This article investigates the trajectory tracking control issue of autonomous underwater vehicles (AUVs) subject to dynamic uncertainties, external disturbances, and input amplitude and rate saturations. Initially, two new stable systems with fixed-time convergence are developed, and their upper bounds of settling time and convergence regions are thoroughly analyzed. Building on these systems, an enhanced fast nonsingular integral terminal sliding-mode (NITSM) surface and a new virtual control law are designed, respectively. Next, a novel saturated adaptive fuzzy fixed-time NITSM controller is proposed, circumventing the restrictions on uncertainties and input saturation in the existing results. The proposed controller ensures that the tracking error converges to a small neighborhood of the origin within a fixed time. Furthermore, to facilitate the adaptive fixed-time stability analysis, two new inequalities are established and rigorously proved. Using the two inequalities, the fixed-time stability of closed-loop systems is demonstrated by the Lyapunov's theory. Finally, representative numerical simulations validate the effectiveness of the proposed control scheme.

Currently, hyperspectral image (HSI) classification methods based on deep learning (DL) are extensively researched. However, the unique imaging characteristics of HSIs introduce a significant amount of spectral-spatial information redundancy, which creates challenges for existing methods in fully processing and utilizing this information. Furthermore, the classical patch-based data augmentation introduces heterogeneous pixels, posing a risk of losing central target information. To tackle these challenges, we propose a novel HSI classification network via spectral-spatial redundancy reduction and TokenLearner-enhanced transformer, named SSRRTL. Leveraging the inherent advantages of convolutional neural networks (CNNs) and transformer architectures, we innovatively introduce three key modules: spectral redundancy reduction and multiscale information extraction (SRRMIE) module, gated spatial feature enhancement (GSFE) module, and TokenLearner module. The SRRMIE module focuses on extracting multiscale features, reducing spectral redundancy, and preserving discriminative features of minor components through separation, feature extraction (FE), and adaptive fusion strategies. To further handle spatial information redundancy, the GSFE module employs a gated mechanism to enhance the extraction of key spatial features and suppress the expression of redundant spatial features. In addition, the TokenLearner method within the transformer architecture integrates feature tokens with central target tokens to enhance semantic feature expression and stabilize central target information, thereby improving the model’s robustness. Experimental results on multiple datasets confirm the effectiveness of SSRRTL and its superiority over other state-of-the-art methods.

Numerous hyperspectral image (HSI) super-resolution (SR) approaches have been proposed and attained remarkable performance recently. However, the enormous computational and memory costs of the deeper and heavier networks make the application challenging in the actual scene. Therefore, a lightweight but efficient local-global feature fusion module network (LGFFMN) is proposed to reduce the computational burden while achieving superior reconstruction performance. In particular, a module LFFM is proposed to utilize CNN for effective local feature extraction. Meanwhile, to exploit global information and avoid the computational burden of traditional Transformers, a multi-scale representation-based feature modulation mechanism is adopted to build a Vision Transformer (ViT)-like block GFFM. In order to reduce the high dimension of HSI while utilizing the higher similarity among adjacent bands, the overall processing manner is group by group. Experiments demonstrate that our LGFFMN achieves a satisfactory balance in SR reconstruction performance and efficiency metrics and is highly competitive compared with other state-of-the-art HSI SR methods.

Crack detection, as one of the structural health monitoring tasks, plays a crucial role in ensuring the reliability of transportation infrastructure. And the convenience of Unmanned Aerial Vehicles (UAVs) has led to their widespread utilization as carriers for crack detection task. However, the motion and vibration of the carrier inevitably lead to image blurring, consequently degrading the performance of subsequent crack detection. While existing image deblurring methods exhibit commendable performance in feature-rich real-world scenarios, their efficacy tends to diminish in feature-limited environments, such as the surfaces of concrete structures and pavements. In response to this challenge, we propose a novel deblurring network, named Motion and Event-guided Cross-modal Fusion Network (MECFNet), by incorporating motion information and event information as constraints to guide the image reconstruction process. For one thing, we devise two novel representation methods: motion field and accumulated event to characterize these two types of information. For another, we introduce a cross-modal transformer module to incorporate these two new modalities of information into the image features. Comparison experiments and ablation experiments have demonstrated the effectiveness of the proposed network and its constituent components. Furthermore, the performance of crack detection achieves a significant enhancement after the reconstruction of blurry images.

For the predominant classification performance, the convolutional neural network (CNN) is becoming quite popular in hyperspectral image (HSI) classification. However, quite a few parameters must be updated in the training procedure. The overfitting problem is also exacerbated by HSI’s high dimensionality and limited training samples. Therefore, an end-to-end model DWTCNN that deeply integrates 3-D discrete wavelet transform (DWT) and CNN is proposed. The 3-D DWT is introduced for the intrinsic feature collection in multi-resolution, and the 3-D DWT modulated kernel with predefined parameters will not increase the trainable parameters of DWTCNN. The impact of 3D-DWT at different scale levels on classification performance is also studied. The amalgamation of 3-D DWT reduces the feature extraction burden of 3-D CNN while increasing the feature extraction capacity of the network. With a compact and light structure, our model is easier to train with less memory and limited training samples. Compared with other methods, our model has a considerable advantage in computational efficiency while maintaining good classification accuracy and clear physical meaning interpretability. Experimental results on three public datasets also demonstrate the superiority of our approach in the training process and classification performance.

Recently, benefiting from the rapid evolution of Deep Learning technology, hyperspectral image (HSI) classification performance is considerably improved. However, the traditional convolutional analysis resulted in too many learnable parameters of the network, and lots of training samples are required to train the network to avoid the overfitting problem. On the other hand, the 3-D discrete wavelet transform can effectively extract both spatial and spectral information, maintaining robust feature representation capabilities and reducing the computational burden of CNN simultaneously. In this letter, the 3-D discrete wavelet transform is adopted to perform preprocessing operations on HSI. Then, the 3D CNN that integrates dense connections attaches great importance to the reuse of features. Furthermore, our approach significantly reduces network parameters, relieves the training complexity of the network, alleviates the training process from overfitting, and improves classification performance with limited training samples. Three benchmark datasets were used for very rigorous HSI classification experiments to test the performance of the proposed approach. The results demonstrate the superiority of the proposed approach compared with existing state-of-the-art (SOTA) HSI classification approaches. The source code is publicly available at https://github.com/xujingran/DWTDENSE.

In this article,we study a novel fixed-time tracking controller for a class of degree-of-freedom nonlinear systems with mismatched and matched perturbations. A modified fixed-time stable system (FTSS) is first proposed, which has an advantage in convergence rate over the existing results. Moreover, an adaptive version of fixed-time disturbance observer (DO) is established to eliminate the requirement for prior knowledge of the perturbation bounds. Based on the aforementioned FTSS and DO, a new fixed-time terminal sliding-mode surface with singularity circumvention is constructed. Then, a novel composite tracking controller is designed, which can guarantee the tracking errors to converge to zero within the fixed time. The stabilities of both the DO and the resulting closed-loop systems are rigorously proved by the Lyapunov theorem. Simulation results of two representative examples demonstrate the efficiency of the proposed approach.

Crack detection plays a crucial role in structural health monitoring tasks to ensure the reliability of the transportation infrastructures. However, the automatic detection of cracks remains a challenging task due to the complicated background. Especially, tiny crack detection should be attached importance because of its weak feature and background interference. Therefore, an end-to-end network Feature Fusion Encoder Decoder Network (FFEDN) with two novel modules is proposed to improve the crack detection accuracy. For one thing, the representation capability for tiny cracks is enhanced by introducing the attention mechanism, which redistributes and fuses different features of both the encoder and the decoder. For another, because high-level feature contains less interference, a shape semantic prior module is developed to learn the shape prior map that provides the rough shape and location information of cracks. This map is fed into the lower-level feature and helps it focus on crack areas, thereby suppressing background interference. To demonstrate the effectiveness of the proposed network, several experiments are implemented on three publicly available crack datasets. Compared with state-of-the-art crack detection methods, the novel network shows better performance on all the six evaluation metrics.

This study presents a foreground detection method capable of robustly estimating the background under the presence of dynamic effects. The key contribution of this study is the use of the dynamic principal component analysis to model the serial correlation between successive frames and construct a robust pixel‐based background model. The frames are normalised in hue, saturation and value colour space to reduce the effect of illumination changes. To restrict the background model, kernel density estimation is used to identify the distribution of the background time‐lagged data matrix and then confidence interval limits are used to determine the corresponding detection thresholds. The foreground is detected using background subtraction. This method is tested on several common sequences such as CDnet 2014, ETSI 2014 and MULTIVISION 2013. The authors also hold comparisons based on quantitative metrics with several state‐of‐the‐art methods. Experimental results show that their method outperforms some state‐of‐the‐art methods and has comparable performance with some depth‐based methods.

The uncertainty principle and cross-term can lead to blur, fake signal components and energy oscillation in time-frequency distribution, deteriorate the results of signal tracking, radar/sonar imaging and parameter estimation. Hence in this paper, we propose a time-frequency image enhancement method based on convolutional neural networks for clearer instantaneous frequency curve. The training data are generated by a frequency modulation signal generator, and then an end-to-end training is performed between Wigner-Ville distributions and time-frequency images. Our networks not only extract underlying features of Wigner-Ville distribution, but also understand the semantic of instantaneous frequency curve and use the priori knowledge of the modulation mode. Therefore, it can correctly recognize and eliminate the cross-terms, and transform the Wigner-Ville distribution to an image that can accurate represent the instantaneous frequency curve. The method is tested by three kinds of frequency modulation signals randomly with Gaussian noise. The results show that it can work properly in most cases and has the generalization ability of multi-component signals.