Search

Pedestrian detection is a crucial task in computer vision, with various applications in surveillance, autonomous driving, and robotics. However, detecting pedestrians in complex scenarios, such as rainy days, remains a challenging problem due to the degradation of image quality and the presence of occlusions. To address this issue, we propose RSTDet-Lite (a robust lightweight network) for pedestrian detection on rainy days, based on an improved version of YOLOv5-x. Specifically, in order to reduce the redundant parameters of the YOLOv5-x backbone network and enhance its feature extraction capability, we propose a novel approach named CBP-GNet, which incorporates a compact bilinear pooling algorithm. This new net serves as a new backbone network, resulting in significant parameter reduction and enhancing the fine-grained feature fusion capability of the network. Additionally, we introduce the Simple-BiFPN structure as a replacement for the original feature pyramid module based on the weighted bidirectional feature pyramid to further improve feature fusion efficiency. To enhance network performance, we integrate the CBAM attention mechanism and introduce the idea of structural reparameterization. To evaluate the performance of our method, we create a new dataset named RainDet3000, which consists of 3000 images captured in various rainy scenarios. The experimental results demonstrate that, compared with YOLOv5, our proposed model reduces the network size by 30 M while achieving a 4.56% increase in mAP. This confirms the effectiveness of RSTDet-Lite in achieving excellent performance in rainy-day pedestrian detection scenarios. [ABSTRACT FROM AUTHOR]

Visible–infrared person re-identification (VI-ReID) is a current focused area in the field of re-identification. In order to reduce the gap between two modalities in VI-ReID and improve recognition accuracy, this paper proposes a four-stream network and nonsignificant feature learning (FS-NSF) method for VI-ReID. First, the dual-intermediate modality images of visible and infrared modalities are generated by two lightweight networks, and the labels are inherited from the visible and infrared images. Second, the ResNet50 backbone network is split in order to reconstruct the network adapted to shared feature learning of the four modalities. Finally, a multi-branch, multi-scale and multi-granularity feature extraction strategy is used to extract both significant and nonsignificant features. The comparison experiments are conducted on SYSU-MM01 dataset and RegDB dataset. The experimental results show that, compared with state-of-the-arts, our method has excellent performance on both datasets, especially on the SYSU-MM01 dataset, with an increase in performance of 1.9–6.28% for each index. [ABSTRACT FROM AUTHOR]

Object detection and analysis in remote sensing images is a critical research subject for many businesses and agencies. At present, object detection based on convolutional neural network (CNN) in natural scenes has good performance. Due to the large number of small objects and similar characteristics between the objects in the VisDrone dataset, the current model cannot extract more small-scale features. Therefore, this paper proposes a stronger feature extraction FasterRCNN (SFE-FasterRCNN) that advances a feature extraction strengthening network to enhance the feature learning ability for different objects. Specifically, the pixel proposal network (PPN) is proposed by combining the low-resolution and strong semantic features with high-resolution and weak semantic features through a top-down approach and reusing these fusion blocks vertically to construct a comprehensive semantic feature map. Then hyperbolic pooling is proposed to minimize the loss of feature information during the activation mapping process. Finally, data clustering is used to adaptively generate better object proposals according to the characteristics of the dataset. Experimental results on the VisDrone dataset show that our method has excellent detection results. [ABSTRACT FROM AUTHOR]

With the improvement of the quality and resolution of remote sensing (RS) images, scene recognition tasks have played an important role in the RS community. However, due to the special bird's eye view image acquisition mode of imaging sensors, it is still challenging to construct a discriminate representation of diverse and complex scenes to improve RS image recognition performance. Capsule networks that can learn the spatial relationship between the features in an image has a good image classification performance. However, the original capsule network is not suitable for images with a complex background. To address the above issues, this paper proposes a novel end-to-end capsule network termed DS-CapsNet, in which a new multi-scale feature enhancement module and a new Caps-SoftPool method are advanced by aggregating the advantageous attributes of the residual convolution architecture, Diverse Branch Block (DBB), Squeeze and Excitation (SE) block, and the Caps-SoftPool method. By using the residual DBB, multiscale features can be extracted and fused to recover a semantic strong feature representation. By adopting SE, the informative features are emphasized, and the less salient features are weakened. The new Caps-SoftPool method can reduce the number of parameters that are needed in order to prevent an over-fitting problem. The novel DS-CapsNet achieves a competitive and promising performance for RS image recognition by using high-quality and robust capsule representation. The extensive experiments on two challenging datasets, AID and NWPU-RESISC45, demonstrate the robustness and superiority of the proposed DS-CapsNet in scene recognition tasks. [ABSTRACT FROM AUTHOR]