The neural network models based on attention mechanism are mainly used in the field of aspect-level sentiment analysis. The dependencies between aspect words and opinion words， as well as the distances between aspect words and context words， are ignored by this type of models， which further leads to inaccurate classification of emotions by this type of models. To solve above problems， a Relational and Interactive Graph ATtention network （RI-GAT） model was established. Firstly， the semantic features of sentences were learned by the Long Short-Term Memory （LSTM） network. Then the learned semantic features were combined with the position information of sentences to generate new features. Finally the dependencies between various aspects words and opinion words were extracted from the new features， realizing efficient and comprehensive use of syntactic dependency information and position information. Experimental results on Laptop， Restaurant， and Twitter datasets show that compared to the suboptimal Dynamic Multi-channel Graph Convolutional Network （DM-GCN）， RI-GAT model has the classification Accuracy （Acc） improved by 0.67， 1.65， and 1.36 percentage points， indicating that RI-GAT model can better establish the relationship between aspect words and opinion words， making sentiment classification more accurate.

At present the insufficient formal business knowledge participation in the process of On-Line Analytical Processing (OLAP) results in restriction and limitation to in-depth analysis. To overcome the limitations, a new approach for building an OLAP system was proposed based on domain ontology. Firstly, by analyzing the limitations of the existing ontology construction methods and using the similarity evaluation algorithm based on multiple features and weighted pattern of entity classes, a semi-automatic domain ontology construction method in which global top-level domain ontology was designed by experts after local ontologies were generated from databases was put forward to implement formalized description of mine-production domain knowledge. And then the key indicators of mine-production capacity were chosen as measurements and Multi-Dimensional Ontology (MDO) with business semantic concepts was built. Finally, the method was tested by a practical project of metal mine decision making system. The experimental results show that the proposed method can dynamically integrate heterogeneous information resource of mine production process and facilitate the unambiguous interpretation of query results, and discover association rules and implicit knowledge through the advantages of formalization expression and reasoning of domain ontology. Meanwhile, by high frequency and general concept views, it avoids query duplication and improves the performance of traditional OLAP systems.

Concerning the MPC8247 target system based on PowerPC, the device tree was discussed and an embedded Linux system was developed, including the transplant and deployment of U-Boot, Linux kernel, Device Tree Blob (DTB) and Ramdisk file system. The actual operation of the system shows that the device tree file is correct, and the system design is rational and efficient.

A novel infrared moving object tracking method based on particle filter and Mean Shift algorithm was presented. Firstly, it utilized the intensity distribution to represent infrared object, and constructed the observation probability model by statistical histogram. Then, Mean Shift algorithm was incorporated into the propagating process of particle filter, which induced particles distributing within the local area of observation. Compared to the conventional particle filter, the proposed method used much fewer particles to maintain the multi-mode distribution, and overcame the degeneration problem effectively. Experimental results on sequential images show that our method can track steadily when the object moves fast or is occluded, the overall performance of the proposed method is better than traditional particle filter algorithm.Experimental results on sequential images show that our method can track steadily when the object move fast or be occluded, the overall performance of the proposed method is better than traditional particle filter algorithm.

To extend the application of traditional edge extraction methods based on gradient, a new method combined with multi-scale idea was proposed. According to the response of the gradient on the edge, the integration of different scale gradients was used to define the edge feature with a new view, which was self-adaptive to the image. Next, non-maxima suppression was employed to obtain the maximal response of the integrative edge feature, which made the method adapted to strong or weak edge based on the local extremum instead of the global one. Experimental results prove that the method is much better than traditional ones and it is not sensitive to the noise.

Traditional gray projection image stabilizing algorithm just works under horizontal and vertical movement; but it is unable to deal with either scaling or rotation of the matched images. Due to the limitation, a gray projection image stabilizing algorithm based on log-polar image transform was introduced. When log-polar image transform was used in scaling or rotation image, the scaling or rotation movement in Descartes reference frame was represented by horizontal and vertical movement in log-polar reference frame. Accordingly, gray projection algorithm could be used in scaling or rotation image.