Ultrasonic through-metal communication is critical for many applications in industrial IoT (IIoT) environments but suffers from frequency-selective fading, requiring orthogonal frequency-division multiplexing (OFDM). Accurate channel estimation with minimal pilot overhead is essential for optimizing OFDM performance. Existing methods for estimating ultrasonic through-metal communication channel ignore the sparsity of the ultrasonic through-metal channel impulse response (CIR), leading to high pilot overhead, while sparse estimation algorithms fail to exploit the CIR’s structural features. Therefore, a periodic sparsity enhanced channel estimation method via improved block sparse Bayesian learning (BSBL) for ultrasonic through-metal communication is proposed, which further improves the estimation accuracy and reduces the pilot overhead by integrating three key features of CIR—the periodic occurrence of echo blocks, their exponential attenuation, and their intrinsic waveform—into the Bayesian prior. Experiments show that the proposed algorithm significantly outperforms standard BSBL in estimation accuracy while substantially reducing pilot overhead compared to conventional non-sparse techniques. This method not only enhances estimation precision but also improves spectral efficiency and link reliability in resource-constrained and harsh IIoT environments.

Deep learning methods have been studied for polarimetric synthetic aperture radar (PolSAR) land-cover classification. However, these methods often lack interpretability. This article proposes a module based on the hidden Markov model (HMM). It retains the overall structure of the neural network while correcting some easily detectable errors made by the neural network using a statistical model. This module is added after the training of the neural network is completed, and it does not require complex training once the neural network model is determined. Furthermore, the classification accuracy is improved when this module is added to various neural networks. Experiments are conducted using the Hainan dataset acquired by the Aerial Remote Sensing System of the Chinese Academy of Sciences. The experimental results demonstrate the superiority of the proposed method.

Local climate zone (LCZ) classification plays a critical role in urban environment research and has attracted extensive attention from many researchers. However, the potential of deep learning-based approaches is not yet fully explored in this field, even though neural networks continue to push the frontier for various applications. In this paper, we propose a novel multimodal multiscale Transformer network for LCZ classification by introducing multiscale patch embedding and multimodal fusion learning in Transformer architecture. The proposed multiscale patch embedding effectively captures hierarchical interrelationships of image contextual neighborhoods, and automatically learns discriminative features. And the proposed multimodal fusion learning enables the network to naturally fuse multispectral and synthetic aperture radar (SAR) data under the guidance of attention mechanism. To further improve classification accuracy, we impose semi-supervised learning to mine unlabeled image data information. Both labeled and pseudo-labeled data jointly drive our network updates. Experiments conducted on the So2Sat LCZ42, CHN15-LCZ and SouthKorea6-LCZ benchmark datasets demonstrate that our proposed approach outperforms other existing methods significantly and achieves state-of-the-art performance. In the generated LCZ maps, urban and natural classes are well distinguished, the urban structure with waters or mountains is well preserved. Finally, we also discuss the impact of the sample receptive field and sample heterogeneity on LCZ classification performance, which provides a new idea for future studies of LCZ classification.

Multi-band polarimetric synthetic aperture radar (PolSAR) has significant advantage in information extraction. However, the demanding acquisition requirement greatly prohibits its development. Typically, compared to low-frequency band PolSAR data, high-frequency band suffers more severe data insufficiency. In this paper, the authors proposed to resolve this issue by simulating Ka-band PolSAR images from X-band images. For this purpose, a conditional Generative Adversial Network (cGAN) based X-to-Ka band PolSAR image transfer network has been proposed. Adaptations in terms of preprocessing and loss function are made to the original cGAN so that it can be better adapted to PolSAR image processing. The proposed method is verified using the X- and Ka-band dataset acquired in Hainan, China by the Aerial Remote Sensing System of the Chinese Academy of Sciences. Experimental results demonstrate the feasibility of the proposed method.

Data insufficiency poses a significant challenge in Ka-band Polarimetric Synthetic Aperture Radar (PolSAR) applications. Traditional PolSAR simulation approaches fail to conquer this issue due to the intricate modeling and computational complexities induced by high-frequency. In this paper, the authors propose to mitigate this issue through neural style transfer. An X2Ka translation network is proposed to transfer X-band PolSAR images to Ka-band. Leveraging the well-verified generative network Pix2Pix, the authors adapt it to accommodate the specific discrepancies between PolSAR and optical data. Experiments are conducted on X- and Ka-band PolSAR images acquired by an Airborne PolSAR system from the Chinese Academy of Sciences. Both qualitative and quantitative evaluation results demonstrate the effectiveness of the proposed network.

In this letter, we present a novel ship detection method for polarimetric synthetic aperture radar (PolSAR) images. Generalized polarization relative entropy (GPRE) is proposed to measure the differences between the target and clutter in scattering mechanism, randomness, and intensity. Since it is difficult to derive a theoretical closed-form of the GPRE, we employ the kernel density estimation to model the distribution of the GPRE in ocean regions. Then, a constant false alarm rate (CFAR) ship detection method is proposed based on the estimated distribution. Experiments performed on both synthetic and real scene images demonstrate the effectiveness of the proposed method.

This letter proposes a novel convolutional neural network (CNN) method for dual-polarized synthetic aperture radar (SAR) ship grained classification. The network employs hybrid channel feature loss that jointly utilizes the information contained in the polarized channels (VV and VH). It is demonstrated that, by adopting the proposed CNN framework and the novel loss function, the classification performance can be efficiently improved. First, instead of the prevalently used threefold or fourfold division (container ship, oil tanker, bulk carrier, and so on), the proposed method can further divide vessels into eight accurate categories. Second, this method can not only effectively classify targets into eight categories but also its accuracy in terms of fewer category classifications surpasses existing methods. Third, the method can achieve good performance on a small training data set. Experiments conducted on the OpenSARShip data sets indicate that the proposed classification method achieves state-of-the-art results.

In this paper, we propose a novel ship detection method for polarimetric synthetic aperture radar (PolSAR) images via graph-based sparse manifold ranking. The main framework comprises a coarse-to-fine scheme. We employ the image intensity for prescreening and introduce graph-based sparse manifold ranking (GSMR) for discrimination. The distance between the ship and clutter in sparse code domain is explored. And the image elements in candidate region are ranked based on the prescreening priors and the new distance in graph labeling framework. The final detection result is produced with the ranking procedure. Experiments performed on two RADARSAT-2 images demonstrate the effectiveness and superiority of the proposed method.