Thoracic Aortic Aneurysm and Dissection (TAAD) is one of the life-threatening cardiovascular diseases, and the histological changes of Medial Degeneration (MD) have important clinical significance for the diagnosis and early intervention of TAAD. Focusing on the issue that the diagnosis of MD is time-consuming and prone to poor consistency because of the great complexity in histological images, a deep learning based classification method of histological images was proposed, and it was applied to four types of MD pathological changes to verify its performance. In the method, an improved Convolutional Neural Network (CNN) model was employed based on the GoogLeNet. Firstly, transfer learning was adopted for applying the prior knowledge to the expression of TAAD histopathological images. Then, Focal loss and L2 regularization were utilized to solve the data imbalance problem, so as to optimize the model performance. Experimental results show that the proposed model is able to achieve the average accuracy of four-class classification of 98.78%, showing a good generalizability. It can be seen that the proposed method can effectively improve the diagnostic efficiency of pathologists.

Aiming at the problem of privacy leakage in the application of traditional clustering algorithm, a spectral clustering algorithm based on differential privacy protection was proposed. Based on the differential privacy model, the cumulative distribution function was used to generate random noise that satisfies Laplasse distribution. Then the noise was added to the sample similarity function calculated by the spectral clustering algorithm, which disturbed the weight values between the individual samples and realized information hiding between sample individuals for privacy protection. Experimental results of UCI dataset verify that the proposed algorithm can achieve effective data clustering within a certain degree of information loss, and can also protect clustered data.

In order to enhance the reconstruction ability of regularized super-resolution technique for noisy image, an adaptive threshold denoising method was proposed based on the extended research of General Total Variation (GTV) regularized super-resolution reconstruction. Firstly, the iterative reconstruction was completed according to GTV regularized super-resolution reconstruction. Then, the deduced adaptive threshold matrix was used to divide GTV cost matrix of each iteration procedure by the threshold. The corresponding pixel points whose costs were less than the threshold continued to be iterated while the points whose costs were greater than the threshold were cut down for re-interpolating and canceled from the iteration of this turn. Finally, the reconstruction result was output when the program met the convergence requirement. The experimental results show that, compared with the single GTV regularized reconstruction method and adaptive parameter method, the proposed adaptive threshold denoising method accelerates the convergence rate and improves the quality of reconstruction image, which makes the regularized super-resolution reconstruction technology perform better for noisy image.

The Attribute Based Encryption (ABE) algorithm can support fine grained access control for cloud data. Concerning the problems that the ABE decryption has huge complexity and is difficult to realize on resource constrained mobile device, a mobile cloud storage-oriented attribute based decryption service middleware was proposed and realized. Without getting the information about the encrypted data, the middleware could delegate the ABE decryption service, a tree-based Linear Secret Sharing Scheme (LSSS) matrix solution was realized, and so the computation and communication cost of the mobile terminal were decreased. Accordingly, the decryption speed was improved. The attribute authority could revoke the user attributes through this middleware instantly with fine-grained control without involving any users. All the services were defined as Restful interface and then the generality of the middleware was ensured. The experimental results show that the middleware improves attribute based decryption performance nearly 30 times, and it has good parallel performance and practical attribute revoking capability.

Considering the problems of low load rate of data centers in cloud providers, uncertainty and variety of cloud user demand; in order to improve the average profit of the cloud providers, an overbooking model of multiple instances under uncertain demand was established. The proposed model combined the influences of overbooking for cloud data center load balancing and Service Level Agreement (SLA) under the actual cloud computing resource market, multi-constraint of overbooking was provided, then the optimal allocation policy of each instance type was put forward. The simulation results show that when the unused rate of reservation is 0.25, the average profit is relatively high, the load rate of data center is 78%, finally the optimal allocation of each instance type is determined.

Current mobile recommendation systems limit the real role of location information, because the systems just take location as a general property. More importantly, the correlation of location and the boundary of activities of users have been ignored. According to this issue, personalized recommendation technique for mobile life services based on location cluster was proposed, which considered both user preference in its location cluster and the related weight by forgetting factor and information entropy. It used fuzzy cluster to get the location cluster, then used forgetting factor to adjust the preference of the service resources in the location cluster. Then the related weight was obtained by using probability distribution and information entropy. The top-N recommendation set was got by matching the user preference and service resources. As a result, the algorithm can provide service resources with high similarities with user preference. This conclusion has been verified by case study.

The meteorological ground minute data has characteristics including various elements, large amounts of information and high frequency generation, therefore the traditional relational database system has some problems such as server overload and low read and write performance in data storage and management. With the research of storage model of distributed databases HBase, the database model of the meteorological ground minute data was proposed to achieve distributed storage of massive meteorological data and meta-information management, in which the row key was designed by the method of time plus station number. When processing the complex meteorological query case, the response time of unique index in HBase is too long. To address this defect and meet the requirements of retrieval time efficiency, with considering the query case, API interface offered by search engine solr was used to establish secondary index for related field. The experimental results show that this system has high efficiency of storage and index, the maximum storage efficiency can be up to 34000 records/s. When generic query cases return, the time consuming can be down to millisecond level. This method can satisfy the performance requirements of large-scale meteorological data in business applications.

To cope with the low accuracy of the mining results in the existing community discovery algorithms and the low quality of intelligent recommendation in the Web services resource, on the basis of the conventional collaborative filtering algorithms, a dynamic community discovery algorithm was proposed based on the nodes' similarity. Firstly, the central node that had the most connected nodes was regarded as the initial network community, and the community contribution degree was taken as the metric to continuously form a plurality of global saturated contribution degree communities. Then, an overlapping calculation was used to merge the communities of high similarity. Finally, the calculated results were arranged in descending order to form neighboring user sets for obtaining community recommendation object by calculating the dynamic similarity between target user and other users in the community. The experimental results show that the user social network community classification by the proposed community discovery algorithms is consistent with the real community classification results. The proposed algorithm can improve the accuracy of the community mining and helps to achieve high-quality community recommendation.

As surgeons do not have effective prediction on the the implant surgery for vocal cord paralysis, resulting in high rate of failure, the finite element method was used for preoperative simulation. Through Computed Tomography (CT) data of larynx, the 3D geometric model of vocal cords and glottis trachea was extracted by Mimics, and then imported into ANSYS-Fluent to simulate the vocal vibration mode and airflow dynamic coupling characteristics before and after implanted surgery. The experimental data and clinical statistics data were compared to prove the feasibility of the finite element analysis techniques for implant surgery simulation of vocal cord paralysis. The experimental result can provide support for the design of surgery program.

To guarantee reliability, traditional cloud storage solutions generally backup data through mirror redundancy, which influences the usage efficiency of storage data space. A storage solution was proposed to reduce the usage of storage data space for redundancy-backup data. The solution introduced: 1) the parity coding backup instead of mirror backup, which reduced the size of backup data; 2) the conflict-jump mechanism to confirm the backup data, which guaranteed reliability while number of backup data copies was reduced. The contrast between running result of simulation program and performance of mainstream cloud storage solutions shows that, by using the proposed solution, the usage of storage space for distributed storage is significantly reduced while the reliability gets guaranteed.

A novel approach for generating binary chaotic sequences based on 4-D chaotic systems was proposed, and the performances of these binary sequences were analyzed. The proposed algorithm was compared with others as well. The results show that the generating binary sequences have good pseudo-randomness and correlation. The binary sequences were also applied to image encryption and good results were achieved.