MRI technology can obtain the brain avatar of patients from multiple modes and angles, which greatly improves the diagnostic efficiency of brain tumor patients. However, the MRI data of brain tumor patients is usually not enough to train the deep learning based segmentation model for sufficient performance. This paper proposes a simple and effective method to augment the data to improve the image segmentation effect of UNet++ model on brain glioma. To enhance the illumination robust of MRI image, we bring in the max filter and min filter for image augmentation, and evaluate the segmentation effect of the UNet++ model using the original dataset and the augmented dataset on glioma, and evaluate the experimental results. We used the public datasets MICCAI_ BraTS_20 to evaluate, The WT PPV values of the two models were 0.9233 and 0.9427 respectively, and other evaluation indicators were also improved to varying degrees. In summary, it is concluded that our proposed simple augmentation method could effective augment the brain glioma MRI segmentation performance of UNet++ model on glioma, especially in the limited data.
The computer vision field is quickly developing, including multiple object tracking, as the big data age approaches. The majority of the effort is focused on tracking methods while less attention is paid to the most important aspect, data. After an analysis of existing datasets, we find that they commonly ignore the breakpoint problem in tracking and have low image quality. Thus we present the dataset named Multiple Object Tracking on Horizontal view (MOT-H). MOT-H is meticulously annotated on crowded scenes from the horizontal view, with the primary goal of proving anti-jamming performance against complicated occlusion or even complete occlusion. The breakpoint issue is emphasized, which means the target object temporarily leaves the scene and returns after a while. The proposed MOT-H dataset has 10 sequences, 20,311 frames, and 337,440 annotation boxes in total, with all pictures having the resolution of 3840 × 2160 and being filmed at 30 frames per second (fps). We establish a fair benchmark for the future object tracking method development. The whole dataset can be found at: https://drive.google.com/drive/folders/1SCUJAdbqXQStyV-F2M9UyGfsuCaxR73a?usp=sharing.
Part segmentation is one of the important tasks in 3D shape analysis. Prior works mostly rely on complex local modeling to learn features directly from raw mesh or point cloud. We proposed a simple, fast and robust approach for 3D shape part segmentation without sophisticated local geometric modeling and ingenious networks. Our main idea is to learn more discriminative features from different geometric descriptors by sparse convolution and attention mechanism. Specifically, an effective shape representation block, consists of a features embedding module, two attention module. The coordinates are fed into the features embedding module to produce embedding vectors. The attention head to learn the intrinsic relations between different descriptors and produce more informative feature maps. In the lower branch, We use some geometric descriptors as local features. And a classification head predicts the corresponding label. Conditional Random Field (CRF) is applied to optimize the segmentation of the network. The experiment results on Princeton Shape Benchmark(PSB) demonstrate that our architecture outperforms different methods, with higher accuracy, lower complexity and faster speed.
Accurate traffic prediction has a wide spectrum of benefits for the society. As Electric Vehicles (EV) are being used more frequently, an accurate EV Charging Station availability forecast becomes necessary. These forecasting models can help alleviate congestion at charging stations and maneuver EV drivers to an ideal location. There have been many data-driven models applied to similar problems related to traffic prediction [1]. Despite numerous efforts to ameliorate already established models and to innovate new models, the complex topological structure of the road networks and the variations of temporal data hinder the models from achieving a good prediction. As for the most classic Recurrent Neural Network (RNN) only extracts time-series information, its lack of information related to spatial dependencies cause a significant loss in long-term accuracy. The Graph Neural Network (GNN) can perform well in extracting spatial dependencies but cannot process time-series information well. Combining the Graph Convolutional Network (a variant of GNN) and the Transformer, we seek to efficiently capture both spatial and temporal data related EV Charging Station Availability. The model is tested for its performances using the Dundee City dataset. And the result reflects that our model surpassed an accuracy of 80% and attained more accurate predictions than the classic baselines.
Proteins are critical for lives. They not only build 10%-35% of our body tissues, but also can be used to understand the structures of different viruses, and then help us to explore effective vaccines. Hence, predicting new protein structures is very important for human health. However, the structure of protein is complicated. Exploration using human experiments is cost-consuming. Recently, artificial intelligence (AI) technology, such as imitation learning and reinforcement learning (RL), has been rapidly developed and significantly improved the efficiency in many different domains. In this project, we will try to use RL to solve the protein folding structure prediction problem.
Software test case selection is committed to select the fewest test cases from test suites to perform a complete test at the least cost. Machine learning and multi-objective optimization techniques have developed rapidly in recent years, and they have been successfully applied to test case selection. In this paper, we present a method called DB-NSGA2, which uses the density clustering algorithm in machine learning combined with the non-dominated ranking algorithm (NSGA2) for test case selection, which can better select the test cases required for testing. In particular, we apply some of the clustering results generated by the clustering algorithm to the crossover and mutation operations of the NSGA2 to improve diversity progeny populations and ensure the transmission of good individuals. Extensive experiments show that the test cases selected by our method can produce a better set of Pareto solutions and can detect more faults at a lower cost than other methods.
When dealing with garbage image classification task, with the deepening of network layer, convolutional neural network will lead to gradient disappearance / explosion and large time consumption. Therefore, a garbage image classification method combining improved convolutional neural network and long short-term memory network is proposed. Taking ResNet-50 as the network backbone, it is optimized by using deep separable convolution and attention mechanism. At the same time, supplemented by LSTM, the features extracted by convolution network and cyclic network are fused to complete classification output. Ablation experiments are carried out on this model and compared with other typical convolutional neural networks. The results show that the accuracy of this model increases by an average of 4.5%. The introduction of deep separable convolution can reduce the training time by about 35.4% compared with the baseline method.
From daily online payments to large financing transactions, online financial services make business behaviour convenient and fast. At the same time, attackers can use many means to damage this system. The Society for Worldwide Interbank Financial Telecommunications (SWIFT) system sets up a series of message standards to regulate the communication message format between all users, which greatly improves the business efficiency. Hackers with financial knowledge can get more familiar with SWIFT messages and make the traditional cyber-attacks more targeted or develop more threatening trojan viruses. This paper focuses on studying the cyber-attacks on SWIFT systems. The paper presents three different attacks and discuss how each attack can damage SWIFT users and profit the hackers.
Software testing is an important means to ensure software quality. The quality and efficiency of software testing can be greatly improved by modeled software testing. Software test maturity model (TMM) is a reference model to guide software organizations to improve test maturity. However, there is a lack of guidance on software testing objectives and process improvement, which leads to poor enforceability and low execution efficiency. To solve the above problems, based on the maturity objectives and content of the five test levels of the TMM model, an improved software testing V model (RAD) is proposed, and a software quality evaluation method is proposed for the improved RAD model.
research-article Share on A Novel Image perceptual hashing algorithm based on frequency decomposition and LoG Authors: Zihao Yang School of Computer Science and Technology, Jiangsu Normal University, China School of Computer Science and Technology, Jiangsu Normal University, China 0000-0001-6046-6433View Profile , Guosheng Hao School of Computer Science and Technology, Jiangsu Normal University, China School of Computer Science and Technology, Jiangsu Normal University, China 0000-0002-5962-0832View Profile , Xiaoyun Zhou School of Computer Science and Technology, Jiangsu Normal University, China School of Computer Science and Technology, Jiangsu Normal University, China 0000-0002-7144-072XView Profile , Wang Ruan School of Computer Science and Technology, Jiangsu Normal University, China School of Computer Science and Technology, Jiangsu Normal University, China 0000-0003-0263-9353View Profile Authors Info & Claims CSSE '22: Proceedings of the 5th International Conference on Computer Science and Software EngineeringOctober 2022Pages 300–305https://doi.org/10.1145/3569966.3570057Published:20 December 2022Publication History 0citation14DownloadsMetricsTotal Citations0Total Downloads14Last 12 Months14Last 6 weeks2 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access
Aiming at the problem that the temperature of display screen of a shipboard cabinet is higher than the limit demand, the thermal simulation software ICEPAK is used to carry out numerical analysis of the internal temperature field of the cabinet and the modified heat dissipation design of the air-cooled channel. The numerical results show that the cooling channel design of the original cabinet is effective for the heat dissipation of the internal heat source of the cabinet, but it can't meet the heat dissipation requirements of the display panel. Through the design of the modified cooling channel of the original cabinet, it is found that the modified cooling channel not only reduce the temperature of the internal heat source of the cabinet but also effectively reduce the maximum temperature of the display panel by about 25℃. The numerical results have important guiding significance for the safe operation design of the shipboard cabinet.
Owing to the complex water environment, the acoustic point-cloud model formed by the detection method based on acoustic reflection mechanism is inevitably disturbed by the noise, which seriously affects the reconstruction effect of the underwater targets. Distinguishing between geometric features and noise is of paramount importance for the underwater point-cloud model filtering. Inspired by the classic image detail enhancement method of unsharp masking, we take the geometric coordinate information of the point as the research object and design a geometric feature-preserving adaptive unsharp masking filtering for the underwater point-cloud model. First, the proposed method directly performed a low-pass filtering using the neighborhood information to obtain the main structure of the input point-cloud model. Second, the detail layer was yielded by the difference between the input point-cloud model and the base layer. Third, the different scaling factors measuring the importance of the points with respect to the whole base layer were used to adaptively enhance the detail layer. Experimental results show that the proposed algorithm can effectively remove noise while maintaining the geometric characteristics of the model, which is obviously better than other comparison methods.
Garbage sorting plays an important and far-reaching role in environmental protection and resource regeneration. At present, garbage sorting is mainly manual assistance. Therefore, because of the characteristics of different object sizes in garbage images, this paper proposes a garbage object detection method based on improved Faster R-CNN. ResNet50 is the backbone network, and a Feature Pyramid Network structure is added to the model. Modified RPN structure parameters, and the original ROI pooling layer is changed to ROI align layer to achieve effective extraction of image features. The experiment shows that the improved Faster R-CNN method with ROI Align and FPN structure is effective in garbage object detection. ROI Align increased by 0.8 percentage points and FPN structure increased by 0.96 percentage points. Finally, the improved Faster R-CNN method achieves 92.81% accuracy on the garbage target detection dataset.
Recent advances in three-generation sequencing technology allow for the rapid generation of large throughput of long reads, and mapping these long reads to a reference sequence is one of the first and most time-consuming steps in the downstream application of genomics. Minimap2, the state-of-the-art long-read sequencing aligner available today, has the advantage of being fast and accurate. However, as NUMA multi-core CPU gradually becomes the processors of mainstream computers, minimap2 is not specifically optimised and adapted for the NUMA multi-core architecture. Frequent remote memory accesses, resource contention and idle hardware resources result in a performance far below the theoretical peak performance of NUMA multi-core CPU. Based on the above problems, we propose three optimisation strategies, namely copying index at each NUMA node and binding threads to the cores of NUMA node, designing new IO and computation overlap mechanism, and adaptively adjusting batch_size based on IO and computation time, to achieve full utilisation of resources. We obtain three sets of human genome sequencing data from the ENA database and performed performance tests on the FT 2000+ MCD-FP92 NUMA multi-core CPU system. The three-point strategies proposed in this paper are effective in improving the performance of minimap2, with a maximum speedup of 13 percentage points.
With the rapid development of deep learning, a variety of neural network models emerge in endlessly, which leads to a huge demand for computing resources. For the intensive numerical computation of neural networks, various computing devices represented by GPUs are favored by researchers. Heterogeneous computing is a kind of technology that can integrate a variety of computing devices with different architectures, and it will be further developed. Therefore, this paper reviews research on some key technologies of heterogeneous computing, including the architecture of heterogeneous computing, the programming language of heterogeneous computing, and the scheduling algorithm for heterogeneous systems. Then, we focus on the research of heterogeneous computing in deep learning, including the parallel technology of neural networks and optimization technology based on heterogeneous systems. Finally, the present research situation is discussed and analyzed, and the future research direction is prospected, aiming to provide some basis for related research.
With the development of cloud computing and cloud storage technology, the data scale has grown rapidly. In order to store and process large-scale data, there are thousands of nodes and devices in the cloud storage center, resulting in a surge in the frequency of failures. In various types of failure events, storage device failure is the most important one. However, most cloud storage systems lack disk failure prediction mechanisms and could only replace disks after disk failures. It is particularly important to predict the potential risks in the system operation environment. In this paper, we propose a disk failure prediction approach that supports multi granularity data fusion, which solves problems of unbalanced samples, single data source, cross scenario model migration and insufficient generalization ability of prediction models in disk failure prediction. Through our proposed approach, the cloud storage system could accurately predict disk failures and actively push prediction results to users, so as to improve the pertinence and planning of the operation and maintenance work. The approach presented in this paper has been validated to be valid through a series of qualitative and quantitative experiments.
This paper collects and sorts out the relevant data of Guangxi's cultural industry and tourism industry from 2010 to 2019, and analyzes and clarifies the degree of integration and existing problems of cultural tourism in Guangxi based on the coupling coordination model, draws a Line charts and histograms reflecting the trend change of its coupling and coordination degree, and proposes corresponding solutions. The results show that from 2010 to 2019, the coupling coordination degree D value of the two is between 0.441 and 0.885, and most of the coupling degree C values are above 0.8. In 2010, the coupling coordination was the lowest, with a value of 0.441, and in 2019, the coupling coordination reached a maximum of 0.885. With the "industrial integration" gradually entering the public vision, tourism, as a pillar industry of the national economy, has been deeply developed at home and abroad in recent years. According to the problem of fluctuation of coupling and coordination degree between the two systems, the paper proposes to focus on the establishment of a long-term mechanism for the development of tourism industry and the development of cultural industry, so as to promote the innovative development of Guangxi's tourism industry, improve the competitiveness of the tourism industry, optimize the industrial structure, and transform the growth momentum.
In order to better preserve the salient targets and detail information of the infrared and visible images, a novel image fusion method based on log-domain decomposition and information interaction is proposed in this paper. Specifically, the infrared and visible images are first transformed into the logarithmic domain for a two-scale decomposition, which helps to extract more high-contrast information compared to decomposing them directly in image space. A visual saliency strategy is then used to fuse the base layer images. As to detail layers, a combined local visual saliency and detail preservation strategy is proposed to determine the final fusion weights. In addition, it is worth noting that the visible image information is introduced into the infrared detail layer before fusion, which achieves the information complementation and interaction of two source images. The experiment results demonstrate that the proposed method outperforms other fusion methods in both qualitative and quantitative assessments.
Facing the new requirements of building a digital engineering ecology, this paper proposes an integration method of PDM system and archive management system based on intermediate table, based on the analysis of the current situation of product electronic data archive management and combined with the requirements of military secrecy, for the process of archiving electronic data in military research institutes. The method realizes the automatic archiving of development models、 processes、documents and other data to the file management system, reduces the manual intervention of archivists, and improves the efficiency of file management. The uniqueness of product data is ensured.
For the military manufacturing field, through a detailed analysis of ERP business processes and finance, a business-financial integration scheme is proposed, which opens the data link between ERP and business finance platform by establishing synchronous update of business and financial information data, and realizes the financial accounting of manufacturing business from sales to delivery, procurement to payment, demand to manufacturing, etc., which includes scientific research and trial production and mass production. The integrated integration improves the efficiency of accounting and the accuracy of data.
