5-layer Neural Network Research Articles

Cross-track infrared sounder cloud fraction retrieval using a deep neural network

The retrieval of cloud fraction in satellite hyperspectral sounder field of view (FOV) is crucial for numerical weather prediction. This study proposes an innovative cloud fraction retrieval model for the hyperspectral infrared sounder - Cross-track Infrared Sounder (CrIS). The model is trained with a deep neural network (DNN), using the CrIS radiation spectra as the predictors and Visible Infrared Imaging Radiometer Suite (VIIRS) cloud mask as the learning target. An ensemble of randomly selected CrIS and VIIRS data are collocated and used as the training dataset. An optimized 5-layer neural network is built to establish the relationship between the CrIS spectra and the cloud fraction calculated from the VIIRS cloud mask within the CrIS FOV. In order to reduce the number of input predictors to enhance the efficiency of the model, a principal component transformation is performed on the original CrIS spectra and only the top 77 principal component scores are adopted as the final predictors. In general, the cloud fraction retrieved from the proposed DNN model are consistent with truth values calculated from the VIIRS cloud mask product. Further analysis on use cases demonstrates a slightly better cloud retrieval result during the daytime than that of the nighttime, and ocean retrievals are more accurate than land retrievals. However, since the relationship between CrIS spectrum and the cloud fraction is nonlinear, the model tends to slightly overestimate the cloud fractions over low cloud coverage regions and underestimate the values over high cloud fraction areas. Even so, the proposed model can still be a useful tool for obtaining cloud fraction information from hyperspectral infrared sounders and has the potential to be used for the numerical weather prediction and climate models, as well as other cloud studies.

Feature Selection Techniques for a Machine Learning Model to Detect Autonomic Dysreflexia.

Feature selection plays a crucial role in the development of machine learning algorithms. Understanding the impact of the features on a model, and their physiological relevance can improve the performance. This is particularly helpful in the healthcare domain wherein disease states need to be identified with relatively small quantities of data. Autonomic Dysreflexia (AD) is one such example, wherein mismanagement of this neurological condition could lead to severe consequences for individuals with spinal cord injuries. We explore different methods of feature selection needed to improve the performance of a machine learning model in the detection of the onset of AD. We present different techniques used as well as the ideal metrics using a dataset of thirty-six features extracted from electrocardiograms, skin nerve activity, blood pressure and temperature. The best performing algorithm was a 5-layer neural network with five relevant features, which resulted in 93.4% accuracy in the detection of AD. The techniques in this paper can be applied to a myriad of healthcare datasets allowing forays into deeper exploration and improved machine learning model development. Through critical feature selection, it is possible to design better machine learning algorithms for detection of niche disease states using smaller datasets.

Fast and Efficient Method for Optical Coherence Tomography Images Classification Using Deep Learning Approach.

The use of optical coherence tomography (OCT) in medical diagnostics is now common. The growing amount of data leads us to propose an automated support system for medical staff. The key part of the system is a classification algorithm developed with modern machine learning techniques. The main contribution is to present a new approach for the classification of eye diseases using the convolutional neural network model. The research concerns the classification of patients on the basis of OCT B-scans into one of four categories: Diabetic Macular Edema (DME), Choroidal Neovascularization (CNV), Drusen, and Normal. Those categories are available in a publicly available dataset of above 84,000 images utilized for the research. After several tested architectures, our 5-layer neural network gives us a promising result. We compared them to the other available solutions which proves the high quality of our algorithm. Equally important for the application of the algorithm is the computational time, which is reduced by the limited size of the model. In addition, the article presents a detailed method of image data augmentation and its impact on the classification results. The results of the experiments were also presented for several derived models of convolutional network architectures that were tested during the research. Improving processes in medical treatment is important. The algorithm cannot replace a doctor but, for example, can be a valuable tool for speeding up the process of diagnosis during screening tests.

High-precision calibration to zoom lens of optical measurement machine based on FNN.

We proposed a calibration method for high-precision zoom lenses of optical measurement machines based on Fully Connected Neural Network (FNN), using a 5-layer neural network instead of a camera calibration model, to achieve continuous calibration of zoom lenses at any zoom setting by calibrating typical zooms. From the experimental verification, the average calibration error of this method is 9.83×10-4mm and the average measurement error at any zoom setting is 0.01317mm. The overall calibration precision is better than that of Zhang's calibration method and can meet the application requirements of a high-precision optical measurement machine. The method proposed in this paper provided a new solution and a new idea for the calibration of zoom lenses, which can be widely used in the fields of precision parts inspection and machine-vision measurement.

Research on Application of Multifractal Spectrum and Neural Network in Financial Market

With the deepening of the financial market’s reform and innovation, to better explain the market's stock price fluctuations, this article uses the traditional efficient market hypothesis to study further. This study is showed that the Shanghai Stock Exchange Index, the Shenzhen Stock Exchange Index, and the Nasdaq Index have the characteristics of "tip fat tail ", which indicates that the distribution is not normal. Based on the theory of multifractal spectrum, empirical analysis and comparison results of multifractal characteristics, three indexes are carried out to analyze the impact of the maturity of different financial markets on stock price fluctuations. Finally, the multifractal spectrum parameters and daily return rate as input variables, an average accuracy rate of 99.436%, are obtained for the Shanghai Composite Index forecast for the next 30 days, using a 5-layer neural network model. It has particular practical significance for controlling and managing financial risks.

Detection of strict left bundle branch block by neural network and a method to test detection consistency

Objective: To develop an automatic algorithm to detect strict left bundle branch block (LBBB) on electrocardiograms (ECG) and propose a procedure to test the consistency of neural network detections. Approach: The database for the classification of strict LBBB was provided by Telemetric and Holter ECG Warehouse. It contained 10 s ECGs taken from the MADIT-CRT clinical trial. The database was divided into a training dataset (N = 300, strict LBBB = 174, non-strict LBBB = 126) and a test dataset (N = 302, strict LBBB = 156, non-strict LBBB = 146). LBBB-related features were extracted by Philips DXL™ algorithm, selected by a random forest classifier, and fed into a 5-layer neural network (NN) for the classification of strict LBBB on the training dataset. The performance of NN on the test dataset was compared to two random forest classifiers, an algorithm applying strict LBBB criteria, a wavelet-based approach, and a support-vector-machine approach. The consistency of NN’s detection was tested on 549 2 min recordings of the PTB diagnostic ECG database. LBBB annotations are not required to measure consistency. Main results: The performance of NN on the test dataset were sensitivity = 91. 7%, specificity = 85.6% and accuracy = 88.7% (PPV = 87.2%, NPV = 90.6%). The consistency score of strict-LBBB and non-strict-LBBB detection was 0.9341 and 0.9973 respectively. Conclusion: NN achieved the highest specificity, accuracy, and PPV. Using random forest for feature selection and NN for classification increased interpretability and reduced computational cost. The consistency test showed that NN achieved high consistency scores in the detection of strict LBBB. Significance: This work proposed an approach for selecting features and training NN for the detection of strict LBBB as well as a consistency test for black-box algorithms.

Estimation of Layered Ink Layout from Arbitrary Skin Color and Translucency in Inkjet 3D Printer

In this paper, we propose a layout estimation method for multi-layered ink by using PSF measurement and machine learning. This estimation can bring various capabilities of color reproduction for the newfangled 3D printer that can apply multi-layered inkjet color. Especially, the control of translucency is useful for the reproduction of skin color that is overpainted flesh color on bloody-red layer. Conventional method of this layer design and color selection depended on the experience of professional designer. However, it is difficult to optimize the color selection and layer design for reproducing complex colors with many layers. Therefore, in this research, we developed an efficiency estimation of color layout for human skin with arbitrary translucency by using machine learning. Our proposed method employs PSF measurement for quantifying the color translucency of overlapped layers. The machine learning was performed by using the correspondence between these measured PSFs and multi-layered printings with 5-layer neural network. The result was evaluated in the CG simulation with the combination of 14 colors and 10 layers. The result shows that our proposed method can derive an appropriate combination which reproduce the appearance close to the target color and translucency.

Identifying twins based on ocular region features using deep representations

In this paper, we tackle the twins identification problem; a challenging task in biometric authentication. The experiments are carried out using deep learning techniques and for that we have proposed a Convolutional Siamese Network (CNN-Siamese) and a multiscale Convolutional Siamese network (MCNN-Siamese). We have also explored and evaluated three pre-trained CNNs namely, ResNet-50, VGG-16 and NASNet-Large, and utilized them in Siamese network after appropriate modifications. In addition, a simple 5-layer neural network (sNN) is also utilised as Siamese subnetwork in the experiments. We have presented compelling experimental evidences in terms of correct classification accuracy (CCR) on CASIA-IrisV4 Twins’ dataset that manifest the effectiveness of ocular biometrics in identifying twins. The results achieve the existing state-of-the-art human level accuracy. Also, the proximity of CCRs of all the models asserts that the ocular regions in twins hold a significant correlation to label them as twins. We have also quantified the cross-domain capability of the proposed subnetworks (i.e. CNN and MCNN) on ND-GFI dataset that outperforms the state-of-the-art methods. Notably, for the ocular biometrics, to the best of our knowledge, there is currently no literature available as of date that explores the association between twins and subsequently unriddles the classification using Deep Learning.

Modeling of facial expression and emotion for human communication system

The goal of this research is to realize a face-to-face communication environment with machine by giving a facial expression to computer system. In this paper, modeling methods of facial expression including 3D face model, expression model and emotion model are presented. Facial expression is parameterized with Facial Action Coding System (FACS) which is translated to the grid's motion of face model which is constructed from the 3D range sensor data. An emotion condition is described compactly by the point in a 3D space generated by a 5-layered neural network and its evaluation result shows the high performance of this model.