Far-field super-resolution ghost imaging with a deep neural network constraint

Ghost imaging (GI) facilitates image acquisition under low-light conditions by single-pixel measurements and thus has great potential in applications in various fields ranging from biomedical imaging to remote sensing. However, GI usually requires a large amount of single-pixel samplings in order to reconstruct a high-resolution image, imposing a practical limit for its applications. Here we propose a far-field super-resolution GI technique that incorporates the physical model for GI image formation into a deep neural network. The resulting hybrid neural network does not need to pre-train on any dataset, and allows the reconstruction of a far-field image with the resolution beyond the diffraction limit. Furthermore, the physical model imposes a constraint to the network output, making it effectively interpretable. We experimentally demonstrate the proposed GI technique by imaging a flying drone, and show that it outperforms some other widespread GI techniques in terms of both spatial resolution and sampling ratio. We believe that this study provides a new framework for GI, and paves a way for its practical applications.

Genetic Diversity and Primary Core Collection Construction of Turnip (Brassica rapa L. ssp. rapifera Matzg) Landraces in Tibet Revealed via Morphological and SSR Markers

Turnip, one of the oldest groups of cultivated Brassica rapa species, is a traditional crop as well as a form of animal fodder, a vegetable, and a herbal medicine that is widely cultivated in farming and farming-pastoral regions in Tibet. Different regions of the Qinghai–Tibet Plateau (QTP) are home to a rich diversity of turnip owing to their high altitudes and variable climate types. However, information on the morphology and genetic diversity of Tibetan turnip remains limited. Therefore, the genetic diversity of 171 turnip varieties from China and elsewhere (Japan, Korea, and Europe) was analyzed using 58 morphological characteristics and 31 simple sequence repeat (SSR) markers in this study. The varieties showed that the genetic distance ranged from 0.12 to 1.00, and the genetic similarity coefficient ranged between 0.73 and 0.95. Cluster tree showed two distinct clusters. Both morphotype and geography contributed to the group classification. A combination of morphological traits and molecular markers could refine the precision of accurate identification compared to the separate morphological and molecular data analyses. The sampling ratio of 15% to utmost precisely represent the initial population was compared to ratios of 10% and 20%, and the sampling ratio of 15% is recommended for future works when a primary core collection of turnip resources is constructed. These results could furnish a foundation for germplasm conservation and effective turnip breeding in future studies.

The Decomposition of Between and Within Effects in Contextual Models

In contextual studies, group compositions are often extracted from individual data in the sample, in order to estimate the group compositional effects [e.g., school socioeconomic status (SES) effect] controlling for interindividual differences in multilevel models. As the same variable is used at both group level and individual level, an appropriate decomposition of between and within effects is a key to providing a clearer picture of these organizational and individual processes. The current study developed a new approach with within-group finite population correction (fpc). Its performances were compared with the manifest and latent aggregation approaches in the decomposition of between and within effects. Under a moderate within-group sampling ratio, the between effect estimates from the new approach had a lesser degree of bias and higher observed coverage rates compared with those from the manifest and latent aggregation approaches. A real data application was also used to illustrate the three analysis approaches.

The Sampling Ratio in Multilevel Structural Equation Models: Considerations to Inform Study Design

Multilevel structural equation modeling (MSEM) allows researchers to model latent factor structures at multiple levels simultaneously by decomposing within- and between-group variation. Yet the extent to which the sampling ratio (i.e., proportion of cases sampled from each group) influences the results of MSEM models remains unknown. This article explores how variation in the sampling ratio in MSEM affects the measurement of Level 2 (L2) latent constructs. Specifically, we investigated whether the sampling ratio is related to bias and variability in aggregated L2 construct measurement and estimation in the context of doubly latent MSEM models utilizing a two-step Monte Carlo simulation study. Findings suggest that while lower sampling ratios were related to increased bias, standard errors, and root mean square error, the overall size of these errors was negligible, making the doubly latent model an appealing choice for researchers. An applied example using empirical survey data is further provided to illustrate the application and interpretation of the model. We conclude by considering the implications of various sampling ratios on the design of MSEM studies, with a particular focus on educational research.

An Ensemble Framework of Multi-ratio Undersampling-based Imbalanced Classification

Class imbalance is commonly observed in real-world data, and it is problematic in that it degrades classification performance due to biased supervision. Undersampling is an effective resampling approach to the class imbalance. The conventional undersampling-based approaches involve a single fixed sampling ratio. However, different sampling ratios have different preferences toward classes. In this paper, an undersampling-based ensemble framework, MUEnsemble, is proposed. This framework involves weak classifiers of different sampling ratios, and it allows for a flexible design for weighting weak classifiers in different sampling ratios. To demonstrate the principle of the design, in this paper, a uniform weighting function and a Gaussian weighting function are presented. An extensive experimental evaluation shows that MUEnsemble outperforms undersampling-based and oversampling-based state-of-the-art methods in terms of recall, gmean, F-measure, and ROC-AUC metrics. Also, the evaluation showcases that the Gaussian weighting function is superior to the uniform weighting function. This indicates that the Gaussian weighting function can capture the different preferences of sampling ratios toward classes. An investigation into the effects of the parameters of the Gaussian weighting function shows that the parameters of this function can be chosen in terms of recall, which is preferred in many real-world applications.

CS-GAN for High-Quality Diffusion Tensor Imaging

Background: Compressed sensing magnetic resonance imaging (CS-MRI) is a promising technique for accelerating MRI speed. However, image quality in CS-MRI is still a pertinent problem. In particular, there is little work on reducing aliasing artefacts in compressed sensing diffusion tensor imaging (CS-DTI), which constitute a serious obstacle to obtaining high-quality images. Method: We propose a CS-DTI de-aliasing method based on conditional generative adversarial (cGAN), called CS-GAN, to tackle de-aliasing problems in CS-DTI with highly undersampled k-space data. The method uses a nested-UNet based generator, a ResNet-based discriminator, and a content loss function defined in both image domain and frequency domain. Result and Concludions: Compared to existing state-of-the-art de-aliasing methods based on deep learning, our method achieves superior imaging quality in terms of both diffusion weighted (DW) image quality and DTI diffusion metrics. Moreover, even at extremely low sampling ratio and low SNR, our method can still reconstruct texture details and spatial information.

Design of Feedforward Neural Networks in the Classification of Hyperspectral Imagery Using Superstructural Optimization

Artificial Neural Networks (ANNs) have been used in a wide range of applications for complex datasets with their flexible mathematical architecture. The flexibility is favored by the introduction of a higher number of connections and variables, in general. However, over-parameterization of the ANN equations and the existence of redundant input variables usually result in poor test performance. This paper proposes a superstructure-based mixed-integer nonlinear programming method for optimal structural design including neuron number selection, pruning, and input selection for multilayer perceptron (MLP) ANNs. In addition, this method uses statistical measures such as the parameter covariance matrix in order to increase the test performance while permitting reduced training performance. The suggested approach was implemented on two public hyperspectral datasets (with 10% and 50% sampling ratios), namely Indian Pines and Pavia University, for the classification problem. The test results revealed promising performances compared to the standard fully connected neural networks in terms of the estimated overall and individual class accuracies. With the application of the proposed superstructural optimization, fully connected networks were pruned by over 60% in terms of the total number of connections, resulting in an increase of 4% for the 10% sampling ratio and a 1% decrease for the 50% sampling ratio. Moreover, over 20% of the spectral bands in the Indian Pines data and 30% in the Pavia University data were found statistically insignificant, and they were thus removed from the MLP networks. As a result, the proposed method was found effective in optimizing the architectural design with high generalization capabilities, particularly for fewer numbers of samples. The analysis of the eliminated spectral bands revealed that the proposed algorithm mostly removed the bands adjacent to the pre-eliminated noisy bands and highly correlated bands carrying similar information.