通讯机构:
[Jiangming Kan] S;School of Technology, Beijing Forestry University, No. 35 Qinghua East Road, Haidian District, Beijing 100083, China
关键词:
monocular depth estimation;unsupervised learning methods;structure from motion;confidence mask
摘要:
Abstract: Monocular depth estimation is a fundamental yet challenging task in computer vision as depth information will be lost when 3D scenes are mapped to 2D images. Although deep learning-based methods have led to considerable improvements for this task in a single image, most existing approaches still fail to overcome this limitation. Supervised learning methods model depth estimation as a regression problem and, as a result, require large amounts of ground truth depth data for training in actual scenarios. Unsupervised learning methods treat depth estimation as the synthesis of a new disparity map, which means that rectified stereo image pairs need to be used as the training dataset. Aiming to solve such problem, we present an encoder-decoder based framework, which infers depth maps from monocular video snippets in an unsupervised manner. First, we design an unsupervised learning scheme for the monocular depth estimation task based on the basic principles of structure from motion (SfM) and it only uses adjacent video clips rather than paired training data as supervision. Second, our method predicts two confidence masks to improve the robustness of the depth estimation model to avoid the occlusion problem. Finally, we leverage the largest scale and minimum depth loss instead of the multiscale and average loss to improve the accuracy of depth estimation. The experimental results on the benchmark KITTI dataset for depth estimation show that our method outperforms competing unsupervised methods. Keywords: monocular depth estimation; unsupervised learning methods; structure from motion; confidence mask
期刊:
Journal of Mathematical Physics,2021年62(10):101507 ISSN:0022-2488
通讯作者:
Zhang, Tingting
作者机构:
[Zhang, Tingting] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.
通讯机构:
[Zhang, Tingting] W;Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.
摘要:
In this paper, we get the invariant domain of 1D non-isentropic gas dynamics equations. First of all, we construct the elementary waves with the characteristic analysis method. According to the characteristic of elementary waves, we divide the u-p plane into five areas. By analyzing the structure of Riemann solutions in each area, we find a new convex bounded domain where if the Riemann data belong to the domain, then the Riemann solutions also belong to the domain. Moreover, it is used to prove the uniform boundedness of approximate solutions built by the difference scheme, so it is the basis for the Riemann problem to be applied to the Cauchy problem.& nbsp;Published under an exclusive license by AIP Publishing.</p>
摘要:
A systematic approach has been developed to estimate the relationship between the permeability and connectivity of two-dimensional fracture networks, in which the network connectivity is evaluated with the concept of geological entropy as informative index of spatial disorder. The geological entropy is quantified by the entropic scale, a metric developed by Bianchi and Pedretti (2018), which is here applied to integrate multiple properties of two-dimensional fracture networks, including aperture, spacing, length, orientation. Through the comparisons among several existing connectivity indicators including the entropic scale in an illustrative example, only the entropic scale is positively correlated with the permeability concerning aperture and can be successfully used to quantify the network connectivity. In order to understand connectivity characteristics dependence of the permeability, a computational method combining fracture network generation and steady-state flow simulation is developed. Based on the results of detailed numerical simulations considering hydraulic behavior and connectivity characteristic in various fracture networks, the entropic scale and permeability are simultaneously inversely proportional to length and proportional to spacing with the same global entropy, but exhibit weak dependence on the orientation variation. A simple closed-form empirical expression in terms of quadratic polynomial model between the permeability and entropic scale is proposed. The results indicated that geological entropy is valid and appropriate to quantify the connectivity and predict the permeability of two-dimensional fracture networks.
期刊:
Mobile Information Systems,2021年2021 ISSN:1574-017X
作者机构:
[Hu, Ying; Chang, Jing] Hubei Univ Educ, Sch Math & Econ, Wuhan 430000, Hubei, Peoples R China.;[Huang, Jian] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430000, Hubei, Peoples R China.;[Hu, Ying] Hubei Univ Educ, Big Data Modeling & Intelligent Comp Res Inst, Wuhan 430000, Hubei, Peoples R China.
摘要:
In the context of the popularization and diversified application of information technology in higher education, efficient information dissemination has a significant impact on the learning effect of the learning community. Improving the efficiency of information dissemination and driving the force of learning to enhance the learning effect are the hot issues in the field of higher education data analysis. This paper proposes a new method of feature fusion using information entropy and ReliefF algorithm, applies the improved PageRank algorithm and K-means algorithm to optimize the information transfer mode, and finally develops a new and efficient network information model. The comparative test results show that the new model can complete the dissemination of the same amount of information with a smaller delivery ratio. The research results can play an advantageous role in information interaction feedback, curriculum quality analysis, and teaching information transmission.
关键词:
Hyperspectral unmixing;weight-sharing architecture;stick-breaking process;Dirichlet distribution
摘要:
Recently, the learning-based method has received much attention in the unsupervised hyperspectral unmixing, yet their ability to extract physically meaningful endmembers remains limited and the performance has not been satisfactory. In this paper, we propose a novel two-stream Dirichlet-net, termed as uTDN, to address the above problems. The weight-sharing architecture makes it possible to transfer the intrinsic properties of the endmembers during the process of unmixing, which can help to correct the network converging towards a more accurate and interpretable unmixing solution. Besides, the stick-breaking process is adopted to encourage the latent representation to follow a Dirichlet distribution, where the physical property of the estimated abundance can be naturally incorporated. Extensive experiments on both synthetic and real hyperspectral data demonstrate that the proposed uTDN can outperform the other state-of-the-art approaches.
期刊:
International Journal of Robust and Nonlinear Control,2021年31(3):806-816 ISSN:1049-8923
通讯作者:
Zhao, Jiemei
作者机构:
[Zhao, Jiemei] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.
通讯机构:
[Zhao, Jiemei] W;Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.
关键词:
inverse;reachable set estimation;time‐varying delay singular systems;uncertain systems;Wirtinger‐based integral inequality
摘要:
This paper is concerned with the reachable set estimation (RSE) problem for singular systems with both time-varying delays and bounded peak disturbances. The objective is to search a bounded set that contains all the system states under zero initial conditions. By utilizing the theory of {1}-inverse and Wirtinger-based integral inequality, an improved criterion is established in terms of the linear matrix inequalities (LMIs) to guarantee that the reachable set of time-varying delay singular system is regular, impulse-free and bounded by the intersection of ellipsoids. Here, a relaxed Lyapunov-Krasovskii functional is employed to solve the addressed RSE problem which does not require all the involved symmetric matrices to be positive definite. Finally, numerical examples are provided to demonstrate the effectiveness of the proposed methods.
期刊:
Journal of Soils and Sediments,2021年21(1):487-498 ISSN:1439-0108
通讯作者:
Zhang, Cong
作者机构:
[Zhang, Cong; Cao, Wenqi] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.
通讯机构:
[Zhang, Cong] W;Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430023, Peoples R China.
关键词:
Deep composite model;Prediction of soil heavy metal content;Radial basis function neural network;Particle swarm optimization;Root mean square back-propagation
摘要:
The content of heavy metals in the soil is directly related to the control of soil pollution, but due to the limitations of manpower and material resources, it is difficult to detect them in detail; researchers usually need to predict the content of soil heavy metals in unknown areas based on existing data. Therefore, how to choose an effective method to complete this process has become a challenging problem. In this paper, a deep composite model (DCM) is proposed. The model is based on radial basis function neural network (RBFNN), then, uses self-adaptive learning based particle swarm optimization algorithm (SLPSO) to generate the weight and bias of the output layer of RBFNN and employs adaptive adjustment based root mean square back-propagation (ARMSProp) to optimize all variables of RBFNN, so as to improve the prediction accuracy of the model for soil heavy metal content. When using this model to predict soil heavy metal content, the Pearson coefficient is used as a comparison index to compare the correlation between different heavy metals and heavy metals to be predicted, and finally the content of heavy metals with a Pearson coefficient greater than 0.5 is selected as the input of the model variable. First in the validation of the proposed SLPSO algorithm, the effectiveness of SLPSO and the feasibility of being applied to the DCM model have been proved. Then, the DCM was applied to the prediction of soil heavy metal content in six new urban areas of Wuhan in China, the experimental results show that the predicted value of soil heavy metal content of DCM is closer to the actual value than other comparison models, and the four error indicator values of DCM are also significantly lower than other comparison models, especially when compared with RBFNN, the MAPE and SMAPE of DCM have dropped by 8.6% and 3.9%, respectively. We can conclude that the deep composite model proposed in this paper obtains a good prediction accuracy when predicting soil heavy metal content; it has certain feasibility and can be used as an effective method for soil heavy metal content prediction.
摘要:
At present, General Regression Neural Network (GRNN) have been more and more used for data prediction in industry, however, because its smoothing factor is difficult to determine, it is easy to obtain poor prediction accuracy when using it to predict complex problems in reality. To tackle these problems, an effective Parallel Integrated Neural Network System (PINN) is proposed in this paper. The model is a combination of GRNN and Adaptive Dynamic Grey Wolf Optimizer (ADGWO), in this model, the smoothing factor and calculation result of GRNN are taken as the individual position information and individual fitness of ADGWO, respectively, and the training of the model is completed through the optimization of ADGWO. Different from Grey Wolf Optimizer (GWO), ADGWO introduces the nonlinear cosine decreasing convergence factor, the weighted position update method and the central disturbance criterion, aiming to balance the exploitation and exploration. Applying PINN to the soil heavy metal datasets from Yinchuan of Ningxia and Wuhan, China for data prediction, the experimental results show that PINN has higher average prediction accuracy than several comparative models, especially an increase of 8.05% compared with Wavelet Neural Network, which proves that PINN can be effectively applied to industrial data prediction. (C) 2021 Elsevier B.V. All rights reserved.
摘要:
Anomaly detection is now a significantly important part of hyperspectral image analysis to detect targets in an unsupervised manner. Traditional hyperspectral anomaly detectors fail to consider spatial information, which is vital in hyperspectral anomaly detection. Moreover, they usually take the raw data without feature extraction as input, limiting the detection performance. We propose a new anomaly detector based on the fractional Fourier transform (FrFT) and a modified patch-image model called the hyperspectral patch-image (HPI) model to tackle these two problems. By combining them, the proposed anomaly detector is named fractional hyperspectral patch-image (FrHPI) detector. Under the assumption that the target patch-image is a sparse matrix while the background patch-image is a low-rank matrix, we first formulate a matrix by sliding a rectangle window on the first three principal components (PCs) of HSI. The matrix can be decomposed into three parts representing the background, targets, and noise with the well-known low-rank and sparse matrix decomposition (LRaSMD). Then, distinctive features are extracted via FrFT, a transformation which is desirable for noise removal. Background atoms are selected to construct the covariance matrix. Finally, anomalies are picked up with Mahalanobis distance. Extensive experiments are conducted to verify the proposed FrHPI detector's superiority in hyperspectral anomaly detection compared with other state-of-the-art detectors.