Attention mechanisms have demonstrated great potential in improving the performance of deep convolutional neural networks (CNNs). However, many existing methods dedicate to developing channel or spatial attention modules for CNNs with lots of parameters, and complex attention modules inevitably affect the performance of CNNs. During our experiments of embedding Convolutional Block Attention Module (CBAM) in light-weight model YOLOv5s, CBAM does influence the speed and increase model complexity while reduce the average precision, but Squeeze-and-Excitation (SE) has a positive impact in the model as part of CBAM. To replace the spatial attention module in CBAM and offer a suitable scheme of channel and spatial attention modules, this paper proposes one Spatio-temporal Sharpening Attention Mechanism (SSAM), which sequentially infers intermediate maps along channel attention module and Sharpening Spatial Attention (SSA) module. By introducing sharpening filter in spatial attention module, we propose SSA module with low complexity. We try to find a scheme to combine our SSA module with SE module or Efficient Channel Attention (ECA) module and show best improvement in models such as YOLOv5s and YOLOv3-tiny. Therefore, we perform various replacement experiments and offer one best scheme that is to embed channel attention modules in backbone and neck of the model and integrate SSAM into YOLO head. We verify the positive effect of our SSAM on two general object detection datasets VOC2012 and MS COCO2017. One for obtaining a suitable scheme and the other for proving the versatility of our method in complex scenes. Experimental results on the two datasets show obvious promotion in terms of average precision and detection performance, which demonstrates the usefulness of our SSAM in light-weight YOLO models. Furthermore, visualization results also show the advantage of enhancing positioning ability with our SSAM.

One open issue of target detection for synthetic aperture radar (SAR) images is the capture effect from the clutter edge and the interfering outliers, including surrounding targets in the multitarget environment, sidelobes, and ghosts. To address this issue, a superpixel-level constant false-alarm rate (CFAR) detector is proposed based on the truncated Gamma statistics for the multilook intensity SAR data. Superpixel segmentation serves as a preprocessing procedure to divide the SAR image into meaningful patches. By automatic clutter truncation in the superpixel-level background clutter window, the real clutter samples are preserved. The experimental results with real SAR images demonstrate that the proposed method achieves better goodness-of-fit performance for the real clutter background with outlier exclusion, yielding a higher target detection rate in the multitarget environments.

In RGB-infrared aerial image object detection, fully utilizing the advantages of both RGB and infrared images for effective detection is a key challenge in this field. In response to the challenges outlined, an RGB-infrared multi-branch progressive feature fusion method for object detection in aerial images is proposed. Specifically, since accurate detection of objects in aerial images usually requires extensive context information, considering the connection between objects and backgrounds, the global-local synergistic attention is constructed. To mitigate the impact of feature domain differences between RGB and infrared images during the complementary information filtering process, a multimodal complementary information filter for RGB and infrared images is designed on the basis of global-local synergistic attention. To fully exploit the advantages of the two types of model images, a multi-branch progressive feature fusion network is designed. The multi-branch progressive feature fusion network progressively fuses the enhanced features from the global-local synergistic attention with the filtered complementary information by the multimodal complementary information filter, resulting in a final fused feature that integrates the rich visual details of RGB images with the distinctive physical characteristics of infrared images. Results against publicly released datasets demonstrate that the proposed method achieves the state-of-the-art detection performance.

In order to improve the efficiency of human–robot interaction (HRI), it is necessary to carry out research on precise control of the manipulator. In this paper, an adaptive non-singular fast terminal sliding mode control scheme is proposed for robot manipulators to solve the trajectory tracking problem with model uncertainty and external disturbances. At first, a novel non-singular fast terminal sliding mode surface is proposed, and by introducing an auxiliary function, the singularity problem caused by the inverse of the error-related matrix could be avoided in the controller design process. Then, the controller is developed by using Lyapunov synthesis. A robust adaptive strategy is used to deal with lumped uncertainty, with an adaptive update law designed to compensate for the upper bound of lumped uncertainty whose upper bound is prior unknown. Finally, a two-link robot manipulators as a simulation example is given to illustrate the effectiveness of the proposed scheme. Compared with other similar algorithms, the proposed adaptive non-singular fast terminal sliding mode control scheme has higher efficiency and smaller computational complexity for the reason that no piecewise continuous function is needed to be constructed during the controller design.

Issues such as complex noise interference and the long-tail distribution of data present many challenges to the multi-category ship detection task in synthetic aperture radar (SAR) images. This paper proposes an efficient multi-category-oriented SAR ship detector, which adopts a powerful lightweight feature enhancement vision transformer backbone (FEViT) for a comprehensive feature representation in SAR ship image and, hence, is referred to as FEVT-SAR. FEViT includes two innovative lightweight modules: localized feature interactive convolution block (LFICB) and dual-granularity attention transformer block (DGTB). LFICB fuses multi-receptive field local features to suppress speckle noise, while DGTB employs a coarse-to-fine self-attention to capture the global dependencies and avoids enormous computational costs. Moreover, a selective CopyPaste augmentation (SCP) paradigm is designed to rebalance ship data distribution through data sampling. Finally, the performance of the FEVT-SAR is evaluated on two typical SAR ship datasets, namely SRSDD and HRSID. Experimental results show that the mean average precision (mAP50) of FEVT-SAR reaches 68.59% and 89.62%, respectively. The proposed FEVT-SAR outperforms several state-of-the-art-oriented bounding box detectors in the multi-category ship dataset SRSDD while demonstrating its robustness in the single-category ship dataset HRSID. The code is available at https://github.com/dmfdmf/FEVT-SAR.