Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-04-01 Cooperative journals: 《计算机应用研究》
Abstract： This paper proposed an improved residual network (ResNet) expression recognition algorithm. The algorithm used small convolution kernels and a deep network structure to solve the problem of accuracy reduction with the increase of depth by the residual module. The experiment overcomes the shortcoming of insufficient data through transfer learning, which can effectively prevent overfitting. The network architecture uses a linear support vector machine (SVM) for classification. The experiment used the ImageNet database to pre-train network parameters to have an excellent ability to extract feature. According to transfer learning, the algorithm used the FER-2013 database and the expanded CK+ database to fine-tune and train network parameters, and overcame the problem that shallow networks rely on manual features and deep networks are difficult to train. The results show the recognition rates is 91.333% and 95.775% on the CK+ database and the GENKI-4K database, respectively. The classification accuracy of SVM in CK + database is about 1% higher than that of Softmax.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-12-13 Cooperative journals: 《计算机应用研究》
Abstract： Aiming at the deficiency of K-means clustering algorithm, this paper proposes an improved algorithm with the use of simplified silhouette coefficient as the evaluation criterion to measure the k value in K-means to boost the algorithm performance. The K-means++ algorithm is used to choose the initial center points in the K-means algorithm. After setting the k value and the initial center point, morphology similarity distance is used as the similarity measurement standard to assign the data points to the cluster formed by the closest center point. And finally calculate the average silhouette coefficient to determine the appropriate k value. The improved algorithm in this paper is implemented on Spark. Experiments on accuracy, run-time and speedup of four standard datasets show that the improved K-means algorithm not only improves the quality of clustering division compared with the traditional K-means algorithm and SKDK-means algorithm, but also shortens the calculation time, showing good parallel performance in a multi-node cluster environment. The experimental results suggest that the improved algorithm proposed in this paper can effectively improve the algorithm execution efficiency and parallel computing ability.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-18 Cooperative journals: 《计算机应用研究》
Abstract： Currently, due to lack of effective iterative stopping criterion (ISC) , most of the ID algorithms are usually implemented with fixed iteration steps, leading to the low efficiency and poor deblurring results. Therefore, this paper proposed a deblurring measure (DM) based on the residual image (the difference between the intermediate estimated image convoluted with blur kernel and blur image during each iterative step) . Based on the proposed DM metric, it designed and applied an adaptive iterative stopping criterion (AISC) to the classic nonlocally燾entralized爏parse爎epresentation (NCSR) . Extensive experiments on uniform blur, Gaussian燽lur燼nd爉otion燽lur爏how爐hat, compared with the original NCSR algorithm using fixed iteration steps, the NCSR algorithm adopting AISC has an obvious advantages in terms of efficiency and the reconstructed爄mages have燼 similar image quality in terms of peak signal-to-noise ratio (PSNR) , structural爏imilarity (SSIM) and爁eature爏imilarity (FSIM) .
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-17 Cooperative journals: 《计算机应用研究》
Abstract： It is a challenging topic to carry out fast computing for three-way decision in stream computing mode. Aim at the phenomenon that the increment and decrement of dynamic objects occur synchronously in the stream computing mode, this paper proposed a fast stream computing method for probabilistic rough set three-way decision. Firstly, ytdiscussed the data mode of single-object increment and decrement updating mode in stream computing. Then, proposed the reasoning of the three-way decision domains in data increment and data decrement dynamic mode respectively based on the pattern of data variation. Finally, proposed a three-way decision dynamic incremental and decremental learning algorithm based on the above theory. The comparison experiments of eight UCI datasets show that the algorithm not only outperforms the classical three-decision algorithm in time consumption, but also has strong stability for the three-way decision thresholds.
Hits
Hits
Downloads
Comment