In order to solve the problem of poor small target detection effect in Synthetic Aperture Radar (SAR) target ship detection, a self-adaptive anchor single-stage ship detection method was proposed. Firstly, on the basis of Feature Selective Anchor-Free (FSAF) algorithm, the optimal feature fusion method was obtained by using the Neural Architecture Search (NAS) to make full use of the image feature information. Secondly, a new loss function was proposed to solve the imbalance of positive and negative samples while enabling the network to regress the position more accurately. Finally, the final detection results were obtained by combining the Soft-NMS filtering detection box which is more suitable for ship detection. Several groups of comparison experiments were conducted on the open SAR ship detection dataset. Experimental results show that, compared with the original target detection algorithm, the proposed method significantly reduces the missed detections and false positives of small targets, and improves the detection performance for inshore ships to a certain extent.

An image can be divided into foreground part and background part, while the foreground is often the visual center. Due to the large categories and complex situations of foreground part, the image colorization is difficult, thus the foreground part of an image may suffer from poor colorization and detail loss problems. To solve these problems, an image colorization algorithm based on foreground semantic information was proposed to improve the image colorization effect and achieve the purpose of natural overall image color and rich content color. First, the foreground network was used to extract the low-level features and high-level features of the foreground part. Then these features were integrated into the foreground subnetwork to eliminate the influence of background color information and emphasize the foreground color information. Finally, the network was continuously optimized by the generation loss and pixel-level color loss, so as to guide the generation of high-quality images. Experimental results show that after introducing the foreground semantic information, the proposed algorithm improves Peak Signal-to-Noise Ratio (PSNR) and Learned Perceptual Image Patch Similarity (LPIPS), effectively solving the problems of dull color, detail loss and low contrast in the colorization of the central visual regions; compared with other algorithms, the proposed algorithm achieves a more natural colorization effect on the overall image and a significant improvement on the content part.

The uneven distribution of cell staining and the diversity of tissue morphologies bring challenges to the automatic segmentation of Hematoxylin-Eosin (HE) stained pathological images. In order to solve the problem, a three-step hierarchical segmentation method of pathological images based on self-supervised learning was proposed to automatically segment the tissues in the pathological images layer-by-layer from coarse to fine. Firstly, feature selection was performed in the RGB color space based on the calculation result of mutual information. Secondly, the image was initially segmented into stable and fuzzy color regions of each tissue structure based on K -means clustering. Thirdly, the stable color regions were taken as training datasets for further segmentation of fuzzy color regions by naive Bayesian classification, and the three complete tissue structures including nucleus, cytoplasm and extracellular space were obtained. Finally, precise boundaries between nuclei were obtained by performing the mixed watershed classification considering both shape and color intensities to the nucleus part, so as to quantitatively calculate the indicators such as the number of nuclei, nucleus proportion, and nucleus-cytoplasm ratio. Experimental results of HE stained meningioma pathological image segmentation show that， the proposed method is highly robust to the difference of staining and cell morphologies, the error of issue area segmentation is within 5%, and the average accuracy of the proposed method in nucleus segmentation accuracy experiment is above 96%, which means that the proposed method can meet the requirements of automatic analysis of clinical images and its quantitative results can provide references for quantitative pathological analysis.

Aiming at the problem that the sub-health state of subway door is difficult to identify, a sub-health state identification method based on time series data mining was proposed. First of all, the angle, speed and current data of the subway door motor were discretized by combining multi-scale sliding window method and Extension of Symbolic Aggregate approXimation (ESAX) algorithm. And then, the features were obtained by calculating the distances among the templates under the normal state of the subway door, in which the Principal Component Analysis (PCA) was adopted to reduce feature dimension. Finally, combining with basic features, a hierarchical pattern recognition model was proposed to identify the sub-health state from coarse to fine. The real test data of subway door were taken as examples to verify the effectiveness of the proposed method. The experimental results show that the proposed method can recognize sub-health state effectively, and its recognition rate can reach 99%.

Aiming at improving the users' high viscosity in social networks, an algorithm based on user loyalty in social network system was proposed. In the proposed algorithm, double Recency Frequency Monetary (RFM) model was used for mining the different loyalty kinds of users. Firstly, according to the double RFM model, the users' consumption value and behavior value were calculated dynamically and the loyalty in a certain time was got. Secondly, the typical loyal users and disloyal users were found out by using the founded standard curve and similarity calculation. Lastly, the potential loyal and disloyal users were found out by using modularity-based community discovery and independent cascade propagation model. On some microblog datasets of a social network, the quantitative representation of user loyalty was confirmed in Social Network Service (SNS), thus the users could be distinguished based on users' loyalty. The experimental results show that the proposed algorithm can be used to effectively dig out different loyalty kinds of users, and can be applied to personalized recommendation, marketing, etc. in the social network system.

To improve the accuracy of traditional method for customer segmentation, the Large Average Submatrix (LAS) biclustering algorithm was used, which performed clusting on customer samples and consumer attributes simultaneously to identify the upscale and high-value customers. By introducing a new value yardstick and a novel index named PA, the LAS biclustering algorithm was compared with K-means clustering algorithm based on a simulation experiment on consumption data of a telecom corporation. The experimental result shows that the LAS biclustering algorithm finds more groups of high-value customers and obtains more accurate clusters. Therefore, it is more suitable for recognition and segmentation of high-value customers.

Recently, the research on clothing simulation is becoming hotter. But the flexibility, sense of reality, real-time and integrity are always difficult to be unified. Therefore, a new dressing simulation system was designed concerning the automatic fitting of any human body and clothes. At first, the surface of Non-Uniform Rational B-Spline (NURBS) was used to complete deformable body modeling. Then, particles were reconstructed from the 3DMAX model and multi-type springs were created to complete arbitrary cloth modeling. Finally, Verlet integrator was adopted to complete dressing simulation, while a new simplification algorithm for cloth models and a new method for judging interior point with a triangle were implemented. The results show that the proposed modeling approach for body and clothes guarantees the diversity of dressing effect, and the model simplification and interior point judgment can increase the simulation performance by 30% or so, which ensures the real-time quality.

The deficiency of Gaussian Mixture Model (GMM) is the high computation cost and cannot deal with the shadow and ghosting. An improved foreground detection algorithm based on GMM is proposed in this paper. By analyzing the stability of the background, intermittent or continuous frame updating is chose to update the parameters of the GMM.It can efficiently reduce the runtime of the algorithm. In the background updating,the updating rate is associated with the weight and this makes it change with the weight.The background pixels which appear after the objects moving set a larger updating rate.It can improve the stability of the background and solve the problem of ghosting phenomenon and the transformation of background and foreground.After objects detection,the algorithm eliminates the shadow based on the RGB color space distortion model and treats the result by Gauss Pyramid filtering and morphological filtering.Through the whole process,a better contour is obtained. The experimental results show that this algorithm has improved the calculation efficiency and accurately segmented the foreground object.