Network abnormal traffic detection is an important part of network security protection. At present， abnormal traffic detection methods based on deep learning treat the port number attribute the same as other traffic attributes， ignoring the importance of the port number. Considering the idea of attention， a novel abnormal traffic detection module based on Convolutional Neural Network （CNN） combining Port Attention Module （PAM） and Convolutional Block Attention Module （CBAM） was proposed to improve the performance of abnormal traffic detection. Firstly， the original network traffic was taken as the input of PAM， the port number attribute was separated and sent to the full connected layer， and the learned port attention weight value was obtained， and the traffic data after port attention was output by dot-multiplying with other traffic attributes. Then， the traffic data was converted into a grayscale map， and CNN and CBAM were used to extract the the channel and space information of the feature map more fully. Finally， the focus loss function was used to solve the problem of data imbalance. The proposed PAM has the advantages of few parameters， plug and play， and universal applicability. The accuracy of the proposed model is 99.18% for the binary-class classification task of abnormal traffic detection and 99.07% for the multi-class classification task on the CICIDS2017 dataset， and it also has a high recognition rate for classes with only a few training samples.

The existing Domain Generalization （DG） methods process the domain features poorly and have weak generalization ability， thus a method based on the feature independence of the frequency domain was proposed to solve the domain generalization problem. Firstly， a frequency domain decomposition algorithm was designed to obtain domain-independent features from phase information by the Fast Fourier Transform （FFT） of depth features of the image， improving the recognition ability of domain-independent features. Secondly， from the independence perspective， the correlation of attributes in frequency domain features was further eliminated by weighting the features of samples， and the most effective domain-independent features were extracted to solve the poor generalization problem caused by correlation between sample features. Finally， the amplitude fusion strategy was proposed to narrow the distance between the source domain and the target domain， so as to further improve the generalization ability of the model to the unknown domain. Experimental results on popular image domain generalization datasets PACS and VLCS show that the average accuracy of the proposed method is 0.44， 0.59 percentage points higher than that of StableNet， and the proposed method achieves excellent performance on all datasets.

A fast intra mode decision scheme for down-sizing video transcoding in H.264 based on hybrid characteristic of multi-scale videos was presented. In order to reduce the high computational complexity of using conventional intra prediction in the H.264 re-encoder, the proposed scheme firstly utilized 2D-histogram to extract the spatial characteristic of macro-blocks in the low-resolution video to choose from intra 16×16 and intra 4×4. Then the Support Vector Machine (SVM) was used to exploit the correlation between coding information extracted from the input high-resolution bit-stream and the coding modes of macro-blocks in down-sized video frames. With the SVM classifier, the improbable modes in the nine intra 4×4 modes were eliminated and only a small number of candidate modes were carried out using the RDO operations. Hence, remarkable computation time can be saved, while maintaining nearly the same quality of the transcoded pictures.

When the mechanical or architectural CAD models are simplified by using the QEM-based mesh simplification algorithm, the features of the simplification models can not be preserved very well. An improved algorithm based on features preserved was proposed. Firstly, the boundary curves were detected by using a hybrid approach to surface segmentation of the models. And the edges of the models were labeled as four types. Then different simplification strategies for each type of edges were used in the simplifying process on principle of avoiding the edges and vertex on the boundary curves to be collapsed during the simplification process. The experimental results show that the improved algorithm is effective.