Search

Given the escalating potential risk associated with X-ray radiation exposure to patients, scholars have been dedicated to investigating advanced algorithms for low-dose CT (LDCT) imaging. The utilization of Generative Adversarial Networks (GANs) has gained significant traction in the field of LDCT denoising research over the past few years, owing to its notable performance advantages. However, variations in the distribution intensity of noise and artifacts in the LDCT image can compromise the robustness and denoising performance of GAN. In this paper, we propose a scale-sensitive generative adversarial network to enhance the robustness of GAN used in LDCT denoising. Our proposed network incorporates an error feedback pyramid generator to improve the denoising network’s sensitivity towards noise and artifacts present in LDCT images by performing feature extraction and noise reduction at different scales. Additionally, we introduce a multi-scale residual discriminator with a wider receptive field and stronger feature extraction ability to enhance the discrimination between true and false samples. Furthermore, specific constraints are incorporated into the loss functions based on the network structure, thereby enhancing the effect of adversarial training between the error feedback pyramid generator and multi-scale residual discriminator while improving the performance of the denoising network. Experimental results demonstrate that our proposed network outperforms state-of-the-art methods in terms of artifact suppression, structural preservation, and model generalization capability for LDCT denoising.

Density Peak Clustering (DPC) is known for its rapid identification of cluster centers and successful clustering tasks. However, traditional DPC encounters several issues, which include simplifications in local density and distance metrics, a non-robust single-allocation strategy and limited fault tolerance. To address these challenges, this study introduces an innovative density peak clustering algorithm, named Graph Distance and Adaptive K-Nearest Neighbors Selection-Based Density Peak Clustering (GAK-DPC). Our goal with the approach is to enhance the algorithm’s adaptability to non-linear and complex data structures. We achieve this by replacing the traditional Euclidean distance with graph distance. Additionally, we redefine the method for computing local density based on information from K-nearest neighbor data points. By introducing the concept of natural neighbors, the neighborhood radius r is obtained when all instances in the dataset have at least one natural neighbor. Then for the current data point, the number of data points falling within a circle centered on it with radius r is counted as the K-value of that data point. Thus, we achieve the adaptive selection of the K-value. This adaptive K-value strategy takes into account the dataset’s characteristics and inter-point neighbor relationships, which enhances the algorithm’s adaptability and robustness. Finally, we optimize the secondary allocation strategy for sample points to improve the algorithm’s fault tolerance. By conducting comparisons with traditional clustering algorithms on UCI datasets and synthetic datasets, we demonstrate the effectiveness of GAK-DPC.

The rapid and reliable detection of oil contaminants is of crucial importance for environmental protection. This paper focused on the classification and identification of six types of oil using unconventional steady-state fluorescence spectroscopy coupled with XY-fused networks (XY-Fs). The excitation emission matrix fluorescence (EEMF) spectroscopy, synchronous fluorescence spectroscopy (SFS) and total synchronous fluorescence spectroscopy (TSFS) of the oil samples were measured by a FS920 steady-state fluorescence spectrometer. Further, the obtained results were compared with those obtained by using partial least square discriminant analysis (PLS-DA), LDA based on score matrix of multivariate curve resolution-alternating least squares (MCR-ALS-LDA) and parallel factor (PARAFAC-LDA) analysis. The experimental results showed that the best accuracy (100%) for the test samples was obtained by TSFS combined with XY-Fs. These results provide important references for discrimination of oil contaminants.

Big data-driven technologies, especially machine learning and deep learning technologies, have been extensively employed in mineral prospectivity prediction. Several approaches have been proposed to learn the deep characteristics of geoscience data, enhance the accuracy of prediction and reduce uncertainty. Nevertheless, the approaches always contain the following two limitations. Firstly, the formation of mineral resources often involves the coupling of multiple factors on a certain spatio-temporal scale, resulting in rare labelled deposits and insufficient number of training samples. Secondly, training Deep Neural Network (DNN) is very challenging. Many approaches are subject to weak interpretability and lack of organic combination with geoscience knowledge. To address these two problems, we propose Geo-Rnet and GCAE (Geological Convolutional Autoencoder). Geo-Rnet is a multi-class mineral prospectivity prediction approach based on improved DNN. GCAE is able to effectively augment multi-disciplinary geoscience data by constructing upon an optimized Convolutional Autoencoder. The experimental results show that most of prospective areas predicted by Geo-Rnet overlap with the labelled mineralization locations, with an average accuracy of 91.1%. In addition, 89.98% of the ore deposits are located in the predicted areas. The results indicate the effectiveness of Geo-Rnet and GCAE for multi-class prediction of mineral resources. Finally, we classify the target area into several mineral prospectiviy areas according to their different mineral types. The research provides an innovative approach for mineral prospectivity prediction in the target area.

The LiDAR and photogrammetric point clouds fusion procedure for building extraction according to U-Net deep learning model segmentation is provided and tested. Firstly, an initial geo-localization process is performed for photogrammetric point clouds generated using structure-from-motion and dense-matching methods. Then, point cloud segmentation is carried out based on U-Net deep learning model. The precision of the U-Net model for buildings extraction reachs 87%, with F-score of 0.89 and IoU of 0.80. It is shown that the U-Net method is effective for high-resolution image extraction. The detailed information can accurately be identified and extracted, such as vegetation located between buildings and roads. After segmentation, each chunk of the LiDAR and photogrammetric point clouds are finely registered and merged based on the iterative closest point algorithm. Finally, the fused point clouds are obtained. It shows that the structure and shape of the buildings could be delineated from the fused point clouds when both enough ground points and a higher point density are available. Furthermore, color information improves both visualization effect and properties identification. The experiments are conducted to extract individual buildings from three types of point clouds in three plots. A DoN (Difference of Normals) approach is used to isolate 3D buildings from other objects in densely built-up areas. It shows that most building extraction results have a Precision > 0.9 and favorable Recall and F-score values. Although the LiDAR extraction results have some advantages over the photogrammetric and fused ones in terms of Precision, the Recall and F-score results appear best for the fused point clouds. It shows that the fused data contains a high point density and RGB color information and could improve the building extraction.

Convolutional neural network (CNN)-based deep learning techniques have enjoyed many successful applications in the field of medical imaging. However, the complicated between-manifold projection from the projection domain to the spatial domain hinders the direct application of CNN techniques in computed tomography (CT) reconstruction. In this work, we proposed a novel CT reconstruction framework based on a CNN, i.e., an intelligent back-projection network (iBP-Net). The proposed iBP-Net method fused a pre-CNN, a back-projection layer, and a post-CNN into an end-to-end network. The pre-CNN adopted CNN techniques to model a filtering operation in the projection domain. In the back-projection layer, a back-projection operation was employed to perform between-manifold projection. Based on CNN techniques, the post-CNN worked together with the pre-CNN to recover reconstructed images from the outputs of the back-projection layer in the spatial domain while maintaining high visual sensitivity. The experimental results demonstrate the feasibility of the proposed iBP-Net framework for CT reconstruction.

Current attention or transform modules in Convolutional Neural Networks (CNNs) are designed pursuing lightweight and in-place. Generally, we need to decrease the channel dimension of input feature maps for reducing computation cost firstly. And then we do some transformation for extracting weight maps or converting to other feature space etc. Finally, we increase the channel dimension back for outputting feature maps with the same size as input. When we change the channel dimension, commonly we choose 1 × 1 convolutional layers or fully connected layers. They are simple and effective, but need learning parameters and consuming more memory with other computation resources. We propose a novel parameter free method named Channel Transformer Network (CTN) to decrease or increase channels for these modules whilst keeping most information with lower computation complexity. We also introduce a Video Co-segment Attentive Network (VCAN) for person re-identification (ReID) to improve pedestrian's noticeable representation across multiple video frames. We embed CTN in Non-local, CBAM, COSAM and VCAN blocks to replace 1 × 1 convolutional or fully connected layers. Experiments of VCAN and CTN embedding models on Mars dataset for person ReID show significant performance in computation efficiency and accuracy, especially VCAN reaches 90.05% in Rank-1. We believe CTN can also be used in other vision tasks like image classification and object detection etc.

In intensity-modulated radiation therapy (IMRT), a network flow is adopted to solve the pricing problem of the generic column generation approach in order to obtain a deliverable aperture. However, excessive computation results from the direct use of a network flow. In addition, a decline in plan quality may result from the direct determination of the leaf position using the gradient information. To overcome these problems, a column generation approach based on region growth is proposed. The proposed method is designed to reduce the computational cost of solving the pricing problem and improve the IMRT plan quality. First, the gradients of the beamlets are obtained by an objective function constructed under the constraint conditions of the organs. Second, the gradients are transformed nonlinearly. Third, the positions of the continuous negative gradient regions in each row of the aperture are determined and stored. Fourth, these gradients are taken as a whole and added to the aperture network flow, which is solved as a shortest-path problem. Finally, the deliverable aperture is obtained and added to the treatment plan. To verify the effectiveness of the proposed method, experiments involving five five-field prostate cancer cases and five nine-field head and neck cancer cases were conducted. Compared with the generic column generation method, the dose distribution of the target is ensured by the proposed method, which also effectively protects organs at risk and reduces the running time. Specifically, in ten groups of comparative experiments, the normal tissue complication probability of the proposed method is reduced by up to 3.37%, and the maximum acceleration rate is 20.44%. According to the experimental results, the proposed method is more consistent with clinical requirements compared with the generic column generation method.

During the process of smart city construction, city managers always spend a lot of energy and money for cleaning street garbage due to the random appearances of street garbage. Consequently, visual street cleanliness assessment is particularly important. However, the existing assessment approaches have some clear disadvantages, such as the collection of street garbage information is not automated and street cleanliness information is not real-time. To address these disadvantages, this paper proposes a novel urban street cleanliness assessment approach using mobile edge computing and deep learning. First, the high-resolution cameras installed on vehicles collect the street images. Mobile edge servers are used to store and extract street image information temporarily. Second, these processed street data is transmitted to the cloud data center for analysis through city networks. At the same time, Faster Region-Convolutional Neural Network (Faster R-CNN) is used to identify the street garbage categories and count the number of garbage. Finally, the results are incorporated into the street cleanliness calculation framework to ultimately visualize the street cleanliness levels, which provides convenience for city managers to arrange clean-up personnel effectively. The overall approach is illustrated and visualized using the street images of Jiangning District in Nanjing, China. The practical application shows the feasibility and usability of the approach.

Photoelectric non-contact measurement methods have been widely implemented in space target location, micro-displacement measurement and space optical communication. In term of the reliability and precision of measurement systems, it is a great significance to obtain the laser center location accurately. Since the two-dimensional position sensitive detector (2-D PSD) can continuously detects the gravity center position of the laser spot on its photosensitive surface, this paper provides a flexible method with high precision for the nonlinear correction of PSD as well as a new idea for PSD-based moving laser center positioning. Above all, a laser device system that rotating around a fixed-axis is designed, and the rotation angle can be controlled with the rotation accuracy of 0.5°. The trace of the gravity center of the rotating spot generated by the device on the PSD surface is detected, and then the laser center location is obtained after further processing of data. In order to improve the detection accuracy of PSD, this research mainly focuses on the nonlinear correction method of the detector and the fitting method of the gravity center trace of the laser spot. Moreover, the effects of different sizes of light spots projected onto the 2-D PSD on the collected photocurrent and estimated coordinates were analyzed by experiments. From the experimental results, the non-linear error of PSD is corrected by the error curved surface interpolation method, the overall error is reduced by more than 32%, and the linearity of the positioning of the laser moving along the diagonal of PSD is 0.2%.

Direct aperture optimization (DAO) is an effective method to generate high-quality intensity-modulated radiation therapy treatment plans. In generic DAO, the direction of negative gradient descent is generally used to determine the aperture shape. However, this strategy can reduce the convergence rate, especially near the optimal value. We propose aperture shape generation based on the direction of gradient descent with momentum, where column generation is implemented as carrier. During aperture shape generation of column generation, the current aperture gradient map is first calculated. Then, the gradient with momentum is calculated based on the existing gradient information. Finally, the direction of gradient descent with momentum is constructed for obtaining the deliverable aperture shape by solving the pricing problem. To verify the effectiveness of the proposed method, we conducted comparative experiments on two head and neck and two prostate tumor cases. Compared with generic column generation, the proposed method can effectively protect the organs at risk while ensuring the required dose distribution to the target. Using the proposed method, the number of apertures and optimization time can be reduced by up to 30.95 and 32.96%, respectively, compared to the conventional approach. The experimental results suggest that the proposed method can accelerate the search speed and improve the quality of treatment plans.

Composite services evolve for various reasons. Test case selection in the regression testing is an effective technique to ensure the correctness of modified versions meanwhile to reduce the cost of testing. However, few work has studied the test case selection problem based on the data flow testing criteria. In addition, there are three observable kinds of changes during the evolution, including Process change, Binding change and Interface change, which all bring impact to the data flow. To address these issues, a test case selection approach is proposed for regression testing of BPEL (Business Process Execution Language) composite service where all-uses criterion is satisfied and all the three change types are involved. BPEL composite service is modeled with a two-level model in which XCFG (eXtended Control Flow Graph) describes the behavior of BPEL process in the first level and WSDM (Web Service Description Model) depicts the interface information of composite service and partner services in the second level. Change impact analysis is performed to identify the affected definition-use pairs by comparing and analyzing two-level models of the baseline and evolved versions. And testing paths are generated to cover the affected definition-use pairs and select test cases based on the path condition analysis. Empirical result shows that the proposed approach is effective.

Cloud computing is the next generation computing model, which has a significant position in the field of scientific and business computing. By predicting cloud service's QoS in next period, it is helpful for end users to choose the most suitable cloud service that meets their needs. The underlying hardware/software resources of cloud architecture may have a certain influence on cloud service QoS. However, existing cloud service QoS prediction approaches do not take this influence into account. As these effects are real during the process of cloud service QoS prediction, ignoring the impact of these effects may create a big gap between the prediction results and the actual results. Therefore, in this paper interactive information is first used to describe the correlation between the hardware/software resources and the QoS attributes of the cloud service. Then, a Bayesian network model is established to predict cloud QoS. Bayesian network prediction reasoning algorithm is used to predict and reason about the future QoS values. A set of dedicated experiments is conducted to validate that our approach can accurately predict QoS of cloud service and the accuracy rate is better than state-of-the-art approaches.

A load-independent wireless power transfer system with constant current and constant voltage output for electric vehicles charging is designed and optimized in this paper. Wireless charging system based on LCL-S or LCL-LCL compensation topology is systematically analyzed. And dynamic LCL-S/LCL switching topology is designed and simplified to achieve constant current in the transmitting coil and load-independent constant current and constant voltage output, which can be controlled easily. Moreover, the coupling structures composed of different coil shapes and shielding structures are comparatively studied to improve the coupling stability under misalignment. Figure-of-merit and coupling change rate Δkt defined in this paper are the key parameters in the process of coupling structure design and optimization. The combination of rounded rectangular spiral coil and splicing magnet core units is optimized as the coupling structure of the wireless charging system. Finally, the resonant wireless charging system prototype is being built and tested. The experimental results show that the load-independent and other characteristics of the implemented system are well correlated with the theoretical analysis and design.

The aperture shape optimization (ASO) is a critical step in the direct aperture optimization (DAO) method. During ASO, the gradient of objective function is calculated with respect to the beamlet weight. These gradient components are directly utilized to generate the new aperture shape. In this way, the beamlet of the large positive gradient value may be grouped into the generated aperture shape. The treatment quality may be deteriorated by adding this aperture into the treatment plan. In order to overcome this drawback, a novel method based on the fuzzy enhancement was proposed to generate the aperture shape. We apply the fuzzy enhancement method to enhance the gradient map that is composed of the gradients of objective function in a beam. The enhanced gradient map was then employed to form a network flow, which was solved to generate the new aperture shape. The optimal aperture shape was generated by removing the beamlet of the large positive gradient value from the new generated aperture shape. To verify the effectiveness, the proposed method was compared with the conventional column generation (CG) method on a prostate cancer case and on a head-and-neck cancer case. Experimental results demonstrate that the new algorithm has a better performance than the CG algorithm. The proposed method can further reduce the dose delivered to the critical structures, when the similar dose coverage is delivered on the targets.

In this paper, we addressed peers' perceived creative idea generation and supervisors' endorsement of these creative ideas in the software industry. Using three sources of data collection techniques for 317 subordinates, subordinate-colleague dyads, and 50 supervisors, we collected data from software engineers, software developers, and software quality specialists working in different hierarchical levels of a software firm and analyzed our hypothesized model with Mplus using random coefficient modeling. We found that the creative idea generation of subordinates is positively related to supervisors' endorsement of creative ideas. Justice in organizations and relationships with a supervisor both independently strengthened the relationship between creative idea generation as well as creative idea endorsement. We found that supervisors are likely to endorse the ideas of their subordinates when there is an overall application of justice in the organization and when they have a good, trustworthy relationship with their subordinates. Research implications, limitations, and future research directions are also discussed.

Massive flows that represent the individual level of movements and communications can be easily obtained in the age of big data. Generalizing spatial and temporal flow patterns from such data is essential to demonstrate spatial connections and mobility trends. Clustering approaches provide effective methods to handle data sets that contain massive individual-level flows. However, existing flow clustering studies obscure the geometric properties of flow data, such as direction and length, which significantly indicate movement trends. In addition, temporal information is often ignored because previous approaches have mainly focused on the perspective of spatial clusters of flow data, resulting in a loss of temporal patterns. In this paper, we introduce new spatial and temporal similarity measurements between flows and propose a new clustering approach of flow data based on a stepwise strategy. This method can identify clusters from distinct flow distributions and discover significant spatio-temporal trends from large flow data. Simulated experiments with synthetic flows and a case study using Beijing taxi trip data are conducted to validate the usefulness of the proposed method.

Automatic and accurate cervical nucleus segmentation is important because nuclei carry substantial diagnostic information for automatic computer-assisted cervical cancer screening and diagnosis systems. In this paper, we propose a cervical nucleus segmentation method in which pixel-level prior information is utilized to provide the supervisory information for the training of a mask regional convolutional neural network (Mask-RCNN), which is then employed to extract the multi-scale features of the nuclei, and the coarse segmentation and bounding box of the nuclei are obtained by forward propagation of the Mask-RCNN. To refine the segmentation, a local fully connected conditional random field (LFCCRF) that contains unary and pairwise energy terms is employed. The nuclear region of interest is determined by extending the bounding box, the coarse segmentation in the nuclear region is used to construct the unary energy, and the pairwise energy is contributed by the position and intensity information of all of the pixels in the nuclear region. By minimizing the energy of the LFCCRF, the final segmentation is realized. We evaluated our method by using cervical nuclei from the Herlev Pap smear data set in this paper, and the precision, recall, and Zijdenbos similarity index were all found to be greater than 0.95 with low standard deviations, demonstrating that our method enables more accurate and stable cervical nucleus segmentation than the current state-of-the-art methods.

Dose-volume (DV)-based objectives are widely used in most intensity-modulated radiation therapy treatment planning, because numerous DV endpoints have been utilized. In clinical practice, DV-based optimization (DVO) with uniform or random initial intensity distributions is utilized, but without considering the non-convexity of the DV objectives. To improve the quality of DVO radiotherapy plan and reduce the local minimum error generated by non-convexity of the DV objective, we proposed an efficient method (an organ-model-based optimization guiding DVO) to determine the initial intensity distributions for DVO. The new approach includes two steps. First, fluence map optimization, based on the organ model that adopts our proposed increasing objective function, to assure organ evaluation criteria Pareto surface, was performed. Second, DVO procedure was performed by using the initial intensity distributions determined in the first step. We demonstrated this technique in two kinds of clinical cases. DV histogram metrics were adopted as the criterion to evaluate the treatment plans. Compared with the conventional DVO plan with uniform initial intensity distributions, the improved DVO plan provided better protection of organ at risk (OAR); the planning target volume coverage was similar. Moreover, the improved DVO plan was better than the plan generated in the first step. The proposed method, with advantages in determining the initial intensity distributions, was highly efficient to improve DVO plans.

Metamorphic testing has been successfully used in many different fields to solve the test oracle problem. However, how to find a set of appropriate metamorphic relations for metamorphic testing remains a complicated and tedious task. Recently some machine learning approaches have been proposed to predict metamorphic relations. These approaches predicting single label metamorphic relation can alleviate this problem to some extent. However, many applications involve multi-group metamorphic relations, and these approaches are clearly inefficient. To address this problem, in this paper we propose a Multi-Label Metamorphic Relations prediction approach based on an improved radial basis function (RBF) neural network named RBF-MLMR. First, RBF-MLMR uses state-of-the-art soot analysis tool to generate control flow graph and corresponds labels from the source codes of programs. Second, the extracted nodes and the path properties constitute multi-label data sets for the control flow graph. Finally, a multi-label RBF neural network prediction model is established to predict whether the program satisfies multiple metamorphic relations. In order to improve the prediction results, affinity propagation and k-means clustering algorithms are used to optimize the RBF neural network structure of RBF-MLMR. A set of dedicated experiments based on public programs is conducted to validate RBF-MLMR. The experimental results show that RBF-MLMR can achieve accuracy of around 80% for predicting two and three metamorphic relations.