As autonomous underwater vehicles (AUVs) expand into complex marine environments, robust integrated navigation becomes critical for long-endurance missions. However, time-varying measurement noise induced by complex underwater conditions significantly degrades navigation performance. To address this challenge, this paper proposes a physics-guided adaptive filtering framework, termed the Attention-based LSTM-UKF (A-LUKF). By embedding an attention-based Long Short-Term Memory (LSTM) network into an error-state Unscented Kalman Filter (ES-UKF) framework, the proposed method enables online estimation and adaptive regulation of measurement noise. Furthermore, a differentiable UKF-informed loss function is introduced to enable label-free learning of noise parameters while preserving the physical consistency of the filtering process. Real-world lake experiments using an INS/PC/DVL-equipped AUV demonstrate that the proposed A-LUKF improved robustness and estimation accuracy under challenging conditions such as DVL beam loss and severe optical field disturbances, outperforming representative adaptive filtering methods with approximately 30%-70% improvement in key navigation states.

The spatial cognitive neural circuit comprising the mammalian hippocampal (HPC) and entorhinal cortex (EC) constitutes the neurobiological foundation that enables organisms to achieve autonomous navigation in complex environments. Brain-inspired navigation technology, as an innovative paradigm for addressing the challenge of autonomous robot localization in complex environments, substantially improves the adaptability of navigation systems in unstructured and dynamic environments through the emulation of mammalian spatial cognitive neural mechanisms. To address the limitations in adaptability and energy consumption faced by current robot navigation systems in unstructured environments, brain-inspired navigation technology provides an innovative solution for the development of autonomous intelligent systems through the emulation of mammalian spatial cognitive mechanisms. Therefore, this review provides a systematic examination and analysis of the neural dynamic properties and information encoding mechanisms of spatial representation units, such as grid cells and place cells, within the EC-HPC neural circuit. Second, we conducted an in-depth analysis of the spatial cognitive computing model inspired by the EC-HPC circuit and reviewed its application progress in robot navigation systems. Finally, the current problems and challenges confronting brain-inspired navigation technology are analyzed, and potential future research directions related to the cross-disciplinary integration of neuroscience and artificial intelligence are discussed.

Due to the increasing security requirements of portable devices, the audio-face cross-modal speaker verification task has gained significant attention. To enhance the robustness of audio-face fusion methods under challenging conditions, this paper proposes a novel audio-face fusion speaker verification method. This method comprises four components: an audio-net, a face-net, a cross-modal fusion strategy, and a metric learning mechanism. The audio-net employs convolution neural network–transformer (CNN-Transformer) architecture, enabling the interactive utilization of global and local feature information. The cross-modal fusion strategy incorporates an adaptive feature fusion mechanism that dynamically adjusts the information flow distribution across modalities. The metric learning mechanism adopts an enhanced adaptive angular margin loss function, which dynamically modulates the gradient of the loss. The proposed method was trained on the Zhvoice, Voxceleb1, and Cnceleb_v2 speech datasets and the CASIA-WebFace face dataset, and evaluated on their corresponding test sets. Relative to the audio-facce direct fusion model ECAPA-TDNN+ResNet50, the proposed model achieves EER improvements of 87.8

Autonomous navigation is a key technology for unmanned motion platforms to perform their tasks smoothly. The current approaches for daytime polarization navigation have been extensively researched. However, the polarization light intensity is the fundamental information within the polarization image, and the light intensity at night is 6-8 orders of magnitude lower than that during the day, which increase the noise and the loss of local polarization information due to occlusion, resulting in a significant decrease in the polarization orientation accuracy. Aimed at the problem, a bio-inspired model is introduced to denoise and enhance weak nighttime polarization patterns. Further, to address the issue of outlier interference in the occluded environment during practical application, a fast-fitting method of the solar meridian based on the anti-symmetric distribution of the polarization angle adjusted by Proportional and Differential (PD) control is proposed. The experimental results show that the method proposed in this paper achieves a dynamic orientation error Root Mean Square Error (RMSE) of 0.7° in the weak polarization mode at night and in the presence of local occlusion. The proposed method has strong robustness under weak polarization occlusion at night, and the orientation accuracy is improved by 97% and 80% in comparison to the least squares method, which provides a new method for polarization navigation at night. This effectively improves the robustness and environmental applicability of the bionic polarization compass for nighttime applications.

To address the nonlinear, time-varying, and short-memory characteristics of magnetic interference during aggressive multirotor Unmanned Aerial Vehicle (UAV) maneuvers, as well as the limited dynamic representation capability of the conventional Tolles–Lawson (TL) model and the instability and overfitting introduced by feature expansion, this paper proposes an enhanced aeromagnetic compensation method, the Enhanced Compensation Tolles–Lawson with Fast Sparse Regularized Estimation via Splitting (ECTL-FRES). Inspired by Volterra-series memory modeling, spatiotemporal coupling terms between the current attitude and first-order historical attitudes are incorporated into the TL framework to construct an Enhanced Compensation Tolles–Lawson (ECTL) model. Furthermore, a Fast Sparse Regularized Estimation via Splitting (FRES) algorithm is developed by integrating elastic-net regularization with operator-splitting optimization, enabling stable coefficient estimation and redundant-feature sparsification under ill-conditioned feature matrices. Experimental results on real flight data show that ECTL-FRES reduces the post-compensation residual standard deviation (STD) by 73.2% and 62.7% compared with TL-LS (Tolles–Lawson Least Squares) and ECTL-PRSM (Enhanced Compensation Tolles–Lawson with Peaceman–Rachford Splitting Method) , respectively, while achieving an average improvement ratio (IR) of 7.02. These results demonstrate the effectiveness, robustness, and engineering applicability of the proposed method for dynamic UAV aeromagnetic compensation.

In the field of visual attitude estimation for unmanned aerial vehicles, precise and reliable horizontal pitch/roll angle measurement is crucial for applications such as autonomous navigation, environmental monitoring, and search-and-rescue missions. However, relying on a single visual information often results in susceptibility to interference and limited accuracy. To address these challenges, we propose the cascaded spatiotemporal fusion attitude estimation architecture. Initial pitch/roll angle estimates are independently obtained through horizon detection and optical flow estimation. Subsequently, a forward-backward optical flow optimization algorithm based on gradient descent is proposed to address the problem of degraded accuracy in conventional optical flow at object boundaries. Finally, in light of the disparities in the noise characteristics of the two types of observation data, an adaptive sequential Kalman filtering algorithm is proposed. This algorithm incorporates a two-stage updating mechanism and dynamically adjusts the measurement noise covariance matrix through an adaptive factor to efficiently fuse the results of horizon and optical flow. The experimental results demonstrate that the suggested method significantly enhances the accuracy of pitch/roll angle measurements compared to the single horizon or optical flow methods and meets the requirements for the stability and reliability criteria for visionbased UAV attitude estimation.

The stable production of crops such as corn, wheat, soybeans, and canola is increasingly threatened by widespread pest infestations. Conventional manual pest surveys are hampered by low operational efficiency, subjective assessment bias, and delayed feedback, thereby impeding their ability to satisfy the demands of precision agriculture. To address these challenges, we proposes an intelligent pest detection framework based on EfficientNet and Feature Pyramid Network (FPN) for fast and accurate field pest identification. EfficientNet is adopted as the lightweight attention-embedded backbone to extract hierarchical features, and multi-scale detection plus hierarchical FPN fusion are integrated to improve recognition performance for tiny, inconspicuous pests. The experimental results on 37 common pest species in field crops showed that the proposed model achieves a mean average precision at Intersection-over-Union (IoU) threshold 0.5 (mAP@0.5) of 98.89%, 1.57% average recognition error rate, and with an average inference time of merely 0.048 s per image, balancing outstanding detection accuracy and real-time performance. Furthermore, this approach delivers a lightweight, reliable, and automated monitoring solution for field pest surveillance, thereby facilitating data-driven, precise pest management and advancing the practice of sustainable, green precision agriculture.

The biomimetic polarization compass, immune to electromagnetic interference and free from cumulative errors, is highly promising for autonomous navigation. However, carrier oscillations or camera vibrations often degrade polarized images, thereby affecting the accuracy of azimuth angle computation. To address this, the paper proposes a cascaded dual-stage deblurring network leveraging atmospheric polarization characteristics. In the initial phase, a Stokes and Polarized Image Feature Fusion (SPIFF) module is utilized to augment the representation of polarization and spatial features, thereby facilitating the estimation of unpolarized images. In the subsequent phase, the sky polarization pattern is reconstructed by integrating atmospheric polarization constraints into the loss function and leveraging the unpolarized images generated from the first phase. Results of UAV flight experiments show that the processed images have more accurate polarization patterns, the fitted solar meridian is closest to the ideal condition, and the root mean square error of the computed azimuth angle is reduced from 14.9629 degrees to 3.8168 degrees. This effectively suppresses the azimuth errors caused by motion blur in polarized images and enhances the robustness and reliability of the polarization navigation system in dynamic environments. The code is available at https://github.com/xzj666zb/Polarized_Deblur.

We developed an intelligent innovative orientation method to improve the accuracy of polarization compasses in harsh conditions: weak skylight polarization patterns resulting from unfavorable weather conditions (e.g., haze, sandstorms) or locally destroyed skylight polarization conditions caused by occlusions (e.g., buildings, trees). First, the skylight polarization status was determined with the degree of linear polarization threshold analysis method and a bionic polarization enhancement sensing model was constructed to simulate the enhanced perception mechanism identified in the Syrphidae visual neural pathway, highly efficient in dark or weakly illuminated environments. The bionic model successfully enhanced the information content extracted from weak polarization patterns. Second, polarization pixel interferences, caused by occlusions under locally destroyed skylight polarization conditions, were removed with a convolutional neural network for image segmentation and the sky area of interest was identified. Finally, the incomplete angle of polarization map derived after image segmentation was fitted using our optimized adaptive antisymmetric ring algorithm. On the basis of the strong angle-of-polarization antisymmetry along the solar meridian, information extracted from the sparse and irregular polarization pixels was analyzed to derive a high-accuracy polarization orientation solution. The whole method intelligently realizes pattern analysis and deep learning intelligent processing, efficiently rotates to manage polarization disorientation. The experimental results demonstrated the performance of the proposed method in compensating for reduced orientation accuracy under degraded polarization conditions, its robustness against perturbations, and its beneficial impact on the environmental adaptability of bionic polarization compasses.

Speaker recognition is essential in smart voice applications for personal identification. Current state-of-the-art techniques primarily focus on ideal acoustic conditions. However, the traditional spectrogram struggles to differentiate between noise, reverberation, and speech. To overcome this challenge, MFCC can be replaced with the output from a self-supervised learning model. This study introduces a TDNN enhanced with a pre-trained model for robust performance in noisy and reverberant environments, referred to as PNR-TDNN. The PNR-TDNN employs HuBERT as its backbone, while the TDNN is an improved ECAPA-TDNN. The pre-trained model employs the Canopy/Mini Batch k-means++ strategy. In the TDNN architecture, several enhancements are implemented, including a cross-channel fusion mechanism based on Res2Net. Additionally, a non-average attention mechanism is applied to the pooling operation, focusing on the weight information of each channel within the Squeeze-and-Excitation Net. Furthermore, the contribution of individual channels to the pooling of time-domain frames is enhanced by substituting attentive statistics with multi-head attention statistics. Validated by zhvoice in noisy conditions, the minimized PNR-TDNN demonstrates a 5.19% improvement in EER compared to CAM++. In more challenging environments with noise and reverberation, the minimized PNR-TDNN further improves EER by 3.71% and 9.6%, respectively, and MinDCF by 3.14% and 3.77%, respectively. The proposed method has also been validated on the VoxCeleb1 and cn-celeb_v2 datasets, representing a significant breakthrough in the field of speaker recognition under challenging conditions. This advancement is particularly crucial for enhancing safety and protecting personal identification in voice-enabled microphone applications.

To rectify significant heading calculation errors in polarized light navigation for unmanned aerial vehicles (UAVs) under tilted states, this paper proposes a method for compensating horizontal attitude angles based on horizon detection. First, a defogging enhancement algorithm that integrates Retinex theory with dark channel prior is adopted to improve image quality in low-illumination and hazy environments. Second, a dynamic threshold segmentation method in the HSV color space (Hue, Saturation, and Value) is proposed for robust horizon region extraction, combined with an improved adaptive bilateral filtering Canny operator for edge detection, aimed at balancing detail preservation and noise suppression. Then, the progressive probabilistic Hough transform is used to efficiently extract parameters of the horizon line. The calculated horizontal attitude angles are utilized to convert the body frame to the navigation frame, achieving compensation for polarization orientation errors. Onboard experiments demonstrate that the horizontal attitude angle estimation error remains within 0.3°, and the heading accuracy after compensation is improved by approximately 77.4% relative to uncompensated heading accuracy, thereby validating the effectiveness of the proposed algorithm.

Aiming at the issue of time-varying measurement noise with heavy-tailed characteristics and outliers generated by the polarization compass (PC) in the micro-electro-mechanical system–inertial navigation system (MEMS-INS) and PC-integrated navigation system when it is subject to internal and external disturbances, an improved Variational Bayesian Innovation Saturation Robust Adaptive Kalman filter (VISKF) algorithm is proposed. This algorithm utilizes the variational Bayesian (VB) method based on Student’s t-distribution (STD) to approximately calculate the statistical characteristics of the time-varying measurement noise of the PC, thereby obtaining more accurate measurement noise statistical parameters. Additionally, the algorithm introduces an innovation saturation function and proposes an adaptive update strategy for the saturation boundary. It mitigates the problem of innovation value divergence in PC caused by outliers through a two-layer structure that can track the changes in the innovation value to adaptively adjust the saturation boundary. To verify the effectiveness of the algorithm, static and dynamic experiments were conducted on an unmanned vehicle. The experimental results show that compared with adaptive Kalman filter (AKF), variational Bayesian robust adaptive Kalman filter (VBRAKF), and innovation saturate robust adaptive Kalman filter (ISRAKF), the proposed algorithm improves the dynamic orientation accuracy by 76.89%, 67.23%, and 84.45%, respectively. Moreover, compared with other similar target algorithms, the proposed algorithm also has obvious advantages. Therefore, this method can significantly improve the navigation accuracy and robustness of the INS/PC integrated navigation system in complex environments.

In the process of using binocular vision for ranging, target detection and image matching are the key to the ranging process. To address the problems of low target detection accuracy and high distance ranging error in traditional binocular ranging methods, this paper proposes an improved binocular vision ranging algorithm based on YOLOv5. First, the binocular camera is calibrated by the checkerboard calibration method, and the imaging plane of the binocular stereo vision is corrected to the ideal structure by the epipolar correction algorithm. Then, the target is detected by the improved YOLOv5 algorithm. This method uses the SimOTA label allocation strategy to further reduce the training time and computational complexity of the model, and introduces LEIOU to solve the problem of the unclear definition of the length–width ratio in the original LCIOU , further improving the speed and accuracy of convergence. Moreover, focal loss is added to compensate for the imbalanced contribution of high- and low-quality samples in the gradient. Next, using the improved multi-scale stereo matching algorithm, the speed of the matching algorithm in large images is enhanced. After the initial matching point pairs have been obtained, the quadratic surface fitting method is used to obtain the sub-pixel disparity. The depth value of the target center point is obtained by conversion from the two-dimensional pixel coordinate system to the three-dimensional space coordinate system. A ranging experiment was carried out in the range of 20-200 m. The MAE index of the ranging result of the proposed method is only 2.85 m, which verifies the effectiveness of the improved algorithm in both its theoretical and experimental aspects.

Inertial navigation systems experience error accumulation over time, leading to the use of integrated navigation as a classical solution to mitigate inertial drift. This provides a novel approach to navigation and positioning by using the combined advantages of inertial and geomagnetic navigation systems. However, inertial/geomagnetic navigation is affected by significant magnetic interference in practical scenarios, resulting in reduced navigation accuracy. This research introduces a new neural network-assisted integrated inertial–geomagnetic navigation method (IM-NN), and utilizes the adaptive cubature Kalman filter to integrate attitude information from geomagnetism and inertial sensors. A model was created utilizing a Long Short-Term Memory Network (LSTM) to represent the relationship between specific force, angular velocity, and integrated navigation attitude information. The dynamics were estimated based on current and previous Inertial Measurement Unit (IMU) data using IM-NN. This study demonstrated that the method effectively corrected inertial accumulation errors and mitigated geomagnetic disruption, resulting in a more accurate and dependable navigation solution in environments with geomagnetic rejection compared to conventional single inertial navigation methods.

Because of their complementary characteristics, intensity images and polarization images are often fused to produce information-rich images. However, the polarization characteristics are easily affected by the object’s environment, and the image fusion process may lose important information. In this paper, we propose an unsupervised end-to-end network framework based on a CNN for intensity images and degree of linear polarization images. First, we construct our own polarization dataset to solve the limitations of the training dataset; a hybrid loss function is designed to form an unsupervised learning process; and a Laplace operator enhancement layer is introduced into the network to further improve the quality of the fused images. Subjective and objective comparison experiments prove that the proposed fusion network is visually superior to several classical fusion methods.

Autonomous orientation technology has major engineering significance for intelligent transportation systems (ITS) especially for the intelligent vehicle. The sky polarization characteristics offer a wealth of navigation data. At present, the most advanced polarization navigators can output this navigation information and do not require complex optical structures. However, these current navigation methods are limited severely by the sky conditions, and it is difficult to achieve orientation accurately under interference from reflected light. Here, we report a sky recognition algorithm based on a region prior approach to reduce the influence of the reflected light. Furthermore, to solve the sun fuzzy problem, a morphometric template matching in transform domain (MTMTD) strategy is proposed based on the angle of polarization (AOP). The application scope and practicability of the method are improved effectively by using all the effective pixels as a navigation unit. In addition, this method turns these pixels into curves, which means that even when only one pixel is observed, the sun fuzzy problem can be solved exactly. Our results efficiently verify the feasibility of the proposed strategy, which may provide an interesting solution for heading measurement of intelligent vehicle.

Autonomous velocity measurement technology based on optical flow plays an important role in applications of the Internet of Things. However, robust velocity measurement results need to provide a robust optical flow field and distance to the ground, especially for complex ground scenes. To solve this problem, the present paper proposes a visual velocity measurement method based on a 3 × 3 camera-array multi-aperture biomimetic compound-eye imaging system. The multi-aperture optical flow field is first obtained from compound-eye through multi-scale analysis and Bayes threshold processing. The real distance to the ground is then obtained by extracting the disparity information of nine apertures for adaptive depth estimation. The velocity measurement error is less than 0.6m/s in low-altitude (8 m and 13 m) flight scenarios with multiple obstacles.

How to obtain reliable vehicle heading information in the event of GNSS rejection is the key part of the current navigation scheme. In this paper, a vehicle-mounted polarization orientation strategy for sloping roads and occlusion is investigated. For inclined road circumstances, a polarization orientation scheme with tilt compensation ability is designed which is not limited to the field of view of the image polarization compass, and a vision modulation model is provided. On this premise, a solar azimuth region orientation (SAO) strategy is developed, which lowers the measurement error caused by occlusion, and improves the image polarization compass's robustness under complex road conditions. Furthermore, outdoor experiments are delivered to testify the effectiveness of the method. In the vehicle experiment, the root mean square error (RMSE) is reduced by more than 82% when compared to the typical polarization orientation technique.