For multimodal representation learning, traditional black-box approaches often fall short of extracting interpretable multilayer hidden structures, which contribute to visualize the connections between different modalities at multiple semantic levels. To extract interpretable multimodal latent representations and visualize the hierarchial semantic relationships between different modalities, based on deep topic models, we develop a novel multimodal Poisson gamma belief network (mPGBN) that tightly couples the observations of different modalities via imposing sparse connections between their modality-specific hidden layers. To alleviate the time-consuming Gibbs sampler adopted by traditional topic models in the testing stage, we construct a Weibull-based variational inference network (encoder) to directly map the observations to their latent representations, and further combine it with the mPGBN (decoder), resulting in a novel multimodal Weibull variational autoencoder (MWVAE), which is fast in out-of-sample prediction and can handle large-scale multimodal datasets. Qualitative evaluations on bimodal data consisting of image-text pairs show that the developed MWVAE can successfully extract expressive multimodal latent representations for downstream tasks like missing modality imputation and multimodal retrieval. Further extensive quantitative results demonstrate that both MWVAE and its supervised extension sMWVAE achieve state-of-the-art performance on various multimodal benchmarks.

The linear geometry distortion caused by time variant bistatic angles induces the sheared shape of the bistatic inverse synthetic aperture radar (bistatic-ISAR) image. A linear geometry distortion alleviation algorithm for space targets in bistatic-ISAR systems is presented by exploiting prior information. First, we analyze formation mathematics of linear geometry distortions in the Range Doppler (RD) domain. Second, we estimate coefficients of first-order polynomial of bistatic angles by least square error (LSE) method through exploiting the imaging geometry and orbital information of space targets. Third, we compensate the linear spatial-variant terms to restore the linear geometry distortions. Consequently, the restored bistatic-ISAR image with real shape is obtained. Simulated results of the ideal point scatterers dataset and electromagnetic numerical dataset verify the performance of the proposed algorithm.

Due to the large size of space targets, migration through resolution cells (MTRC) are induced by a rotational motion in high-resolution bistatic inverse synthetic aperture radar (Bi-ISAR) systems. The inaccurate correction of MTRC degrades the quality of Bi-ISAR images. However, it is challenging to correct the MTRC where sparse aperture data exists for Bi-ISAR systems. A joint approach of MTRC correction and sparse high-resolution imaging for Bi-ISAR systems is presented in this paper. First, a Bi-ISAR imaging sparse model-related to MTRC is established based on compress sensing (CS). Second, the target image elements and noise are modeled as the complex Laplace prior, and the Gaussian prior, respectively. Finally, the high-resolution, well-focused image is obtained by the full Bayesian inference method, without manual adjustments of unknown parameters. Simulated results verify the effectiveness and robustness of the proposed algorithm.