Aspect-level sentiment analysis aims to classify the sentiment of text in different aspects. In the aspect-level sentiment analysis of long text, the existing aspect-level sentiment analysis algorithms do not fully extract the features of aspect related information in the long text due to the redundancy and noise problems, leading to low classification accuracy. On the datasets with coarse and fine aspects, existing solutions do not take advantage of the information in the coarse aspect. In view of the above problems, an algorithm named TFN+BERT-Pair-ATT was proposed based on text filtering and improved Bidirectional Encoder Representation from Transformers (BERT). First, the Text Filter Network (TFN) based on Long Short-Term Memory (LSTM) neural network and attention mechanism was used to directly select part sentences related to the coarse aspect from the long text. Next, the related sentences were associated with others in order, and after combining with fine aspects, the sentences were input into the BERT-Pair-ATT, which is with the attention layer added to the BERT, for feature extraction. Finally, the sentiment classification was performed by using Softmax. Compared with the classical Convolutional Neural Network (CNN) based models such as Gated Convolutional network with Aspect Embedding (GCAE) and LSTM based model Interactive Attention Network (IAN), the proposed algorithm improves the related evaluation index by 3.66% and 4.59% respectively on the validation set, and improves the evaluation index by 0.58% compared with original BERT. Results show that the algorithm based on text filtering and improved BERT has great value in the aspect-level sentiment analysis task of long text.

For the problem that the existing methods for noise type recognition and intensity estimation all focus on single noises, and cannot estimate the intensity of source noises in the mixed noises, a K-Nearest Neighbors ( KNN) algorithm with distance threshold was proposed to recognize the single and mixed noises, and estimate the intensity of source noises in the mixed noises by combining the recognition results of mixed noises and the reconstruction of noise bases. Firstly, the data distribution in frequency domain was used as feature vector. Then the signals were identified by the noise type recognition algorithm, and the frequency domain cosine distance between reconstructed noise and real noise was adopted as the optimal evaluation criterion in the process of reconstruction of noise bases. Finally, the intensity of source noises was estimated. The experimental results on two test databases indicate that, the proposed algorithm has the average accuracy of noise type identification as high as 98.135%, and the error rate of intensity estimation of mixed noise of 20.96%. The results verify the accuracy and generalization of noise type recognition algorithm as well as the feasibility of mixed noise intensity estimation algorithm, and this method provides a new idea for the mixed noise intensity estimation. The information of mixed noise type and intensity obtained by this method contributes to the determination of denoising methods and parameters, and improves the denoising efficiency.

Aiming at the low recognition accuracy and poor practicability of the traditional character feature extraction methods to International Phonetic Alphabet (IPA), a You Only Look Once (YOLO) network character recognition method with variable candidate box density for IPA was proposed. Firstly, based on YOLO network and combined with three characteristics such as the characters of IPA are closely arranged on X-axis direction and have various types and forms, the distribution density of candidate box in YOLO network was changed. Then, with the distribution density of candidate box on the X-axis increased while the distribution density of candidate box on the Y-axis reduced, YOLO-IPA network was constructed. The proposed method was tested on the IPA dataset collected from Chinese Dialect Vocabulary with 1360 images of 72 categories. The experimental results show that, the proposed method has the recognition rate of 93.72% for large characters and 89.31% for small characters. Compared with the traditional character recognition algorithms, the proposed method greatly improves the recognition accuracy. Meanwhile, the detection speed was improved to less than 1 s in the experimental environment. Therefore, the proposed method can meet the need of real-time application.

Concerning the problem that distant historical signals cannot be transmitted to the current time under the network structure of Recurrent Neural Network (RNN), Long Short Term Memory (LSTM) network was proposed as a variant of RNN. On the premise of inheriting RNN's excellent memory ability for time series, LSTM overcomes the long-term dependence problem of time series and has a remarkable performance in natural language processing and speech recognition. For the long-term dependence problem of human behavior data as a time series and the problem of not real-time detection caused by using the traditional sliding window algorithm to collect data, the LSTM was extended and applied to the human posture detection, and then a human posture detection method based on LSTM was proposed. By using the real-time data collected by the accelerometers, gyroscopes, barometers and direction sensors in the smartphones, a human posture dataset with a total of 3336 manual annotation data was produced. The five kinds of daily behavior postures such as walking, running, going upstairs, going downstairs, calmness as well as the four kinds of sudden behavior postures of fallling, standing, sitting, jumping, were predicted and classified. The LSTM network was compared with the commonly used methods such as shallow learning algorithm, deep learning fully connected neural network and convolution neural network. The experimental results show that, by using the end-to-end deep learning method, the proposed method has improved the accuracy by 4.49 percentage points compared to the model of human posture detection algorithm trained on the produced dataset. The generalization ability of the proposed network structure is verified and it is more suitable for posture detection.

In underwater acoustic communication, multipath effect channel can cause severe Inter-Symbol Interference (ISI). In view of the problems of the existing equalization algorithms when dealing with ISI, including slow convergence speed and huge steady-state error, as well as the complicated algorithm and being difficult to carry out hardware migration, a new variable step Least Mean Square (LMS) algorithm was proposed with anticosine step function and three adjustment parameters within the Feed-Forward Equalizer and Decision Feed-back Equalizer (FFE-DFE) structure. Firstly, simulations of three adjustment parameters including α, β, r were given to optimize the algorithm and compare it with traditional LMS algorithm, Modified Arctangent based Variable Step LMS (MA-VSLMS) and Hyperbolic Secant function based Variable Step size LMS algorithm (HS-VSLMS) in convergence and steady-state error. The simulation results showed that compared with the traditional LMS algorithm, the convergence speed of the proposed algorithm was 57.9% higher, and the steady-state error was reduced by 2 dB; compared with HS-VSLMS and MA-VSLMS, the convergence speed of the proposed algorithm was 26.3% and 15.8% higher, respectively, and the steady-state error was reduced by 1-2 dB. Finally, the proposed algorithm was transplanted to signal processing module and tested in an underwater experiment. Experimental results indicate that the signal is recovered very well after the equalizer, and the ISI problem caused by multipath effect is solved in the actual scene.

The current Radio Frequency IDentification (RFID) system just simply integrates the collision algorithm and security mechanism together. Based on the analysis of classical adaptive dynamic anti-collision algorithm, an anti-collision strategy of embedded security mechanism was proposed. It combined the first traversal mechanism and Boolean mutual authentication protocol to solve the problem that traditional RFID tag identification system is not efficient and has high cost; it also has high security. Compared with the backward binary, dynamic adaptive and binary tree search algorithms, the proposed strategy can greatly reduce the times of the system search and improve the label throughput.

The traditional multi-threshold image segmentation method has high complexity and the results are poor. In order to solve the problems, a new multi-thresholding method based on combining Fisher criterion and potential function was proposed. First, the Fisher criterion function was simplified and a recursive algorithm was used to reduce the computational complexity. Then the number of segmentation class was obtained by utilizing the histogram potential function method. Finally, the simplified Fisher criterion function was used for multi-thresholding and the segmentation results were further processed. The experimental results show that compared with the traditional multi-thresholding methods, the proposed method has better segmentation performance and that its running time is less. It can be used in the real-time applications.

With reference to the creation of the independent game StarMusiX, this paper provided a new method of designing game scenes in order to solve such problems as high cost and low efficiency in development and production of game scenes. The method is that external data was analyzed in real-time and the result was recognized as the driving factors of real-time game scenes building, meanwhile the details of game scenes were generated by program. In the experimental game, scenes can be generated by inputting and analyzing "music-data" and a full expression can be found in "music-visualization". This verifies the rationality and feasibility of the above methods. The experimental results indicate that the method effectively improves the efficiency of game scenes design.

It is found that real systems including Internet have the features of complex networks.A new holistic sight is given to the research of virus spreading on the complex Internet.Based on the previous work,a complex network model of Internet was constructed and the spread behavior on it was simulated.The similarity to the real spread data of Sasser shows that this model can reflect the features of the real Internet and the virus can be controlled through varying several parameters.

The proposed JPEG 2000 real-time quantization watermarking algorithm was used in an improved online bank pension distribution system. The system was based on fingerprint recognition and digital watermarking technologies. In the client side, real-time quantization watermark was embedded into the sampled fingerprint image in the JPEG 2000 coding pipeline; then the compressed bit-stream was sent to the server side. In the server side, the watermark was extracted from the compressed bit-stream in the JPEG 2000 decoding pipeline; then the decompressed fingerprint image and extracted watermark were used to verify users identification. Experiments showed when typical fingerprint image was compressed to 1/4~1/20 of its original size, the embedded watermark could be exactly extracted, and fingerprint recognition rate remained almost the same after lossy compression. The system has a better interaction performance in the band-limited network situation, and is very promising in the E-business applications.

Semiconductor manufacturing process's complexity and randomness make it difficult to build determinate prediction model. A new method was presented which used RBF neural network to model this process. Manufacturing lines with various release control and scheduling policies were simulated by software simul8, and the sampling data got from the simulation model were used in the training and test of the prediction model. Results demonstrate that the model's output and the real samples output are basically identical and the model has great generalization ability. So the well-trained network can be used to forecast the state of the process rapidly and accurately, which lays foundation to prediction control and real time scheduling.

Auther proposed a brand new encryption method for video sequence, based on Chaos theory. This method adopted multilayer chaotic transformation,therefore,the key can be changed during theencryption process, and the security performance is greatly improved.Besides, this algorithm inherited the selective encrytion idea, it only processed the key messages in the video sequence, so encryption efficiency is enhanced remarkably to satisfy the requirements of realtime communication. Furthermore, through computer simulations, the security performance and encryption efficiency of this method was demonstrated.