现有的视网膜血管分割算法存在特征提取能力不足和分割效率低等问题。针对该问题,对UNet网络进行改进,提出一种基于多尺度特征提取的U型网络(Multi-scale feature extraction based on UNet, MF-UNet)。该算法在编码和解码部分构建反卷...现有的视网膜血管分割算法存在特征提取能力不足和分割效率低等问题。针对该问题,对UNet网络进行改进,提出一种基于多尺度特征提取的U型网络(Multi-scale feature extraction based on UNet, MF-UNet)。该算法在编码和解码部分构建反卷积分割模块替代传统卷积块,使网络保留更多的血管细节信息。之后,在编码和解码中间连接部引入混合池化(Mix Pooling Moudle, MPM)和模板卷积(Template convolution, TConv),提升网络对多尺度特征的提取能力,从而提升血管的分割质量和分割效率。在两个眼底数据库DRIVE和STARE上进行实验验证,结果表明,MF-UNet算法在准确性、灵敏度、特异性和AUC表现优异,更优于UNet与其他视网膜血管分割算法。展开更多
Inspired by the immune network theory, an adaptive anomaly detection paradigm based on artificial immune network, referred as APAI, is proposed. The implementation of the paradigm includes: initially, the first is to...Inspired by the immune network theory, an adaptive anomaly detection paradigm based on artificial immune network, referred as APAI, is proposed. The implementation of the paradigm includes: initially, the first is to create the initial antibody network; then, through the learning of each training antigen, the antibody network is evolved and updated by the optimal antibodies. Finally, anomaly detection process is accomplished by majority vote of the k nearest neighbor antibodies in the network. The experiments used the famous Sonar Benchmark dataset in our study, which is taken from the UCI machine learning database. The obtained detection accuracy of APAI was 97.7%, which was very promising with regard to the other classification applications in the literature for this problem. In addition to its nonlinear classification properties, APAI possesses biological immune network properties such as clonal selection, immune network, and immune memory, which can be applied to pattern recognition, classification, and etc.展开更多
基金Supported by the National High Technology Research and Development Program of Chin(a863 Program)(2006AA01Z435)the National Natural Science Foundation of China (60573130, 60502011).
文摘Inspired by the immune network theory, an adaptive anomaly detection paradigm based on artificial immune network, referred as APAI, is proposed. The implementation of the paradigm includes: initially, the first is to create the initial antibody network; then, through the learning of each training antigen, the antibody network is evolved and updated by the optimal antibodies. Finally, anomaly detection process is accomplished by majority vote of the k nearest neighbor antibodies in the network. The experiments used the famous Sonar Benchmark dataset in our study, which is taken from the UCI machine learning database. The obtained detection accuracy of APAI was 97.7%, which was very promising with regard to the other classification applications in the literature for this problem. In addition to its nonlinear classification properties, APAI possesses biological immune network properties such as clonal selection, immune network, and immune memory, which can be applied to pattern recognition, classification, and etc.