To address the problem that significant performance degradation of Transformer-based object detection methods when facing long-tailed distributions， a collaborative query optimization-based single-stage end-to-end long-tailed object detection framework， named CQ-DETR （Collaborative Query optimization-based DEtection TRansformer）， was proposed. In the framework， a Layer-Adaptive Encoder Fusion （LAEF） module was designed to integrate multi-scale encoder features dynamically， so as to consider both high-level semantics and low-level details； a Feature-aware Query Generation （FQG） module was designed to generate content-aware queries dynamically from image features， thereby enhancing the representation capability of initial content queries for potential objects； a Category-Localization Joint-aware query Selection （CLJS） mechanism was proposed to achieve collaborative optimization of category coverage and localization accuracy. Experimental results indicate that CQ-DETR is superior to RichSem （Rich Semantics） method on the long-tailed object detection benchmark dataset LVIS v1.0， with the Average Precision （AP） and the AP of rare categories （AP_r） improved by 1.4 and 1.5 percentage points， respectively， verifying the effectiveness of the proposed framework in category-imbalanced scenarios； meanwhile， on the relatively balanced COCO 2017 dataset， compared with DINO （DETR with Improved deNoising anchOr box）， CQ-DETR has the AP improved by 1.1 percentage points， verifying the good generalization ability of this framework in general object detection scenarios.

Identifying important features from high-dimensional complex industrial data is crucial for production process anomaly monitoring. Aiming at the problem that the existing feature selection algorithms are difficult to model the complex intrinsic structure of data in the face of noise disturbance， a Multi-view Consistency-driven Robust feature selection method （MCR） was proposed. Firstly， a consistency-guided denoising mechanism with structure preservation was designed， in which multi-view collaborative modeling and inconsistency region detection were used to eliminate local noise disturbance while improving structural fidelity and integrity of the raw data. Then， a joint discriminative and consistency-driven feature fusion module was constructed， where high-quality multi-view embedding representations and a feature weight matrix were learned simultaneously， thereby enhancing the ability to perceive key feature dimensions. Finally， a cooperative sparse regularization-based feature selection strategy was introduced， so as to select the most discriminative and structurally consistent subset of features from the fused embedding space. Without relying on labeled information， this method achieves perception and selection of key feature dimensions through multi-view collaborative modeling and consistency-driven optimization. Extensive experimental results on several public benchmark datasets and a real-world cigarette production dataset demonstrate that MCR outperforms the existing mainstream feature selection methods such as Binary Horse herd Optimization Algorithm （BinHOA） and Improved Binary DJaya Algorithm （IBJA）， achieving classification accuracy improvements of 0.23 to 12.15 percentage points on public datasets and 2.22 to 5.00 percentage points on real industrial dataset， validating its robustness and effectiveness in complex scenarios.

To address the issue of low channel estimation accuracy in Reconfigurable Intelligent Surface （RIS） assisted communication systems， a channel estimation scheme based on Channel Denoising Network （CDN） was proposed， which modeled the channel estimation problem as a channel noise elimination problem. Firstly， traditional algorithms were employed to estimate the received pilot signal preliminarily. Then， the estimated signals were input into the channel estimation network to learn noise features and execute denoising， thereby recovering accurate channel coefficients. Finally， to improve the denoising capability of the network， a Weighted Attention Block （WAB） and a Dilated Convolution Block （DCB） were designed to enhance the network's extraction of dominant noise features， and a multi-scale feature fusion module was designed to prevent the loss of shallow features. Simulation results demonstrate that compared with classical DnCNN （Denoising Convolutional Neural Network） and CDRN （Convolutional neural network-based Deep Residual Network） schemes， the proposed scheme reduces the Normalized Mean Square Error （NMSE） by 2.89 dB and 2.01 dB averagely at different Signal-to-Noise Ratios （SNRs）.

An improved method based on Local Binary Pattern (LBP) was proposed to solve the problem that the representing ability of LBP is bad because only the relationship between neighbors and the central pixels are considered while the floating relationship of the gray values in the neighbor region is ignored. Firstly, each neighbor was compared clockwise with its next adjacent neighbor before threshold and an LBP-like code was generated. Secondly, the code was encoded to a decimal number named as Float-LBP (F-LBP). Thirdly, the features extracted by the F-LBP and the basic LBP operators were combined together. The experimental results show that the combination of the F-LBP and the basic LBP operators can improve the retrieval accuracy by extracting more discriminative information while reserving the local micro-texture.

According to the shape features of mechanical design images, an asynchronous retrieval method based on weighted distance was proposed. The algorithm firstly got preliminary results from the image database by using the circumcircle distance feature, and then calculated the weighted distances between the input image and the preliminary results, by considering both the formal output positions and the Hu invariant moments feature. The experiments show that compared with the traditional methods, the proposed method gets higher precision and recall ratio.

The initial clustering centers of K-means algorithm are randomly selected, which may result in low accuracy and unstable clustering. To solve these problems, a K-means initial clustering center selection algorithm was proposed. The locations of data points were determined by analyzing Difference of K-dist (DK) graph. One point with the least k-dist value on the main density curves was selected as an initial clustering center. The experimental results demonstrate that the improved algorithm can select unique initial clustering center, gain stable clustering result, get higher accuracy and reduce times of iteration.

In order to improve the classification performance of minority class instances in imbalanced dataset, a new algorithm named ISMOTE (Improved Synthetic Minority Over-sampling TEchnique) was proposed. ISMOTE improved the imbalanced distribution of data through randomizing interpolation in the ball space constituted of the minority class instances and its nearest neighbor. The experiment was given on real data set. The experimental results show that the ISMOTE has substantial advantages over SMOTE (Synthetic Minority Over-sampling Technique) and direct classifying imbalanced data algorithm in prediction accuracy, and it can effectively improve the performance of classifier.

In order to meet the resolution requirement of hundreds of microseconds，microsecond global clock synchronization must be achieved in Sequence of Event (SOE) system. By means of assessing the existing clock synchronization methods and analyzing the cause of the error in the clock synchronization process, a new method based on the combination of improved Network Time Protocol (NTP) server synchronization and 1PPS synchronization was proposed. That is: advanced NTP server was used to eliminate the clock error between the control stations, "cross-second" phenomenon was avoided in this situation; 1 PPS was used to synchronize the millisecond counters in order to eliminate the crystal accumulated error of Field Programmable Gate Array (FPGA). This method is simple, precise and stable. The resolution of SOE system by using this technique can be up to 0.5 ms, and it has been applied to the turbine protection system in a power plant successfully.

Query sharing is an effective way to share the same or similar storage structures and query operations during the query procession so as to lessen the repetitive storage and resources occupation in a data stream system. For query storage sharing, a middle-result storage structure was designed, and accordingly an index-based algorithm with two-level indirect storage of the sharing queue was presented to enable the proper sharing of middle storage results, which can improve the flexibility for the data tuple to migrate as well. Meanwhile, for the multi-queries sharing, an algorithm to pick up the same query operations from several data streams was proposed, which can reduce the system processing resources by sharing the same processing resources in query operations. The model and algorithm were analyzed and discussed.

Based on the analysis of the relation between the Karush-Kuhn-Tucker (KKT) conditions of Support Vector Machine（SVM） and the distribution of the training samples, the possible changes of support vector set after new samples are added to training set were analyzed, and the generalized Karush-Kuhn-Tucker conditions were defined. Based on the equivalence between the original training set and the newly added training set, a new algorithm for SVM-based incremental learning was proposed. With this algorithm, the useless samples were discarded and the useful training samples of importance were reserved. Experimental results with the standard dataset indicate the effectiveness of the proposed algorithm.

The key to body movement emotion recognition lies in extracting emotional features existed in human body movements. To solve the problems of poor emotional feature learning capability and difficulty in improving emotion recognition accuracy in existing models， a body movement emotion recognition method based on Emotional Latent Space Learning （ELSL） and Contrastive Language-Image Pre-training （CLIP） model was proposed. Firstly， CLIP model was introduced to improve the emotional feature learning capability of the model. Secondly， for the fine-grained multi-label emotion classification task， ELSL method was proposed. By learning discriminative mappings from emotional latent space to various subspaces， the subtle differences between emotion categories and the feature information beneficial to the classification of each emotion category in various emotional subspaces. Experiments were carried out on real-world open scenarios-oriented Body Language Dataset （BoLD） The results demonstrate that the proposed method makes use of the advantages of CLIP model and latent space learning in feature learning effectively， leading to significant performance improvement. In specific， compared to Movement Analysis Network （MANet）， the proposed method has a 1.08 percentage points increase in mean Average Precision （mAP） and a 1.32 percentage points improvement in mean Area Under Receiver Operating Characteristic Curve （mRA）.