Knowledge graph embedding is to embed symbolic relations and entities of the knowledge graph into low dimensional continuous vector space. Despite the requirement of negative samples for training knowledge graph embedding models, only positive examples are stored in the form of triplets in most knowledge graphs. Moreover, negative samples generated by negative sampling of conventional knowledge graph embedding methods are easy to be discriminated by the model and contribute less and less as the training going on. To address this problem, an Adversarial Negative Generator (ANG) model was proposed. The generator applied the encoder-decoder pipeline, the encoder readed in positive triplets whose head or tail entities were replaced as context information, and then the decoder filled the replaced entity with the triplet using the encoding information provided by the encoder, so as to generate negative samples. Several existing knowledge graph embedding models were used to play an adversarial game with the proposed generator to optimize the knowledge representation vectors. By comparing with existing knowledge graph embedding models, it can be seen that the proposed method has better mean ranking of link prediction and more accurate triple classification result on FB15K237, WN18 and WN18RR datasets.
To solve the time-consuming and error-prone problem in the diagnosis of fundus images by the ophthalmologists, an unsupervised automatic detection method for hard exudates in fundus images was proposed. Firstly, the blood vessels, dark lesion regions and optic disc were removed by using morphological background estimation in preprocessing phase. Then, with the image luminosity channel taken as the initial image, the low rank matrix and sparse matrix were obtained by combining local entropy and Robust Principal Components Analysis (RPCA) based on the locality and sparsity of hard exudates in fundus images. Finally, the hard exudates regions were obtained by the normalized sparse matrix. The performance of the proposed method was tested on the fundus images databases e-ophtha EX and DIARETDB1. The experimental results show that the proposed method can achieve 91.13% of sensitivity and 90% of specificity in the lesional level and 99.03% of accuracy in the image level and 0.5 s of average running time. It can be seen that the proposed method has higher sensitivity and shorter running time compared with Support Vector Machine (SVM) method and K-means method.
Ultrasound image segmentation of left ventricle is very important for doctors in clinical practice. As the ultrasound images contain a lot of noise and the contour features are not obvious, current Convolutional Neural Network (CNN) method is easy to obtain unnecessary regions in left ventricular segmentation, and the segmentation regions are incomplete. In order to solve these problems, keypoint location and image convex hull method were used to optimize segmentation results based on Fully Convolutional neural Network (FCN). Firstly, FCN was used to obtain preliminary segmentation results. Then, in order to remove erroneous regions in segmentation results, a CNN was proposed to locate three keypoints of left ventricle, by which erroneous regions were filtered out. Finally, in order to ensure that the remained area were able to be a complete ventricle, image convex hull algorithm was used to merge all the effective areas together. The experimental results show that the proposed method can greatly improve left ventricular segmentation results of ultrasound images based on FCN. Under the evaluation standard, the accuracy of results obtained by this method can be increased by nearly 15% compared with traditional CNN method.
In traditional pulmonary nodule detection algorithms, there are problems of low detection sensitivity and large number of false positives. To solve these problems, a pulmonary nodule detection algorithm based on deep Convolutional Neural Network (CNN) was proposed. Firstly, the traditional full convolution segmentation network was simplified on purpose. Then, in order to obtain high-quality candidate pulmonary nodules and ensure high sensitivity, the deep supervision of partial CNN layers was innovatively added and the improved weighted loss function was used. Thirdly, three-dimensional deep CNNs based on multi-scale contextual information were designed to enhance the feature extraction of images. Finally, the trained fusion classification model was used for candidate nodule classification to achieve the purpose of reducing false positive rate. The performance of algorithm was verified through comparison experiments on LUNA16 dataset. In the detection stage, when the number of candidate nodules detected by each CT (Computed Tomography) is 50.2, the sensitivity of this algorithm is 94.3%, which is 4.2 percentage points higher than that of traditional full convolution segmentation network. In the classification stage, the competition performance metric of this algorithm reaches 0.874. The experimental results show that the proposed algorithm can effectively improve the detection sensitivity and reduce the false positive rate.
In order to ensure that the Wireless Sensor Network (WSN) delay requirements while minimizing power consumption, a sensor network queue management algorithm based on duty cycle control and delay guarantees (DQC) was proposed. According to changing network conditions, in order to better control node duty cycle and queue thresholds, a two-way controller was used. The controller provided a delay notification mechanism to determine an appropriate sleep time and queue length for each node based on application requirement and time-varying delay requirement. And the stability of the state of two-way controller was derived based on control theory to obtain a condition of the control parameters for guaranteeing asymptotically stable steady state. Simulation results show that compared with the algorithm based on adaptive duty cycle control and performance improvement queue-based congestion management mechanism, the proposed algorithm shortened end-to-end delay of time period by 38.8% and 36.0%, reduces the average power consumption by 46.5 mW and 27.5 mW. It show better performances on the control of delay time and energy efficiency.
During the automatic segmentation of cardiac structures in echocardiographic sequences within a cardiac cycle, the contour with weak edges can not be extracted effectively. A new approach combining Speeded Up Robust Feature (SURF) and Chan-Vese model was proposed to resolve this problem. Firstly, the weak boundary of heart chamber in the first frame was marked manually. Then, the SURF points around the boundary were extracted to build Delaunay triangulation. The positions of weak boundaries of subsequent frames were predicted using feature points matching between adjacent frames. The coarse contour was extracted using Chan-Vese model, and the fine contour of object could be acquired by region growing algorithm. The experiment proves that the proposed algorithm can effectively extract the contour of heart chamber with weak edges, and the result is similar to that by manual segmentation.
Studying the constructing mechanism of micro-blog transmission network help to understand the information spreading process on the micro-blog platform deeply, and then obtain effective strategies and suggestions. As for this issue, a directed and weighted network model was proposed. In the model building process, according to the phenomenon that micro-blogs can be transmitted more than one time, triad formation was introduced. Different directions of links were used to represent the various characteristics of active and famous users. Besides, the dynamic evolution process of the link weight was considered. The theory analysis and simulation experiment results indicate the strength distribution, the degree distribution and the correlation of strength and degree obey power-law distribution, and the power exponents are between 1 and 3. Also, this model is characterized by high clustering coefficient and short average path length. Average clustering coefficient is 0.7, and average length is less than 6. As well, actual data of micro-blog transmission were collected to prove the model's correctness.
To improve the accuracy of Centroid Localization (CL) algorithm in Wireless Sensor Network (WSN), an Optimal Beacon nodes-based Centroid Localization (OBCL) algorithm was proposed. In this algorithm, four mobile beacon nodes were used. First, the path for each mobile beacon node was planned. Second, the optimal beacon nodes were selected from the candidate beacon nodes by each unknown node to estimate location according to Set Deviation Degree (SDD). Besides, a role-change mechanism that an unknown node can assist other unknown nodes to locate as the expectant beacon node after it got its estimated location was adopted to solve the problem of beacon nodes' shortage. At last, to ensure that each unknown node could get its location, a relocation procedure was executed after the completion of the initial locating. The simulation results show that, the average locating error is respectively reduced by 67.7%, 39.2%, 24.4% comparing with the CL, WCL (Weighted Centroid Localization), RR-WCL (Weighted Centroid Localization based on Received signal strength indication Ration) algorithms. For the reason that OBCL can achieve better locating results using only four mobile beacon nodes, it is suitable for scenes which require low network cost and high locating accuracy.