No abstract

This paper describes a new type of image segmentation method based on deep convolutional neural networks (DCNN) in the actual autonomous driving scene. The spatial pyramid pooling model is used to identify and segment the actual scene to complete the machine-aware task. In order to improve the information aggregation of the whole image, we use atrous convolution for multi-scale feature extraction based on the pyramid structure of image cascade network (ICNet), removing a residual module in the fifth stage of the network, in order to reduce the scale of the convolutional layer. The feature map is preprocessed by padding and atrous convolution before the four-level spatial pyramid model. Then, conventional convolutions are introduced to compose the atrous spatial pyramid pooling (ASPP) structure. Finally, the four feature maps in the pyramid are merged with the feature maps before input into the pyramid. This paper analyzes the spatial pyramid model, receptive field, and dilation convolution in detail and propose atrous image cascade network (AtICNet). Experiment results in the cityscape dataset have shown that AtICNet has some improvements over ICNet, by improving the accuracy of the segmentation.

In this paper, the direction of arrival (DOA) estimation of signals in the presence of impulsive noise environment is studied. Complex isotropic symmetric alpha-stable (SαS) random variables are modeled as impulsive noise, then a novel second-order statistic method that correntropy-based covariance matrix (CBCM) is defined, based on the combination of the CBCM of the array sensor outputs with the signal subspace technique (e.g., multiple signal classification (MUSIC)), which can be achieved source localization under impulsive noise environments. The Monte-Carlo simulation results illustrate the improved performance of CBCM-MUSIC for DOA estimation under a wide range of impulsive noise conditions.

No abstract

Two methods for improving the detection performance of neural networks are introduced in this paper, multifeature map detection and multi-branch convolution structure. The former is to analyze the features of each convolution layer in the network separately, because these features have different resolutions and correspond to objects of different sizes. Finally, the comprehensive judgment of the analysis results can give better consideration to the overall situation and improve the accuracy of detection. The multi-branch convolution structure uses convolutions of different sizes on multiple branches to process input in parallel, and these branches are independent of each other. Finally, the feature maps corresponding to different receptive fields from each branch are combined and analyzed comprehensively. In this paper, the application process of the above two methods is described in combination with classical neural networks, such as the single shot multibox detector (SSD) and receptive field block (RFB) net.

No abstract

No abstract

No abstract