Emotion recognition is a technology that allows computers to recognize and understand human emotions. It plays an important role in many fields and is an important development direction in the field of artificial intelligence. Therefore， the research status of bimodal emotion recognition based on speech and text was summarized. Firstly， the representation space of emotion was classified and elaborated. Secondly， the emotion databases were classified according to their emotion representation space， and the common multi-modal emotion databases were summed up. Thirdly， the methods of bimodal emotion recognition based on speech and text were introduced， including feature extraction， modal fusion， and decision classification. Specifically， the modal fusion methods were highlighted and divided into four categories， namely feature level fusion， decision level fusion， model level fusion and multi-level fusion. In addition， results of a series of bimodal emotion recognition methods based on speech and text were compared and analyzed. Finally， the application scenarios， challenges， and future development directions of emotion recognition were introduced. The above aims to analyze and review the work of multi-modal emotion recognition， especially bimodal emotion recognition based on speech and text， providing valuable information for emotion recognition.

With the rapid development of Large Language Models （LLMs）， dialogue assistants based on LLM have emerged as a new learning method for students. These assistants generate answers through interactive Q&A， helping students solve problems and improve learning efficiency. However， the existing conversational assistants ignore students’ personalized needs， failing to provide personalized answers for “tailored instruction”. To address this， a personalized conversational assistant framework based on student capability perception was proposed， which is consisted of two main modules： a capability perception module that analyzes students’ exercise records to explore the knowledge proficiency of the students， and a personalized answer generation module that creates personalized answers based on the capabilities of the students. Three implementation paradigms — instruction-based， data-driven， and agent-based ones were designed to explore the framework’s practical effects. In the instruction-based assistant， the inference capabilities of LLMs were used to explore knowledge proficiency of the students from students’ exercise records to help generate personalized answers； in the small model-driven assistant， a Deep Knowledge Tracing （DKT） model was employed to generate students’ knowledge proficiency； in the agent-based assistant， tools such as student capability perception， personalized detection， and answer correction were integrated using LLM agent method for assistance of answer generation. Comparison experiments using Chat General Language Model （ChatGLM） and GPT4o_mini demonstrate that LLMs applying all three paradigms can provide personalized answers for students， the accuracy of the agent-based paradigm is higher， indicating the superior student capability perception and personalized answer generation of this paradigm.

Network traffic anomaly detection is a network security defense method that involves analyzing and determining network traffic to identify potential attacks. A new approach was proposed to address the issue of low detection accuracy and high false positive rate caused by imbalanced high-dimensional network traffic data and different attack categories. One Dimensional Convolutional Neural Network（1D-CNN） and Bidirectional Gated Recurrent Unit （BiGRU） were combined to construct a model for traffic anomaly detection. For class-imbalanced data， balanced processing was performed by using an improved Synthetic Minority Oversampling TEchnique （SMOTE）， namely Borderline-SMOTE， and an undersampling clustering technique based on Gaussian Mixture Model （GMM）. Subsequently， a one-dimensional CNN was utilized to extract local features in the data， and BiGRU was used to better extract the time series features in the data. Finally， the proposed model was evaluated on the UNSW-NB15 dataset， achieving an accuracy of 98.12% and a false positive rate of 1.28%. The experimental results demonstrate that the proposed model outperforms other classic machine learning and deep learning models， it improves the recognition rate for minority attacks and achieves higher detection accuracy.

In order to solve the problems of slow accurate search speed due to mixed data storage and difficult security governance caused by unclassified and graded data management， a data classified and graded access control model based on master-slave multi-chain was built to achieve classified and graded protection of data and dynamic secure access. Firstly， a hybrid on-chain and off-chain trusted storage model was constructed to balance the storage bottleneck faced by blockchain. Secondly， a master-slave multi-chain architecture was proposed and smart contracts were designed to automatically store data with different privacy levels in the slave chain. Finally， based on Role-Based Access Control， a Multi-Chain and Level Policy-Role Based Access Control （MCLP-RBAC） mechanism was constructed and its specific access control process design was provided. Under the graded access control policy， the throughput of the proposed model is stabilized at around 360 TPS （Transactions Per Second）. Compared with the BC-BLPM scheme， it has a certain superiority in throughput， with the ratio of sending rate to throughput reaching 1∶1. Compared with no access strategy， the memory consumption is reduced by about 35.29%； compared with the traditional single chain structure， the memory average consumption is reduced by 52.03%. And compared with the scheme with all the data on the chain， the average storage space is reduced by 36.32%. The experimental results show the proposed model can effectively reduce the storage burden， achieve graded secure access， and suitable for the management of multi-class data with high scalability.

The target size of the Unmanned Aerial Vehicle （UAV） is small， and the characteristics among multiple UAVs are not obvious. At the same time， the interference of birds and flying insects brings a huge challenge to the accurate detection and stable tracking of the UAV targets. Aiming at the problem of poor detection performance and unstable tracking of small target UAVs by using traditional target detection algorithms， a real-time tracking algorithm for multiple UAVs based on improved PaddlePaddle-YOLO （PP-YOLO） and Simple Online and Realtime Tracking with a Deep association metric （Deep-SORT） was proposed. Firstly， the squeeze-excitation module was integrated into PP-YOLO detection algorithm to achieve feature extraction and detection of UAV targets. Secondly， the Mish activation function was introduced into ResNet50-vd structure to solve the problem of vanishing gradient in the back propagation process and further improve the detection precision. Thirdly， Deep-SORT algorithm was used to track UAV targets in real time， and the backbone network that extracts appearance features was replaced with ResNet50， thereby improving the original network’s weak perceptual ability of small appearances. Finally， the loss function Margin Loss was introduced， which not only improved the class separability， but also strengthened the tightness within the class and the difference between classes. Experimental results show that the detection mean Average Precision （mAP） of the proposed algorithm is increased by 2.27 percentage points compared to that of the original PP-YOLO algorithm， and the tracking accuracy of the proposed algorithm is increased by 4.5 percentage points compared to that of the original Deep-SORT algorithm. The proposed algorithm has a tracking accuracy of 91.6%， can track multiple UAV targets within 600 m in real time， and effectively solves the problem of "frame loss" in the tracking process.

Nitrogen oxide （NOx） is one of the main pollutants in the regenerated flue gas of Fluid Catalytic Cracking （FCC） unit. Accurate prediction of NOx emission can effectively avoid the occurrence of pollution events in refinery enterprises. Because of the non-stationarity， nonlinearity and long-memory characteristics of pollutant emission data， a new hybrid model incorporating Ensemble Empirical Mode Decomposition （EEMD） and Long Short-Term Memory network （LSTM） was proposed to improve the prediction accuracy of pollutant emission concentration. The NOx emission concentration data was first decomposed into several Intrinsic Mode Functions （IMFs） and a residual by using the EEMD model. According to the correlation analysis between the IMF sub-sequences and the original data， the IMF sub-sequences with low correlation were eliminated， which could effectively reduce the noise in the original data. The IMFs could be divided into high and low frequency sequences， which were respectively trained in the LSTM networks with different depths. The final NOx concentration prediction results were reconstructed by the predicted results of each sub-sequences. Compared with the performance of LSTM in the NOx emission prediction of FCC unit， the Mean Square Error （MSE）， Mean Absolute Error （MAE） were reduced by 46.7%， 45.9%，and determination coefficient （R²） of EEMD-LSTM was improved by 43% respectively， which means the proposed model achieves higher prediction accuracy.

The effective storage mode for time series was designed on Hadoop Distributed File System (HDFS), the sub-series were distributed to the compute nodes on Hadoop cluster by applying Distributed Cache tool, and the matrix of dynamic time warping distances was partitioned into several sub-matrixes. Based on MapReduce programming mode, by parallel computing sub-matrixes in each back-diagonal iteratively, the parallel computation of dynamic time warping distances was implemented, and an efficient parallel algorithm for searching similar patterns from data streams was developed by improving pruning redundant computation. The experimental results on the data set of snow depth long time series in China show that when the length of each time series is equal to or longer than 5000, the required time of parallel computing dynamic time warping distances is less than that of the corresponding sequential computation, and when the length of each time series is equal to or longer than 9000, the more the compute nodes used, the less the required parallel computation time; furthermore, when the length of each pattern is equal to or longer than 4000 and the number of compute nodes is equal to or larger than 5, the required time of parallel searching similar sub-series from data streams is 20% of the corresponding sequential searching time.

Focusing on the issue that non-zero genus surface parameterization has large deformation and high computational complexity, an improved global parameterization approach based on holomorphic 1-form was proposed, which starts from the gradient field and adapts easier and faster method to compute homology and cohomology group. Firstly, a simplified cut graph method was used to construct homology group to determine the topology. Secondly, cohomology group of the linear space formed by the gradient field was calculated by defining special harmonic function to figure out closed 1-form. Thirdly, homology group was diffused to harmonic 1-form through minimizing harmonic energy. Finally, holomorphic 1-form was computed by combining linearly harmonic 1-form and the parameterization was obtained by integrating holomorphic 1-form on the surface basic domain. Theoretical analysis of homology group and cohomology group shows that the parameterization is a global, border-free conformal mapping. Experimental results based on non-zero genus model show that, compared with the former global parameterization based on holomorphic 1-form, the proposed algorithm has better visual effect, smaller average error and higher operation efficiency.

According to the capacity of multi-level caches, the population individuality and ant data in CPU main memory were assigned to L3 cache, L2 cache and L1 cache to reduce data transfer overhead among multiple caches during parallel computing. The asynchronous and incomplete transmission was performed between CPU and GPU, and multiple flows were asynchronously executed by multiple GPU kernel functions. The thread number of GPU block was set to the size of 16 times and GPU public memory was divided into bank with the size of 32 times. GPU constant memory was used to store read-only parameters such as cross probability and mutate probability which were read frequently. The read-only big data structure such as string set and overlap matrix were bound to GPU texture memory, and a computation, cache and communication-efficient parallel algorithm for CPU and GPU to coordinate solving shortest common superstring problem was designed and implemented. The experimental results for solving shortest common superstring problem with several sizes show the proposed CPU and GPU parallel algorithm is faster over 70 times than the sequential algorithm.

In order to solve the problem of privacy disclosure in medical image's interest of region, a new reversible visible watermarking based privacy detection algorithm was proposed. The method embedded a binary watermark image into the interest of region of original medical image to protect privacy, and used visual masking of Human Visible System (HSV) and pixel's mapping mechanism to dynamically adjust the visibility and transparence of watermark. In addition, a shrinking projection technology was utilized to solve the problem of potential overflow and underflow during the embedding procedure. Finally, a random key was introduced to enhance the embedded watermark's robustness. The experimental results show that the proposed method achieves better performance in visibility and transparence of watermark, and the number of additional information produced by embedding is only 65608bits. In addition, the deletion of watermark is difficult without knowing the correct key, and the quality difference between the watermarked image and the recovered image is less than 1dB.

To achieve the goal that personal and original information is not disclosed to each other when several parties cooperatively mine several data tables at different computational sites, based on secure triple-party protocol, a triple-site cross multi-table frequent item sets mining algorithm with privacy preserving was proposed in distributed environment with multiple tables at each site. The proposed algorithm disturbed data by generating random numbers, mined frequent item sets of inter-site in parallel, and linked the data with equal-value by common link attribution of the tables among the sites and applied secure protocol to compute the global support of inter-site cross-table frequent item sets. The experimental results show that the proposed algorithm is efficient, and it can not only mine the cross multi-table frequent item sets, but also preserve the private data at each site.

After analyzing inverted index structure and access mode of an open source search engine-ASPSeek, this paper gave an abstract definition of "inverted index". In order to solve the difficulties of inverted index updating and the efficiency issues caused by directly accessing inverted index through file caching of operating system in ASPSeek, considering the characteristics of 1.25 million Chinese agricultural Web pages, this article proposed a new blocking inverted index storage scheme with a buffer mechanism which was based on CLOCK replacement algorithm. The experimental results show that the new scheme is more efficient than ASPSeek whether the buffer system is disabled or enabled. When the buffer system got enabled and 160 thousand Chinese terms or 50 thousand high-frequency Chinese terms were used as a test set, the retrieval time of new scheme tended to be a constant after one million accesses. Even when using entire 827309 terms as a test set, the retrieval time of new scheme began to converge after two million accesses.

In this paper, the main idea was to improve the foreground detection method based on statistical model. On one hand, historical maximum probability of which feature vector belongs to background was recorded in the background model, which could improve the matched vectors updating speed and make it blended into the background quickly. On the other hand, a method using spatial feature match was proposed to reduce the shadow effect in the foreground detection. The experimental results show that, compared with the MoG method and Lis statistical model method, the method proposed in this paper has obvious improvement in shadow remove and obscured background restoration of big target object.

A data distribution strategy and a communication-efficient parallel algorithm for sorting integers sequence were proposed on the heterogeneous cluster with multi-core machines. The presented data distribution model properly utilized different computation speed, communication rate and memory capacity of each computing node to dynamically compute the size of the data block to be assigned to each node to balance the loads among nodes. In the proposed parallel sorting algorithm, making use of the characteristic of integers sequence, master node distributed the data blocks to the salve nodes and received the sorted subsequences with two-round mode, each salve node returned its sorted subsequence to master node by bucket-packing method, and master node linked its received sorted subsequences to form directly a final sorted sequence by the bucket mapping in order to reduce the data merge operations with large communication cost. The analysis and experimental results on the heterogeneous cluster with multi-core machines show that the presented parallel sorting integers sequence algorithm is efficient and scalable.

A new reversible data hiding algorithm for medical images was proposed. The hidden information was embedded into Region Of Interest (ROI) and non-interest respectively. In ROI, an adaptive integer transform scheme was employed to enhance the embedding capacity and control distortions. And in Region of Non-Interest (RONI), the classical Least Significant Bit (LSB) method was used to keep the marked image’s quality. The experimental results show that, compared with previous works, the performance of the proposed method has been significantly improved in terms of capacity and image quality. The proposed method’s embedding capacity is between 1.2bpp and 1.7bpp, while the Peak Signal-to-Noise Ratio (PSNR) can maintain the 43dB or so. Moreover, the proposed method with high run efficiency can be applied into the practical hospital information system.

Some E-commerce sites cannot deploy independent recommender systems themselves due to limited resources. In order to help these sites deploy recommender systems quickly and conveniently, an E-commerce recommendation platform based on Software-as-a-Service (SaaS) pattern was proposed and implemented. This platform used unified scripts to collect user action information and provided recommendation services through standard interface. It realized a low coupling between platform and E-commerce sites so that the cost of implementation was reduced. The results of online operations show that this platform can help E-commerce sites increase the conversion rate and the volume of orders.

The analysis results on PPStream-VOD System client behavior characteristics were presented in this paper. This study began from researching on peer-distributing protocol and the architecture of Buffer-Map based on passive measurement. A dedicated PPS-VOD crawler was deployed to capture clients’ Buffer-Map and study the characteristics of client watching behavior. By accurate data analysis, the client behavior was classified as Long-Smoother, Short-Smoother and Jumper. Then the proportion of three kinds of clients and their different watching behaviors were proposed. The concept watching viscosity was put forward to reveal the attraction of program to users, which is in direct proportion to average watching time, and in inverse proportion to slope of probabillty accumulation curve.

A Web news recommendation method based on multiple topic tracking was proposed to improve the precision of recommendation. The proposed algorithm used multiple user profiles to represent user's interests in different topics, and dynamically updated user's profile to reflect the changing of user's interests. The central recommendation algorithm was implemented, and experiments on Reuters Corpus Volume 1 were carried out. The experimental results show that the proposed algorithms can effectively improve the precision of recommendation.

Based on the intrinsic features of spatial distribution, temporal distribution and high dynamic of parking activities, a negotiation approach was introduced to the design of an intelligent parking guidance system. The IEEE FIPA compliant multi-Agent system called active negotiation-based intelligent parking guidance system (AIPGIS) was proposed. The architecture, operation mechanism, negotiation algorithms and characteristics were analyzed and presented. The AIPGIS can implement effective sharing of urban traffic state information and strengthen the coordination and decision-making capacities of the active Agents.

The types of embedded device platform become more and more diversified. In order to make GIS systems developed run on several kinds of platforms, this paper put forward a cross-platform embedded GIS data model. This model isolated the data processing part, which was irrelevant to display, and based on which, it was further divided into higher layer interface model and lower layer data engine model. This model is able to not only meet user's requirement of secondary development but also cross several operating systems and integrate several heterogeneous GIS data resources. Finally, the platform-independence of the model is certified, through the implementation of a traffic navigation system.