The segmentation of liver and tumor images plays a supportive role in clinical diagnosis. Liver tumors vary significantly in size and exhibit diverse morphological variations. Additionally, liver tumors are similar to their surrounding tissues. These obstacles hinder algorithms from enhancing their effectiveness. In this paper, PVTv2 is employed as the encoder on the left side of the model. Multiscale information extraction modules are used at the skip connections to capture rich feature representations. To distinguish similar tissues, the model enhances uncertain areas by attention mechanism. The global dice coefficient for liver on the LiTS dataset reached 97.33%, with 82.24% for tumor. Our model outperforms the other models in comparison. Our ablation experiments validate the effectiveness of our modules.

The color filter array of the camera is an effective fingerprint for digital forensics. Most previous color filter array (CFA)-based forgery localization methods perform under the assumption that the interpolation algorithm is linear. However, interpolation algorithms commonly used in digital cameras are nonlinear, and their coefficients vary with content to enhance edge information. To avoid the impact of this impractical assumption, a CFA-based forgery localization method independent of linear assumption is proposed. The probability of an interpolated pixel value falling within the range of its neighboring acquired pixel values is computed. This probability serves as a means of discerning the presence and absence of CFA artifacts, as well as distinguishing between various interpolation techniques. Subsequently, curvature is employed in the analysis to select suitable features for generating the tampering probability map. Experimental results on the Columbia and Korus datasets indicate that the proposed method outperforms the state-of-the-art methods and is also more robust to various attacks, such as noise addition, Gaussian filtering, and JPEG compression with a quality factor of 90.

This paper addresses the problem of vascular segmentation of retinal fundus images, which is prone to breakage and missing microvessel segmentation, from the perspective of enabling the network to acquire more abundant and significant semantic information. First, using U-Net as the backbone network, the novel residual attention module RSCA and multiscale transformer as generators, and the binary classification model with dense jump connections as discriminators, which are combined to form a multiscale GAN network. Second, setting the joint loss function for parameter update. The accuracy, sensitivity, specificity, and ACC of the constructed model on the DRIVE dataset reached 98.57%, 84.13%, 98.63%, and 97.80%, respectively, which is at the level of human experts.

This work suggests a medical image segmentation approach based on local-global Gaussian-weighted attention as a remedy for the U-Net model’s lack of explicit modeling of long-range dependencies for medical image segmentation problems. The approach uses a Transformer module built on a local-global Gaussian weighted attention mechanism and a unified decoder module in the U-Net model to enable relevant relationship mining and centralized multi-resolution feature processing. Additionally, this paper validates the efficacy of the method put forth in this paper on two openly accessible datasets, Synapse and ACDC, by expanding the dataset with random angle rotation and mirroring operations for pre-processing and using a loss function combining HD95 and Dice for training. According to the experimental findings, the suggested method is competitive with other existing methods and outperforms TransUNet and U-Net in terms of the model’s generalization and segmentation effects.

In recent years, using convolutional neural network (CNN) to segment medical images has achieved good results already. The CNN model with U-shaped structure of encoder and decoder and jump connection mechanism has been widely used in various medical tasks. However, CNN is unable to learn global information and conduct remote semantic information interaction due to its inductive bias. With the application of Transformer in Computer Vision field, the global remote dependency modeling capability brought by Transformer can make up for the locality of CNN to some extent. However, in recent studies, most proposed methods only use Transformer or combinations of Transformer and CNN at both encoder and decoder. In this paper, we propose a new U-shaped network framework with incomplete symmetry, using Transformer module for image feature extraction, and using CNN for image recovery, to seek more possibilities for the combination of Transformer and CNN in the medical segmentation field. Experiments on cardiac MRI segmentation ACDC data sets show that giving up Transformer in the decoding layer will not reduce the overall segmentation performance, and half-transformer and half-CNN network structure can give a good balance between computation and segmentation performance.

Powerful image editing software makes the process of image manipulation easy, which increases security risks. Therefore, it is urgent to locate the tampered region to uncover the processing history. However, previous research has mainly focused on feature extraction, with little discussion on classifiers for classifying original and tampered regions. We improve the splicing forgery localization method from a statistical perspective. The refined color filter array feature provides sufficient data for statistical analysis, and the geometric mean is used to eliminate anomalous data. Subsequently, a classifier that combines the expectation-maximization algorithm and Bayesian theory is proposed to binarize the original and tampered regions. The two steps of feature extraction and feature classification are associated from a statistical perspective, which ultimately improve the performance of the method effectively. Extensive experimental results demonstrate that the refined feature used for classification has several advantages, and the proposed classifier is appropriate for handling complex image manipulation across different statistical distributions. The proposed method outperforms the reference methods in both the Columbia and Korus datasets.

According to the characteristics of kidney tumours in abdominal multi-organ CT images, analyzing the CNN network to deal with long-distance connection defects and the shortage of Transformer structure detail processing, we designed a multimodal residual continuous convolution CNN and an improved multiscale dual-path multi-head attention Transformer combined with an encoder, deconvolution up-sampling, and learnable attention gating mechanism of the decoder with a jump junction introduced into the multiscale convolutional fusion module to form an asymmetric U-Net model framework for kidney tumour segmentation. Experimental comparison results show that the proposed method achieves an average dice coefficient of 0.9352, an average cross-merge rate of 0.9894, and a class-average pixel accuracy of 0.98235 for kidney segmentation metrics, which are more accurate and detailed than the existing and newer network models for segmentation.

With the improvement of medical consumption level, patients have more and more demand for the prediction of treatment costs. However, the prediction accuracy of existing research methods is low when the amount of data is small. In order to solve this problem, a weighted lasso regression method is proposed to predict the treatment cost based on the electronic medical record. Firstly, a set of transformation method of text-based medical record data is established, and the missing values are supplemented according to the clustering distance to realize the data representation of medical records. Then, in view of the low prediction accuracy of the traditional regression model, the lasso regression model with local weighting is established by introducing the data feature weight into lasso regression method. Finally, the model is verified by the medical record data provided by the hospital, and the results show that the model has higher prediction accuracy.

Noise is the inherent intrinsic fingerprint in digital images and is often used for forgery localization. Most noise-based methods assume that the noise is similar over the whole image and can be considered as white Gaussian noise. However, the noise is different in various regions, which degrade the performance of these noise-based methods. To reduce the impact of impractical assumptions, in this paper, we propose an effective noise fingerprint incorporated with CFA configuration for splicing forgery localization. The noise of interpolated pixels is expected to be suppressed after interpolation, and the relationship between the noise levels of adjacent acquired and interpolated pixels is only related to the interpolation algorithm, which is constant in the original image. We utilize a dual tree wavelet based denoising algorithm to extract the noise from the green channel and compute the standard deviation of the noise for acquired and interpolated pixels, respectively. The noise level of acquired and interpolated pixels are then obtained by the geometric mean of the noise standard deviations. Finally, the ratio of noise levels between acquired and interpolated pixels can be a fingerprint to locate tampered regions. Experiments conducted on publicly available databases demonstrate that the proposed approach outperforms previous methods for detecting splice tampering. Moreover, the proposed method is robust to Gaussian filtering and JPEG compression attacks.

随着面向医疗服务WBAN应用场景的不断增加,对WBAN的QoS要求也日益多样化,由于本地设备的计算能力、通信带宽、能量和存储容量限制,难以保证业务对实时性等方面的需求.因此,如何在本地资源受限的情况下,满足不同类型业务的多样化需求是面向医疗服务WBAN的一个挑战,本文提出了一种基于边云协同模式的无线体域网资源调度方法,根据用户偏好采用多属性决策方法建立任务卸载效用模型,对体域网中多样化业务进行卸载资格预判,在此基础上建立任务卸载模型,对本地设备、边缘和云端资源的均衡调度和管理,提升资源利用率和用户体验,实现整个系统的负载均衡.实验结果证明,本文提出资源调度方法,可以根据用户的偏好设置,有效满足了医疗服务的多样化需求.

Based on the actual monitoring historical data of photovoltaic power station, combined with the actual engineering demand of photovoltaic microgrid on the user side, the lightweight algorithm of ultra short-term photovoltaic power prediction is studied, which is conducive to improving the operation efficiency and economy of power system. In this paper, the ultra short-term power prediction of photovoltaic power station is carried out by combining the LSTM algorithm with attention mechanism. Firstly, Pearson correlation coefficient method is used to reduce the dimension of the data set. The data with low correlation between weather variables and power to be predicted and historical power are eliminated, and the algorithm model structure is simplified. Then, the attention mechanism is combined with LSTM network to improve the effectiveness of the prediction model for long time series input. The proposed model is trained and compared with the data of a photovoltaic power station. The results show that the model achieves good experimental results in different weather conditions, and can effectively improve the prediction accuracy.

Charging according to disease is an important way to effectively promote the reform of medical insurance mechanism, reasonably allocate medical resources and reduce the burden of patients, and it is also an important direction of medical development at home and abroad. The cost forecast of single disease can not only find the potential influence and driving factors, but also estimate the active cost, and tell the management and reasonable allocation of medical resources. In this paper, a method of Bayesian network combined with regression analysis is proposed to predict the cost of treatment based on the patient's electronic medical record when the amount of data is small. Firstly, a set of text-based medical record data conversion method is established, and in the clustering method, the missing value interpolation is carried out by weighted method according to the distance, which completes the data preparation and processing for the realization of data prediction. Then, aiming at the problem of low prediction accuracy of traditional regression model, this paper establishes a prediction model combined with local weight regression method after Bayesian network interpretation and classification of patients' treatment process. Finally, the model is verified with the medical record data provided by the hospital, and the results show that the model has higher prediction accuracy.

片上网络的拓扑结构和路由算法直接影响片上网络的传输延迟和传输效率.基于2D-Torus拓扑结构,提出了一种新的片上网络无死锁路由算法.通过改变数据包在片上网络路由过程中受限制转弯的位置,保证片上网络的自适应路由条件,从而有效降低片上网络的延迟.在FPGA硬件平台上,设计并实现了基于该路由算法的2D-Torus片上网络,并对其进行测试.实验结果表明,基于该路由算法的片上网络,可以满足片上网络多方向数据通信及多路数据并行通信等性能要求.

A high-precision measurement method, based on multiple sampling, is proposed for the time interval of two signals in this study. A time interval measurement circuit integrated into the time-to-digital converter (TDC), is designed based on this high-precision measurement method. In the TDC, the authors use two identical delay lines as the holding module to ensure the two signals with a constant time interval. The TDC samples the two signals multiple times by a clock signal, whose period is shorter than that of the delay line. Consequently, the problem of limited resolution caused by a mismatch between delay lines in the delay-line structure can be avoided, and the precision of the output can be improved. The proposed TDC is designed and simulated in Semiconductor Manufacturing International Corporation (SMIC) 0.18 μm complementary metal–oxide–semiconductor process. Simulation results show that the differential non-linearity and the integral non-linearity are always less than one least significant bits. The proposed TDC achieves input dynamic range of 0–32.13 ns and time resolution of 9 ps.

This paper first introduces the principle of target detection based on Gaussian mixture model, and then introduces the improved method of the original Gaussian mixture model to detect moving targets in detail. The main content is the improvement of initialization method. In this paper, the fast initialization method is used, and the learning rate in the training stage is adjusted, so that it can quickly adapt to the changes in the environment in the training stage. In the aspect of parameter updating, the concept of change rate is introduced, which aims to describe the size of changes in the current frame and the previous frame, so as to adjust the learning rate adaptively In terms of the total number of Gaussian components, this paper adopts the method of initializing a single Gaussian model to increase the Gaussian components with the complexity of description, and the method of combining the removal of Gaussian components and the combination of the removal of Gaussian components in each frame. The removal of Gaussian components can directly remove the Gaussian components with small weight and reduce unnecessary calculation, while the combination of Gaussian components can reduce calculation to describe the background more precisely. In the last part of the improvement, the frame difference method is combined with the Gaussian mixture model to make the edge detection more accurate. Finally, a GUI interface is designed.

提出了一种新的基于FPGA的立体图像差异性算法,它以块匹配算法为基础,根据FPGA的特点,对图像相关性的公式进行设计优化,并结合穷尽方式搜索和预测方式搜索,提高算法的执行速度.设计基于FPGA的立体图像差异性算法IP核,充分利用FPGA独特的并行处理机制和强大的运算能力以提高系统的处理速度和性能.系统测试结果表明,基于FPGA的立体图像差异性算法,可以达到每秒33帧的处理能力,处理速度能够达到PC机的二百倍以上,具有较好的实时性;且能够连续处理500帧图像数据,具有较好的稳定性.

In this paper, a fusion algorithm of background difference and Sobel is proposed. This algorithm can not only accurately detect the moving target, but also well describe the appearance and contour of the target, and also eliminate a lot of irrelevant noise. The detection and tracking system is based on the FPGA development platform of DE2-115 development board. Firstly, video images are collected by TRDB-D5M camera, raw image format is converted into RGB format and stored in SDRAM, and image acquisition is realized. The image is read from SDRAM, and then the median filter is carried out after the gray-scale transformation to realize the image preprocessing. Then we use the fusion algorithm of background difference and Sobel edge detection mentioned above, as well as the centroid tracking algorithm to detect and track the moving target. Finally, we display the real-time image and tracking results through VGA. After many times of debugging, the system can process 640 * 480 @ 60Hz image, realize the real-time tracking of moving target, and the tracking rate can reach higher than 90%. The system realizes the expected function and meets the requirements of real-time, and has a large potential for progress and application prospects.

This paper proposes a 6 bit floating window asynchronous level crossing ADC in 180nm CMOS technology for applications with sparse input signals. Two different hysteresis comparators are used to detect the rise and fall of the input signals respectively. The hysteresis width can changing with the change speed of input signal, so that the level crossing ADC can not only reduce the collection of invalid data when the signal changes slowly, but also collect the peak signal when the signal changes rapidly. Operating under a supply voltage of 1.8 V, the ADC consumes 31.8-41.5 μnW from 1 Hz to 50 kHz.

针对传统遗传算法交叉、变异过程过于繁琐和神经网络在极值判断及收敛速度受限等问题,提出了一种并行的量子遗传算法优化神经网络权值的算法.首先引入了量子计算的概念,在量子计算的过程中使用量子旋门实现染色体的训练,然后引入量子交叉克服了早熟收敛现象,避免了遗传算法中繁琐的交叉、变异过程.最后设计实现了并行的卷积神经网络,使用并行量子遗传算法优化了卷积神经网络权值,实现了并行量子遗传神经网络人脸识别系统.实验结果表明,相对于原来的遗传算法,该算法在鲁棒性和实验速度上都有明显的提高.

The communication mechanism of Network on Chip(NoC)directly influence the transmission delay and information interac-tion capability. A dual-mode-fusion communication mechanism is proposed in the paper. According to the characteristics of the trans-mitted data in NoC,the data is divided into control message and data message. And the data transmission is carried out by different communication mechanisms. At the same time, a configurable dual-mode-fusion data communication interface circuit is proposed. Therefore,the commonality of the routing nodes is increased,and the problems of current asynchronous FIFO with depth and single da-ta transmission channel are solved. Based on reconfigurable technology,the NoC based on the dual-mode-fusion communication mech-anism is designed and realized in the FPGA platform. And the video processing algorithm is used as an example to construct the video processing system. The test results show that the NoC with dual-mode-fusion communication mechanism has a smaller transmission de-lay,greater throughput,and better video processing performance for video processing applications.