Search

Vessel segmentation is critical for disease diagnosis and surgical planning. Recently, the vessel segmentation method based on deep learning has achieved outstanding performance. However, vessel segmentation remains challenging due to thin vessels with low contrast that easily lose spatial information in the traditional U-shaped segmentation network. To alleviate this problem, we propose a novel and straightforward full-resolution network (FR-UNet) that expands horizontally and vertically through a multiresolution convolution interactive mechanism while retaining full image resolution. In FR-UNet, the feature aggregation module integrates multiscale feature maps from adjacent stages to supplement high-level contextual information. The modified residual blocks continuously learn multiresolution representations to obtain a pixel-level accuracy prediction map. Moreover, we propose the dual-threshold iterative algorithm (DTI) to extract weak vessel pixels for improving vessel connectivity. The proposed method was evaluated on retinal vessel datasets (DRIVE, CHASE_DB1, and STARE) and coronary angiography datasets (DCA1 and CHUAC). The results demonstrate that FR-UNet outperforms state-of-the-art methods by achieving the highest Sen, AUC, F1, and IOU on most of the above-mentioned datasets with fewer parameters, and that DTI enhances vessel connectivity while greatly improving sensitivity. The code is available at: https://github.com/lseventeen/FR-UNet. [ABSTRACT FROM AUTHOR]

As one of the main pollution sources in China, haze can blur the railway monitoring and threaten railway safety. In this paper, we propose an end-to-end multi-scale residual network (MSRN) which can achieve remarkable dehazing effect on railway monitoring images. The method optimizes the image dehazing algorithm in three aspects: network structure, loss function and hazy dataset. Firstly, inspired by the residual network, the paper presents a method of fusing multi-scale feature information based on the residual network, which can extract more effective information at different scales. Secondly, a combined loss function is designed to achieve better convergent results by balancing training time, training calculations, and precision. Thirdly, the paper synthesizes an outdoor dataset specifically for railway scenarios, which relies on real depth maps and various outdoor images. Extensive experimental results on both full reference image quality assessment and no reference image quality assessment of image restoration demonstrate that the proposed algorithm shows higher performance than the state-of-the-art algorithms. Moreover, the haze of railway monitoring images is removed under hazy weather, and the detection algorithm achieved higher detection accuracy on the images after dehazing. The proposed network structure, loss function, and hazy dataset are discussed and analyzed in detail to verify the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Polar codes are known as the first capacity-achievable codes with low encoding and decoding complexity. The sequential decoding nature of traditional polar decoding algorithms such as successive cancellation (SC) results in high decoding latency, which is not suitable for services that require high reliability and low latency. Deep learning for decoding, referred to as neural network decoder (NND), has shown strong competitiveness because of its non-iterative and full-parallel feature. Whereas, the bit-error-rate (BER) performance of NND is still not satisfactory. In this paper, we first propose a residual learning denoiser (RLD) for polar codes. The RLD can remarkably improve the signal-to-noise ratio (SNR) and reduce the symbol-error-rate (SER) of received symbols. In order to decode polar codes more efficiently, we subsequently propose a residual neural network decoder (RNND) for polar codes. Different from the traditional pure-NND (PNND) which directly uses neural network for decoding received symbols, the proposed RNND concatenates a RLD used for denoising and a NND used for decoding. We provide a novel multi-task learning (MTL) strategy to jointly optimize the denoiser and decoder, and find that the denoising gain of Joint-RLD is more significant than the Independent-RLD. Numerical results show that the proposed RNND outperforms its counterpart PNND with regard to the BER performance. In addition, the optimal RNND(MLP-MLP) approaches the traditional SC decoding performance, while saving more than one hundred times of computation time. Eventually, scalability of the RNND to longer polar codes as well as LDPC codes further demonstrates the superiority of our proposed scheme. [ABSTRACT FROM AUTHOR]

Fine-grained classification and grading of breast cancer (BC) histopathological images are of great value in clinical application. However, automatic classification and grading of BC histopathological images are complicated by (1) small inter-class variance and large intra-class variance exist in BC histopathological images, and (2) features extracted from similar histopathological images with different magnification are quite different. To address these issues, an improved deep convolution neural network model is proposed and the procedure can be divided into three main stages. Firstly, in the representation learning process, multi-class recognition task and verification task of image pair are combined. Secondly, in the feature extraction process, a prior knowledge is built, which is "the variances in feature outputs between different subclasses is relatively large while the variance between the same subclass is small." Additionally, the prior information that histopathological images with different magnification belong to the same subclass are embedded in the feature extraction process, which contributes to less sensitive with image magnification. The experimental results based on three different histopathological image datasets show that the performance of the proposed method is better than state of the art, with better robustness and generalization ability. [ABSTRACT FROM AUTHOR]

Scene recognition is an important branch of computer vision and a common task for deep learning. As is known to all, different scenes are supported by different "key objects". Therefore, the neural network used for the scene recognition task needs to extract the features of these key objects in the scene, sometimes even has to integrate the positional relation between objects to determine the class to which the scene belongs. Under some circumstances, key objects in the scenes are very small and the features of them become extremely inconspicuous or even disappear in the deep layers of the network. Such kind of phenomenon is called "small object-supported scenes". In this paper, Multi-Level Ensemble Network (MLEN), a convolutional neural network, has been proposed, to improve the recognition accuracy of these "small object-supported scenes". Features from multiple levels of the net are used to make separate predictions. Then ensemble learning is performed within the net to make the final prediction. Apart from all this, "Feature Transfer Path" is added and feature fusion methods are adopted to make full use of low-level and high-level features. Moreover, a class-weight loss function for the problem of non-uniform class distribution has been designed. This function can help further improve accuracy in most scene recognition datasets. The experiments involve the Urban Management Case (UMC) dataset collated from two smart urban management system databases by ourselves, and the Places-mini dataset, which is a subset of the well-known Places dataset [36]. The results show that our Multi-Level Ensemble Network achieves much higher accuracy than the state-of-the-art scene recognition networks on both datasets. [ABSTRACT FROM AUTHOR]

• The proposed method can effectively protect the railway from foreign objects in a complex and open environment. • The proposed method combines the advantages of traditional algorithms and deep learning and is significant for the application of deep learning in railways. • This work does not compress the input images so significantly improves the accuracy of detecting small objects. • An effective railway intrusion detection algorithm based on depthwise separable convolutions is designed. It is significant to detect railway intrusion for railway safety. Unlike indoors, the railway is a complex environment with many interferences leading to false positives in the intrusion detection algorithm. However, there is a lack of research on effective intrusion detection in complex railway scenes. To solve these problems, the paper proposes RailDet, an effective railway intrusion detection method using dynamic intrusion region and lightweight neural network, which has high accuracy, low false-positives and high speed. RailDet consists of two stages. Firstly, the intrusion localization algorithm is used to obtain dynamic intrusion regions. Secondly, the object recognition based on lightweight neural network processes intrusion regions to obtain the coordinates and category. Finally, RailDet is validated on Chinese railways with various weathers and locations, which achieves 96.90% accuracy, 0.24% false-positive rate and 0.043 s/image. Moreover, the proposed method is very meaningful for the application of deep learning and not limited to railway. [ABSTRACT FROM AUTHOR]

Trains shuttle in semiopen environments, and the surrounding environment plays an important role in the safety of train operation. The weather is one of the factors that affect the surrounding environment of railways. Under haze conditions, railway monitoring and staff vision could be blurred, threatening railway safety. This paper tackles image dehazing for railways. The contributions of this paper for railway video image dehazing are as follows: (1) this paper proposes an end-to-end residual block-based haze removal method that consists of two subnetworks, namely fine-grained and coarse-grained network can directly generate the clean image from input hazy image, called RID-Net (Railway Image Dehazing Network). (2) The combined loss function (per-pixel loss and perceptual loss functions) is proposed to achieve both low-level features and high-level features so to generate the high-quality restored images. (3) We take the full-reference criterion (PSNR&SSIM), object detection, running time, and sensory vision to evaluate the proposed dehazing method. Experimental results on railway synthesized dataset, benchmark indoor dataset, and real-world dataset demonstrate our method has superior performance compared to the state-of-the-art methods. [ABSTRACT FROM AUTHOR]