Yulong Qiao

Infrared moving maritime vessel segmentation is a crucial image processing task for maritime security, which is a challenging problem due to the complex backgrounds and targets with varying sizes. To address these issues, we propose an end-to-end segmentation network based on a multi-scale spatiotemporal vision transformer (ST-VT) for segmenting the moving maritime vessels in the infrared image sequence. Specifically, in the feature extraction module, we introduce a multi-scale feature encoding structure that combines a multi-scale backbone and Feature Pyramid Network technology. Then, the multi-scale deformable encoder structure and a cross-scale fusion module with the pixel decoder are proposed to generate the multi-scale spatiotemporal features. Subsequently, we employ the improved attention blocks that are the core blocks of the coarse-to-fine framework (across scales) of the prompt decoder to obtain the prompts. Finally, a multi-scale mask decoder is applied to achieve the final target segmentation. The experiments are conducted on the benchmark dataset IPATCH and our labeled dataset LAS-MassMIND. The results demonstrate that the proposed method achieves state-of-the-art performance, especially within complex backgrounds and targets of varying sizes.

This paper focuses on the application of image dehaze algorithms in highway scenarios, proposing a novel dehaze algorithm and a self-supervised transfer learning method for practical highway surveillance applications. The new lightweight dehazing network with the pyramid network structure is designed by combining the information multi-distillation network (IMDN), the channel and pixel attention module. In the deployed highway monitoring application, the self-supervised transfer learning method by proposed by integrating the pre-trained dehazing model with a dynamic target detection network. Through multiple alternating learning processes, the dehazing model continuously transfer and suitable for the current real-world application scenarios. The proposed algorithm is rigorously tested on an RTX 3090 GPU by using several public standard datasets and real-world highway datasets. The results demonstrate that the new algorithm outperforms state-of-the-art algorithms, achieving significantly higher Peak Signal-to-Noise Ratio (PSNR) and structural similarity (SSIM) on the public datasets. Furthermore, the visual quality of the dehazed images from new algorithm after transfer learning is markedly superior compared to other algorithms in the real-world highway scenarios. In terms of speed, the new algorithm exhibits faster inference speed than other comparative algorithms, achieving a frame rate 25 frames per second (FPS) for the $1920\times 1080$ real video. On the 4KID dataset, the inference speed can reach 26ms.

Traditional semantic segmentation methods have problems such as poor multi-scale feature extraction ability, weak lightweight backbone network feature extraction ability, lack of effective fusion of context information, resulting in edge segmentation errors and feature discontinuity. In this paper, a novel semantic segmentation model based on multi-layer information fusion and dual convolutional attention mechanism is proposed. In this method, SegFormer network is used as the backbone network, and multi-scale features of encoder output are fused with overlapping features. The feature extraction subnetwork is optimized by constructing the object region enhancement module, and the intermediate feature map is refined adaptively in each convolutional block of the deep network, so as to strengthen the fine extraction of multi-dimensional feature information of complex images. Dual convolutional attention module is used to fusion high-level semantic information to avoid the loss of feature information caused by up-sampling operation and the influence of introducing noise, and refine the effect of target edge segmentation. At the same time, the feature pyramid grid is proposed to process the overlapping features, obtain the context information of different scales, and enhance the semantic expression of features. Finally, the features processed by the feature pyramid grid module are combined to improve the segmentation effect. The experimental results on the public data set show that the proposed method has better performance than the existing methods, and has better segmentation effect on the object edge in the scene.

Underwater image denoising technology is of great significance in underwater operation. Underwater operations (such as offshore oil drilling, undersea tunnels, pipeline construction, underwater archaeology, biological research, and lifesaving) require stable and clear underwater images to aid analysis. Due to the scattering and absorption of light by water bodies, obtaining high-quality underwater images is a challenging task. Underwater images are prone to low contrast, low resolution and edge distortion. Therefore, it is difficult to accurately separate the effective signal when removing the underwater image noise, which leads to the image contrast reduction. Also the edge contour is not clear, and the detail loss is serious. Therefore, we propose a novel underwater image denoising method based on curved wave filter and two-dimensional variational mode decomposition. Firstly, the noisy image is decomposed by two-dimensional variational mode decomposition, and a series of modal components with different center frequencies are obtained. The effective modal components are selected by correlation coefficient and structural similarity. And the effective modal components are processed by the curve-wave filter. Finally, the filtered modal components are reconstructed to remove the noise in the image. The experimental results show that, compared with other state-of-the-art methods, the proposed method has clearer denoising results, less mean square error, and better denoising effect.

Efficient detection of pavement cracks can effectively prevent traffic accidents and reduce pavement maintenance costs. In order to overcome the complicated and uneconomical disadvantages of traditional crack detection methods, this paper introduces a pavement crack detection network based on deep learning, which can automatically detect pavement cracks and achieves excellent detection accuracy. And the network can easily use the sensors to collect data to facilitate industrial applications. In additional, considering that most cracks have slim feature, we apply the latest Transformer module in the network to improve the effect of cracks detection. Transformer has a strong ability to capture the long-range dependence of the cracks, which enables the network to learn the context information of the crack region. Furthermore, the network also utilizes some techniques to improve the ability of algorithm to detect various cracks. Our network is trained on pavement data sets containing India, the Czech Republic and Japan. It achieved F1 scores of 0.6739 and 0.6650 on two online test sets with fewer network parameters.

Maritime target tracking can be applied in the fields of intelligent marine transportation and resource protection. The spectral filter tracking method focuses on the adaptability of the local appearance changes of targets, with the graph representation. Considering the complexity and diversity of the marine environment, we propose a maritime target tracking algorithm based on kernelized spectral filter. The spectral filtering is modeled as a tracking framework based on kernel regression. According to graph signal processing and spectral graph theory, based on the description of kernel regression on graph signals, we deduce and discuss the construction of the filter, the calculation of kernel matrix, the solution of kernel regression model, and the prediction of tracking position. With the help of a nonlinear model, it can effectively improve the precision under complex tracking conditions. The experimental results on the benchmark dataset verify the effectiveness of the proposed method, especially when the target tracking is affected by waves or wakes.

When the traditional semantic segmentation model is adopted, the different feature importance of feature maps is ignored in the feature extraction stage, which results in the detail loss, and affects the segmentation effect. In this paper, we propose a BiSeNet-oriented context attention model for image semantic segmentation. In the BiSeNet, the spatial path is utilized to extract more low-level features to solve the problem of information loss in deep network layers. Context attention mechanism is used to mine high-level implied semantic features of images. Meanwhile, the focus loss is used as the loss function to improve the final segmentation effect by reducing the internal weighting. Finally, we conduct experiments on open data sets, and the results show that pixel accuracy, average pixel accuracy, and average Intersection-over-Union are greatly improved compared with other state-of-theart semantic segmentation models. It effectively improves the accuracy of feature extraction, reduces the loss of feature details, and improves the final segmentation effect.