to improve the clarity of objects in a dark-light environment, and to facilitate the identification and detection of targets behind. People perceive the color and brightness of a point not only depending on the pixel ...
详细信息
to improve the clarity of objects in a dark-light environment, and to facilitate the identification and detection of targets behind. People perceive the color and brightness of a point not only depending on the pixel value of the point but also the absolute light entering the human eye is related to the color and brightness around the point. The self-calibration model we used considers the surrounding information, which avoids the information pollution, such as large regions of dark background information. In this model, we introduce a heterogeneous convolution filter, which makes reasonable use of different parts of the filter. Through this operation, information from multiple different scale spaces can be fused, and the field of view when applying the convolution layer is greatly increased without increasing the hyper-parameters, thus producing a more distinctive feature representation. Then the self-calibration model is combined with the backbone reinforcement network, which can not only retain the information of the original scale space but also efficiently collect the latent space information to guide the feature transformation in the original space. After the different channels of the image are processed separately, the dependency between the channels can be established by using the heterogeneous convolution filter. Finally, the test on the ExDark data set proves that our dark target enhancement effect has been significantly improved.
Heterogeneity is a fundamental and challenging issue in federated learning, especially for the graph data due to the complex relationships among the graph nodes. To deal with the heterogeneity, lots of existing method...
详细信息
Remote sensing object detection is an important research area in computer vision, widely applied in both military and civilian domains. However, challenges in remote sensing image object detection such as large image ...
Remote sensing object detection is an important research area in computer vision, widely applied in both military and civilian domains. However, challenges in remote sensing image object detection such as large image sizes, complex backgrounds, and significant variations in target scales are prevalent. To address these issues, this paper proposes a new Feature Denoising and Fusion Module (FDFM) aimed at enhancing the accuracy and robustness of object detection. This module comprises a Multi-Scale Denoising Submodule(MDS) and an Attention Optimization Submodule(AOS). The Multi-Scale Denoising Module aims to suppress lower-level texture noise by utilizing higher-level semantic features before the fusion process, reducing the impact of lower-level noise on subsequent multi-scale feature fusion. Meanwhile, the Attention Optimization Module seeks to enhance the precision of self-attention computations within the Multi-Scale Denoising Module without increasing the parameter count. The efficacy of this method was evaluated on public datasets DOTA, VisDrone, VOC and COCO, showing improvements in comparison to baseline models.
Estimating the Ratio of Edge-Users (REU) is an important issue in mobile networks, as it helps the subsequent adjustment of loads in different cells. However, existing approaches usually determine the REU manually, wh...
详细信息
Estimating the Ratio of Edge-Users (REU) is an important issue in mobile networks, as it helps the subsequent adjustment of loads in different cells. However, existing approaches usually determine the REU manually, which are experience-dependent and labor-intensive, and thus the estimated REU might be imprecise. Considering the inherited graph structure of mobile networks, in this paper, we utilize a graph-based deep learning method for automatic REU estimation, where the practical cells are deemed as nodes and the load switchings among them constitute edges. Concretely, Graph Attention Network (GAT) is employed as the backbone of our method due to its impressive generalizability in dealing with networked data. Nevertheless, conventional GAT cannot make full use of the information in mobile networks, since it only incorporates node features to infer the pairwise importance and conduct graph convolutions, while the edge features that are actually critical in our problem are disregarded. To accommodate this issue, we propose an Edge-Aware Graph Attention Network (EAGAT), which is able to fuse the node features and edge features for REU estimation. Extensive experimental results on two real-world mobile network datasets demonstrate the superiority of our EAGAT approach to several state-of-the-art methods.
Generative self-supervised learning demonstrates outstanding representation learning capabilities in both Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). However, there are currently no generative...
详细信息
This paper presented an improved linear discriminant analysis (LDA) algorithm for face recognition, which can effectively deal with the two problems in traditional LDA-based approaches: (1) the small sample size probl...
详细信息
This paper presented an improved linear discriminant analysis (LDA) algorithm for face recognition, which can effectively deal with the two problems in traditional LDA-based approaches: (1) the small sample size problem, and (2) the Fisher criterion is nonoptimal with respect to classification rate. In particular, the proposed algorithm can also improve the classification rate of one or several appointed classes. The key to this method is to use the technique that it can reserves the significant discriminatory information for dimension reduction and meanwhile utilize a modified Fisher criterion. The comparative experiments on ORL face database verify the effectiveness of the proposed method.
De-interlacing is very important when converting interlaced pictures to progressive pictures in format ***-formats digital broadcast and progressive display requires the de-interlacing technique. An adaptive weight de...
详细信息
De-interlacing is very important when converting interlaced pictures to progressive pictures in format ***-formats digital broadcast and progressive display requires the de-interlacing technique. An adaptive weight deinterlacing method is proposed. It combines motion compensation technique with directional-based spatio-temporal filter efficiently. Experiment results indicate that the method can keep edge continuity and sharpness effectively, reduce the artifacts in motion areas, and shows better visual performance when the estimated motion vectors are inaccurate.
A C64x-based multi-DSP real-time imageprocessing system is introduced, which uses high performance TMS320C6414 DSP to process image and FPGA device to realize LINK port to transport image data with LVDS signal. Requi...
详细信息
A C64x-based multi-DSP real-time imageprocessing system is introduced, which uses high performance TMS320C6414 DSP to process image and FPGA device to realize LINK port to transport image data with LVDS signal. Requirements of imageprocessing performance and image data communication of image fusion are met. Based on the hardware system, a real time microkernel based distributed operating system is designed and implemented. At the end, its real-time performance is analyzed from three aspects. It's shown that the real time imageprocessing system can reach the requirements of real time imageprocessing.
A major difficulty in multivariable control design is the cross-coupling between inputs and outputs which obscures the effects of a specific controller on the overall behavior of the system. This paper considers the a...
详细信息
A major difficulty in multivariable control design is the cross-coupling between inputs and outputs which obscures the effects of a specific controller on the overall behavior of the system. This paper considers the application of ker nel method in decoupling multivariable outputfeedback controllers. Simulation results are presented to show the feasibility of the proposed technique.
An integrated memory array processor (IMAP) ULSI with 64 processing elements and a 2 Mb SRAM has been developed to build a compact real-time imageprocessing system. The chip attains a 3.84 GIPS peak performance throu...
详细信息
An integrated memory array processor (IMAP) ULSI with 64 processing elements and a 2 Mb SRAM has been developed to build a compact real-time imageprocessing system. The chip attains a 3.84 GIPS peak performance through the use of SIMD parallel processing and 1.28 GByte/s on-chip processor-memory bandwidth. The IMAP is capable of parallel indexed addressing, which increases applications for parallel algorithms. Created using 0.55 mu m BiCMOS double layer metal process technology, the IMAP contains 11 million translators in a 15.1 x 15.6 mm(2) die area.
暂无评论