The existing image captioning methods only focus on the grid spatial location features without enough grid feature interaction and full use of image global features. To generate higher-quality image captions， a global image captioning method based on Graph ATtention network （GAT） was proposed. Firstly， a multi-layer Convolutional Neural Network （CNN） was utilized for visual encoding， extracting the grid features and entire image features of the given image and building a grid feature interaction graph. Then， by using GAT， the feature extraction problem was transformed into a node classification problem， including a global node and many local nodes， and the global and local features were able to be fully utilized after updating the optimization. Finally， through the Transformer-based decoding module， the improved visual features were adopted to realize image captioning. Experimental results on the Microsoft COCO dataset demonstrated that the proposed method effectively captured the global and local features of the image， achieving 133.1% in CIDEr （Consensus-based Image Description Evaluation） metric. It can be seen that the proposed image captioning method is effective in improving the accuracy of image captioning， thus allowing processing tasks such as classification， retrieval， and analysis of images by words.

The performance of Ceph system is significantly affected by the configuration parameters. In the optimization of configuration of Ceph cluster， there are many kinds of configuration parameters with complex meanings， which makes it difficult to achieve fast and accurate optimization. To solve the above problems， a parameter tuning method based on Random Forest （RF） and Genetic Algorithm （GA） was proposed to automatically adjust the Ceph parameter configuration in order to optimize the Ceph system performance. RF algorithm was used to construct a performance prediction model for the Ceph system， and the output of the prediction model was used as the input of GA， then the parameter configuration scheme was automatically and iteratively optimized by using GA. Simulation results show that compared with the system with default parameter configuration， the Ceph file system with optimized parameter configuration has the read and write performance improved by about 1.4 times， and the optimization time is much lower than that of the black box parameter tuning method.

Focusing on the problems of complex process and long response time of task offloading in multi-cloudlet mode， a computing task offloading model based on multi-cloudlet collaboration was constructed， and a Weighted self-Adaptive Inertia Weight Particle Swarm Optimization （WAIW-PSO） algorithm was proposed to solve the optimal offloading scheme quickly. Firstly， the task execution process of mobile terminal-cloudlet-remote cloud was modeled. Secondly， considering the competition of computing resources by multiple users， the task offloading model based on multi-cloudlet collaboration was constructed. Finally， since the complexity of solving the optimal offloading scheme was excessively high， the WAIW-PSO was proposed to solve the offloading problem. Simulation results show that compared with the standard Particle Swarm Optimization （PSO） algorithm and the PSO algorithm with Decreasing Inertia Weight based on Gaussian function （GDIWPSO）， WAIW-PSO algorithm can adjust the inertia weight according to evolutionary generation and individual fitness， and it has the better optimization ability and the shortest time for finding the optimal offloading scheme. Experimental results on different task unloading schemes with different numbers of equipments and tasks show that the WAIW-PSO algorithm based offloading schemes can significantly shorten the total task completion time.