Images captured in underwater environments often suffer from color distortion, low contrast, and reduced visual quality. Most existing methods solve underwater image enhancement (UIE) by applying supervised training on synthetic images or pseudo references. However, the synthetic paired data fail to accurately replicate real-world data due to the inherent differences, and the quantity and quality of pseudo references are limited, which seriously reduces the generalization ability and performance of the model when testing on real underwater images. In contrast, unsupervised-based method is not constrained by paired data, which is more robust and potentially more promising for practical applications. Nevertheless, existing unsupervised-based methods cannot effectively constrain the network to train a model that can adapt to various degradation. Inspired by the fact that people often resolve problems from opposing but complementary perspectives, we maintain that there is implicit cooperation between the removal and generation of water layers, as they can constrain and promote each other at the same time. Based on the above analysis, a new unsupervised-based UIE framework that jointly learns water layer generation and removal based on disentangled representations is proposed. Specifically, we propose a bidirectional disentangling network in which each unidirectional network contains a loop consisting of water layer removal and generation, and restricts the image to remain consistent after a loop. Meanwhile, a novel double-order contrastive loss is proposed to improve the ability of disentanglement by utilizing the joint implicit constraint of first-order features and second-order features. Extensive experimental results demonstrate that the model outperforms the state-of-the-art methods in both qualitative and quantitative evaluation with a relatively high processing speed. The experimental results of the ablation study demonstrate the usefulness of the various components.

The landing safety and accuracy of a Mars lander will be seriously degraded due to multiple unknown disturbances or perturbations in the powered descent phase. A novel composite guidance algorithm is proposed to improve the landing performance in this paper. The presented guidance algorithm, with a composite hierarchical framework, is developed by the combination of disturbance observer-based control and multiple sliding surfaces guidance theory. The major disturbance owing to the Mars wind could be estimated through a disturbance observer and incorporated in the feed-forward compensation in the inner loop, other disturbances or perturbations could be attenuated by the multiple sliding surfaces technique in the outer loop. The composite guidance algorithm could be utilized without a pre-computed reference trajectory and it has enhanced anti-disturbance capability compared with the multiple sliding surfaces guidance law. Its global stability is verified using a Lyapunov-based approach. Monte Carlo simulation results show that the composite guidance law has a better performance on guiding a Mars lander from the point of engine ignition to the desired landing point in the presence of disturbances and perturbations.

The polarization-based attitude and heading reference system (PAHRS) consisting of inertial navigation system (INS) and polarization sensor (PS) offers an effective solution for attitude and heading determination in the case of global navigation satellite system (GNSS) signal degradation. Its performance depends largely on the state estimation accuracy. In the existing work, the Kalman filtering (KF), as a low complexity scheme, is employed to achieve the state estimation of PAHRS. However, in practice, the performance of PAHRS could be affected by weather conditions and the maneuvering state of the vehicle. The accurate noise statistics of INS and PS is often encountered, which leads to a degradation of PAHRS. To improve the adaptability and accuracy of the system, in this article, we conduct a KF flow-based deep neural networks (KFDNNs), a real-time state estimator that learns from PS and INS data to carry out Kalman filter. In the constructed KFDNNs, deep neural networks (DNNs) are inserted into the flow of the KF to learn the optimal Kalman gain from PAHRS data, which we can retain data efficiency and interpretability of the classic algorithm while circumvents the dependency of the KF on knowledge of the noise statistics. Moreover, a two-stage training strategy consisting of warm-up training stage and task-oriented training stage is presented for the KFDNNs, which mitigate the gradient explosion caused by the unstable random initialization of KFDNNs while improve the flexibility of sequence length selection. Finally, the simulation and vehicle test are carried to verify the performance of PAHRS. The experimental results confirm the KFDNNs outperforms KF-based INS/PS method especially in complex weather and maneuvering scenarios.

Underwater image enhancement (UIE) can improve the effectiveness of applications, such as underwater object detection. In UIE, synthetic images or pseudoimages are commonly used as enhancement references. However, the synthetic images cannot simulate underwater scenes, and there is interdomain difference between synthetic images and the underwater environment. Since different enhancement methods produce different effects on different semantic regions of the same image, pseudolabels lose valuable information from other enhancement methods. Moreover, different semantic information has different requirements for enhanced effectiveness. If the enhancement reference is provided on the semantic scale instead of the image scale, the advantages of various enhancement methods will be better utilized. To achieve this, this article introduces a method of semantic labels, which uses the results of different enhancement methods to label different semantic regions of the same image to build a more comprehensive reference. At the same time, enhancement images are often restricted by insufficient texture detail in advanced applications. Existing UIE methods employ the encoder–decoder architecture, which is effective in encoding broad contextual information but not reliable in preserving spatial image details. To simultaneously capture spatial information and retain image details, a two-stage spatial and detail network is proposed, which integrates U-Net and original resolution network to enhance the network’s ability to retain details. The experimental results show that the proposed method has better qualitative and quantitative evaluation results than the existing methods. Ablation studies demonstrate the effectiveness of the various components.

The camera-based polarization sensor provide an effective solution for orientation in satellite-denied environments. A key issue in polarization image processing is the solar meridian extraction (SME). However, in the presence of global noise in the polarization image, such as low exposure conditions at night, the performance of existing SME methods are significantly degraded due to the lack of effective polarization information. To alleviate this problem, some methods reduce the noise by increasing the exposure time, but at the expense of dynamics. To improve the environmental adaptability and dynamic performance of the system in the global noise environment, we conduct an unsupervised image restoration (IR) framework and a task mutual enhancement module (TMEM). The IR framework is based on disentangled representation, with each unidirectional subnetwork embedding a loop for noise removal and generation, which can remove the global noise in the angle of polarization (AoP) image. Moreover, TMEM is proposed to promote task-level information interaction between IR and SME, thereby achieving simultaneous improvement in the performance of IR and SME. Specifically, SME constrains the SM in the restored polarization images to be straight, obvious and azimuth-invariant. Conversely, the noise-free polarization images in IR improves the accuracy of SME. Extensive experimental results demonstrate that the proposed method can significantly improve the quality of AoP images and implement accurate and robust orientation under global noise environment. Under 0.2 second (reduced by 25 times) exposure at low-light condition, the dynamic orientation accuracy of the proposed method can achieve $0.74^\circ$. Ablation study demonstrate the necessity of jointly learning SME and IR and the effectiveness of TMEM. The code is available at: https://github.com/Wiggins-Carter/ROTME.