Unsupervised change detection between SAR images based on hypergraphs

Robust unsupervised small area change detection from SAR imagery using deep learning

“…8, while the quantitative accuracy metrics are given in Table 3. In addition, other methods used for comparison on datasets D, E and F (Table 4, 5, 6) include saliency-guided detection with k-means clustering (SGK) (Zheng et al, 2017), stacked autoencoder and FCM with CNN (SAEFCNN) (Gong et al, 2017), saliency-guided deep neural network (SGDNN) (Geng et al, 2019), adaptive generalised likelihood ratio test (AGLRT) (Zhuang et al, 2020), PCANet with Saliency detection (SDPCANet) , spatial FCM and CNN (SFCNN) , FCM and Deep Belief Network (FDBN) (Gong et al, 2016) and hypergraph-based change detection framework (HCDF) (Wang et al, 2020), the numerical results of these methods are acquired from published articles.…”

“…DI generation: most DI generation methods adopt rectangular windows to characterise local spatial information. Such strategy will smooth out small area changes or fine details along the changed and unchanged region, where changed pixels are difficult to identify (Wang et al, 2020).…”

“…Further, an adaptive generalized likelihood ratio test (AGLRT) was developed to weaken the geometric degradation of the DI caused by integrating heterogeneous pixels (Zhuang et al, 2020). Recently, Wang et al (2020) developed a method of generating DI using irregular spatial information, in which the clear boundaries and textures were preserved by matching hypergraphs.…”

“…Thus, unsupervised approaches are developed and employed widely in this field (Li et al, 2015;Jia et al, 2016). The major components of unsupervised approaches include: 1) image pre-processing, 2) difference image (DI) generation, and 3) analysis of the DI and the classification of pixels into changed and unchanged classes (Li et al, 2020;Wang et al, 2020).…”

Multilayer Cascade Screening Strategy for Semi-Supervised Change Detection in Hyperspectral Images

“…On each HSI dataset, we take p samples from each class for first training, and the remaining pixels are used for testing. Comparison methods include CVA [22], AAD [36], TDI-GC [37], MAD [23], IR-MAD [24], ISFA [25], and DPCA [26], which employ different techniques to detect changes, such as simple algebraic operation and image transformation. The first three algorithms (i.e., CVA, AAD and TDI-GC) are unsupervised CD methods, which do not require prior information of samples.…”