Unmanned boats will encounter many static and dynamic obstacles during navigation, and only real-time obstacle sensing can ensure safe navigation and long endurance of unmanned boats. In this paper, LGCGNet is proposed to perform real-time water surface semantic segmentation on the images captured by the on-board camera. In order to ensure that the model adapted to obstacles with extremely variable scales, a local-global module is proposed in this paper. The local-global module consisted of residual dense dilated module and context-enhanced separable self-attention. Residual dense dilated module enabled the enhancement of local detail information and context-enhanced separable self-attention enabled model receptive field expansion. In addition, the sub-pixel downsampling module is used to avoid the loss of feature information to improve segmentation accuracy. Experiments on the MaSTr1325 dataset showed that LGCGNet apprpached the segmentation accuracy of state-of-the-art semantic segmentation models with only 689,000 parameters and 9.068G floating point operations per second, with an mIoU of 84.14

To balance the accuracy and speed of image fusion, a lightweight infrared and visible image fusion algorithm is proposed in this paper. Firstly, to improve the feature extraction capability, ECA attention is introduced to enhance information exchange among channels. Secondly, to improve the feature reconstruction ability and lightweight network, RFB is used to enlarge the receptive field and minimize parameters. Finally, experimental results show that the proposed method outperforms other algorithms in quality and speed. The quantitative results demonstrate that our method effectively preserves infrared features and visible details, with minimal color distortion and uniform brightness. Among them, params decreased by 59.5

Infrared and visible image fusion plays an important role in subsequent advanced visual tasks, especially for complex marine environments. To improve the real-time performance of infrared and visible fusion tasks, a lightweight residual attention deep-learning method is introduced to address the challenges associated with the huge parameter quantities and long fusion time in fusion networks. Firstly, residual blocks are designed for global feature extraction, while channel attention is used to focus on local features to fully extract and distinguish infrared features and visible details. Secondly, RFB is introduced into the feature reconstruction module to improve the network reconstruction performance by expanding the receptive field. Finally, the simple and convenient addition fusion strategy is selected to ensure the lightness of the network. According to the experimental results, our method shows remarkable improvements in image fusion effect and fusion time consumption compared to other fusion methods.

A lightweight feature fusion SSD for USV (unmanned surface vessel) platforms with limited memory is proposed in this paper. Firstly, MobileNetV2 is used to replace VGG to reduce parameter redundancy and improve speed without compromising accuracy. Secondly, a lightweight feature fusion module is utilized to improve accuracy for small targets. On the Flow dataset the average accuracy of the network proposed in this paper is improved by nearly 10

A lightweight stacked residual low-light image en-hancement network was proposed to overcome the difficulty in accurately sensing the environment under low-light conditions for unmanned surface vessels(USV).Firstly,a pyramid multi-scale pooling was introduced into feature fusion to better preserve image details.Secondly,depthwise separable convo-lution was introduced to reduce network burden and improve the image processing speed.Thirdly,a new composite loss function with color loss was designed to reduce color distor-tion.Finally,LeakyReLU activation function was used to pre-vent neuronal death.Results show that compared to the low il-lumination image enhancement network of stacked attention residual network(SARN),the proposed method improves im-age quality while accelerating image processing speed.The structural similarity and peak signal-to-noise ratio are im-proved by 3.31％and 2.08％,respectively.The model com-putation,parameter count,and single image processing time are reduced by 81.88％,75％,and 43.02％,respectively.

Image deraining ensures the visual quality of images to prompt ship detection for visual perception systems of unmanned surface vessels. However, due to the insufficiency of captured rain streaks features and global information, current image deraining methods often face the issues of rain streaks remaining and image blurring. Consider that the visual perception system captures the same useful information during rainy and hazy days, and only the way in which the image degrades is different. In addition, rainy days are usually accompanied by hazy days at the same time. In this paper, a two-stage and two-channel attention single image deraining network is proposed. Firstly, the subpixel convolution up-sampling module is introduced to increase the range of captured features and improve the image clarity. Secondly, the attention mechanism is integrated with the pyramid multi-scale pooling layer, so that the network can accumulate context information in a local to global way to avoid the loss of global information. In addition, a new composite loss function is designed, in which a regular term loss is introduced to maintain the smoothness and a perceptual loss function is employed to overcome the problem of large differences in the output of the loss function due to outliers. Extensive experimental results on both synthetic and real-world datasets demonstrate the superiority of our model in both quantitative assessments and visual quality by comparing with other state-of-the-art methods. Furthermore, the proposed deraining network is incorporated into the visual perception system and the detection accuracy of ships on rainy seas can be effectively improved.