sensing image retrieval

This research work deals with the problem of content-based image retrieval (CBIR) of very high resolution (VHR)remote sensing (RS) images using the notion of a novel Siamese graph convolution network (SGCN). The GCNmodel has recently gained popularity in learning representations for irregular domain data including graphs.In the same line, we argue the effectiveness of region adjacency graph (RAG) based image representationsfor VHR RS scenes in terms of localized regions. This technique captures important scene information whichcan further aid in a better image to image correspondence. However, standard GCN features, in general, lacksdiscriminative property for fine-grained classes. These features may not be optimal for the task of CBIR inmany cases with coherent local characteristics.

Figure: A pipeline of the proposed SGCN network. The images are first segmented, and their segment wise features are extracted. From this segmented image, a RAG is formed,where each node contains their corresponding segment feature vector. These graphs are fed into the GCN layer for a pre-training, which then is fed to the SGCN layer for completetraining. The SGCN layer is trained like a Siamese twin pair with two images at a time, and the contrastive loss is minimized.

As a remedy, we propose the SGCN architecture for assessingthe similarity between a pair of graphs which can be trained with the contrastive loss function. Given the RAGrepresentations, the aim is to learn an embedding space that pulls semantically coherent images closer whilepushing dissimilar samples far apart. In order to ensure a quick response while performing the retrieval usinga given similarity measure, the embedding space is kept constrained. We implement the proposed embeddingsfor the task of CBIR for RS data on the popular UC-Merced dataset and the PatternNet dataset where improvedperformance can be observed.

Ushasi Chaudhuri, Biplab Banerjee, Avik Bhattacharya, “Siamese Graph Convolutional Network for Content Based Remote Sensing Image Retrieval”. Computer Vision and Image Understanding, vol. 184, pp. 22-30, 2019.

Tensorization of Multi-Frequency PolSAR Data for Classification Using an AutoEncoder Network

A novel tensorization framework is proposed, which utilizes the Kronecker product to combine multifrequency polarimetric synthetic aperture radar data in conjunction with an artificial neural network (ANN) for classification. The ANN comprises of two stages, where an unsupervised stochastic sampling autoencoder learns an efficient representation and a supervised feed forward network performs classification. The proposed framework is demonstrated using multifrequency (C-, L-, and P-bands) data sets collected by the AIRSAR system. The classification performance of single tensor product of dual- and triple-band combinations is evaluated. It is observed that the classification accuracy of the tensor products outperforms single, as well as, the simple augmentation of the frequency bands.

Shaunak De, Debanshu Ratha, Dikshya Ratha, A. Bhattacharya, Subhasis Chaudhuri. “Tensorization of Multi-Frequency PolSAR Data for Classification Using an AutoEncoder Network”. IEEE Geoscience and Remote Sensing Letters, vol. 15, no. 4, pp. 542-546, 2018.