To address the lack of an effective method for generating test cases for Chinese text error correction software， and to measure and optimize the correction performance of software， a multi-user engineering-oriented method was designed， called Selective Generation Method of Test cases for Chinese text error Correction Software （SGMT-CCS）. Two different criteria were defined for evaluating test cases that users can choose from： error quantity density and error type density. SGMT-CCS consists of three modules： test case automatic generation module， test case selection module， and test case priority sorting module. Users can： 1） customize the minimum error interval and the size of the test case set during the automated generation of test cases； 2） customize the minimum error interval and expected value during the selection process； 3） select different criteria for evaluating and prioritizing test cases to meet the requirements of different datasets. Experimental results show that SGMT-CCS can generate effective test cases in a short period of time. The selection module satisfies the user’s customized goals under simulated requirements， and the priority sorting module effectively improves test efficiency in different time periods under different evaluation criteria than before sorting.

YOLOv3 （You Only Look Once version 3） algorithm is widely used in target detection tasks. Although some improved algorithms based on YOLOv3 have achieved some results， there are still problems of insufficient representation ability and low detection accuracy， especially for the detection of small targets. In order to solve the above problems， a small target detection algorithm for remote sensing images based on YOLOv3 was proposed. Firstly， K-means Transformation （K-means-T） algorithm was used to optimize the size of anchor box， so that the matching degree between the priori box and ground truth box was improved. Secondly， the confidence loss function was optimized to solve the problem of uneven distribution of hard and easy samples. Finally， attention mechanism was introduced to improve the algorithm’s ability to perceive the detailed information. Results of the experiments carried out on RSOD dataset show that compared with the original YOLOv3 algorithm and YOLOv4 algorithm， the proposed algorithm has the detection Average Precision （AP） on the small target class “aircraft” increased by 7.3 percentage points and 5.9 percentage points respectively， illustrating that the proposed improved algorithm can detect small targets in remote sensing images effectively， with higher accuracy.

In the field of image recognition, images with insufficient label data cannot be well recognized by the supervised method model. In order to solve this problem, a semi-supervised method model based on Generative Adversarial Network(GAN) was proposed. That is, by combining the advantages of semi-supervised GANs and deep convolutional GANs, and replacing the sigmoid activation function with softmax in the output layer, the Semi-Supervised Deep Convolutional GAN(SS-DCGAN) model was established. Firstly, the generated samples were defined as pseudo-samples and used to guide the training process. Secondly, the semi-supervised training method was adopted to update the parameters of the model. Finally, the recognition of abnormal(cerebral infarction) images was realized. Experimental results show that the SS-DCGAN model can recognize abnormal images well with little label data, which achieves 95.05% recognition rates. Compared with Residual Network 32(ResNet32) and Ladder networks, the SS-DCGAN model has significant advantages.

Stereoscopic 3D (S3D) image retargeting aims to adjust aspect ratio of S3D images. To objectively and accurately assess the quality of different retargeted S3D images, a retargeted S3D image quality assessment database was constructed. Firstly, 45 original images were retargeted by eight representative retargeting algorithms with two retargeting scales to generate 720 retargeted S3D images. Then, the subjective quality evaluation score of each retargeted image was obtained via subjective testing. Finally, the subjective scores were converted to MOS (Mean Opinion Score) values. Based on all above, an objective quality assessment method was proposed for retargeted S3D images. In this method, three types of features including depth perception, visual comfort and image quality of left and right views were extracted to calculate the retargeted S3D image quality with the use of support vector regression prediction. Experimental results on the proposed database show that the proposed method has the Pearson linear correlation coefficient and the Spearman rank-order correlation coefficient higher than 0.82 and 0.81 respectively, demonstrating its superiority in retargeted S3D image visual quality assessment.

In order to improve clarity of foggy images and solve the problem of unicity of fractional order, a new adaptive fractional differential image enhancement method was proposed. Based on an approximate formula of Riesz fractional differential with six order accuracy, a new high precision fractional differential mask:RH operator (Riesz Higher order operator) was constructed, and the IRH operator (Improved Riesz Higher order operator) was proposed by improving RH operator. A fractional differential function was established based on image local features, and a selection criterion of fractional differential order was proposed, and then the adaptive selection method of order point by point was implemented. Combining with IRH operator, an adaptive fractional differential image enhancement algorithm was formed. For color images, due to low independence among components in RGB space, color distortion may occur after enhancement of each channel. Therefore, the image was converted from the RGB space to the HSV space and only the luminance channel was enhanced. A group of foggy images was selected compared with Tiansi operator, the segmentation-based adaptive fractional differential image enhancement algorithm and the adaptive fractional-differential compound bilateral filtering algorithm. The results show that the proposed method has obvious enhancement effect by calculating the information entropy and average gradient in comparison with methods in the reference, which further demonstrates the effectiveness of the proposed algorithm.

Visual Background extractor (ViBe)model for moving target detection cannot avoid interference caused by irregular flicker pixels noise in dynamic outdoor scenes. In order to solve the issue, a flicker pixels noise-suppression method based on ViBe model algorithm was proposed. In the initial stage of background model, a fixed standard deviation of background model samples was used as the threshold value to limit the range of background model samples and get suitable background model samples for each pixel. In the foreground detection stage, an adaptive detection threshold was applied to improve the accuracy of detection result. Edge inhibition of image edge background pixels was executed to avoid error background sample values updating to the background model in the background model update process. On the basis of above, morphological operation was added to fix connected components to get more complete foreground images. Finally, the proposed method was compared with the original ViBe algorithm and the ViBe's improvement with morphology post-processing on the results of multiple video sequences. The experimental results show that the flicker pixels noise-suppression method can suppress flicker pixels noise effectively and get more accurate results.

On the private cloud platform, it cannot be flexible to monitor and distribute the virtual machine memory resources effectively using the existing methods. To solve this problem, a Memory Monitor and Scheduler (MMS) model was put forward. And the real-time monitoring and dynamic scheduling of the virtual machine memory shortage and memory free were realized by using the libvirt function library and libxc function library provided by Xen. A small private cloud platform was built using Eucalyptus with regarding one physical machine as master node and two physical machines as child nodes. In the experiments, when the state of host was in memory shortage, MMS system effectively released the memory space by starting the virtual machine migration strategy; when the memory of the virtual machine was approaching the initial maximum memory, MMS system assigned it with a new maximum memory; when the occupied memory decreased, MMS system recycled part of free memory resource, which has little effect on the performance of virtual machines if the release memory did not exceed 150MB (maximum memory is 512MB). The results show that the MMS model of private cloud platform is effective for real-time monitoring and dynamic scheduling of the memory.

Alignment and normalization are needed in the process of floating point summation in computer. Normalization operation conducts rounding processing which will generate errors. Accumulated operations of floating-point numbers will result in the accumulation of error, the lack of precision of the calculations and even wrong calculation results. This paper discussed the influence of different binding order of floating-point numbers on the error of accumulative sum in the process of single precision floating-point accumulation through experimental methods, and aimed to explore the law of caused calculation errors by the binding sequence, provided a basis for a method of selective binding for computing paradigm and calculating structure as multi-core calculation, GPU computation and multi-processor calculation, and facilitated the advantage of parallel computing.

According to the electrocardiograph signal characteristics of human body, a wireless heart rate monitoring system based on Global System for Mobile communications (GSM) was proposed. It was composed of the acquisition terminal and the monitoring terminal. The optical pulse sensor HKG-07B of the acquisition terminal acquired pulse signal on the human body which was adjusted as the output signal and then the SPCE061A completed the data computation, storage, display and voice broadcast. On the basis of the calculated value of the heart rate, the TC35I module used the mean of short message to achieve the remote alarm received by the monitoring terminal. The experimental results show that: the accuracy of the acquired heart rate data meets the performance requirements, and real-time wireless heart rate monitoring can be achieved.

Traditional Linear Discriminant Analysis (LDA) faces the problem of tending to keep the separability of the class pairs having large within-class distances, while discarding the separability of those having small within-class distances. Based on the viewpoint that the feature subspace should uniformly keep the separability of each class pair, a new criterion, i.e., the Proportion of Divergence (PD), was presented. PD criterion was the mean of the proportion of the subspace divergence to original space divergence of each class pair. The solution of the Linear Discriminant Analysis (LDA) maximizing PD criterion (PD-LDA) was also presented. PD-LDA was used to perform feature extraction in the amplitude spectrum space of High Resolution Range Profile (HRRP). Shortest Euclidian distance classifier and Support Vector Machine (SVM) classifier were designed to evaluate the recognition performance. The experimental results for measured data show that, compared with traditional LDA, PD-LDA reduces data dimension remarkably and improves recognition rate effectively.

Concerning the characteristics of complex components of irregular images and random alignment of irregular spot without proper fitting mathematics model, an adaptive segmentation algorithm with improved Pulse Coupled Neural Network （PCNN） was proposed in the paper. On the basis of basic PCNN model, the neurons feedback input function and dynamic threshold function were modified and multi-level output model for the neuron output was designed to implement the segmentation process as well. Simulation experiment shows that the improved PCNN has better robustness and can realize the adaptive segmentation of irregular image.

The advantages of fractal geometry in representing natural objects with scrambling features and building their geometric models were introduced and the methods of producing the geometric data models of the natural objects based on fBm with mountain were exemplified. Then, how OpenGL was applied to display and render the realistic graphics of the mountain terrain was investigated. Because of the fractal features of the mountain terrains, the default function of OpenGL that could be used to compute the normals of the surfaces of the fractal models was ineffective. Thus, an effective method to remedy the flaw above was suggested. In the end, in order to prove the validity of the methods above, a prototype program was developed and a few vivid graphics produced by the program were displayed.