Improved Training for Machine Learning: The Additional Potential of Innovative Algorithmic Approaches.

2021 ◽  
Author(s):  
Octavian Dumitru ◽  
Gottfried Schwarz ◽  
Mihai Datcu ◽  
Dongyang Ao ◽  
Zhongling Huang ◽  
...  
Keyword(s):  
Machine Learning ◽  
Remote Sensing ◽  
Test Data ◽  
Satellite Images ◽  
Training Data ◽  
Data Selection ◽  
Generative Adversarial Networks ◽  
Radar Images ◽  
Basic Work ◽  
Selection Of

<p>During the last years, much progress has been reached with machine learning algorithms. Among the typical application fields of machine learning are many technical and commercial applications as well as Earth science analyses, where most often indirect and distorted detector data have to be converted to well-calibrated scientific data that are a prerequisite for a correct understanding of the desired physical quantities and their relationships.</p><p>However, the provision of sufficient calibrated data is not enough for the testing, training, and routine processing of most machine learning applications. In principle, one also needs a clear strategy for the selection of necessary and useful training data and an easily understandable quality control of the finally desired parameters.</p><p>At a first glance, one could guess that this problem could be solved by a careful selection of representative test data covering many typical cases as well as some counterexamples. Then these test data can be used for the training of the internal parameters of a machine learning application. At a second glance, however, many researchers found out that a simple stacking up of plain examples is not the best choice for many scientific applications.</p><p>To get improved machine learning results, we concentrated on the analysis of satellite images depicting the Earth’s surface under various conditions such as the selected instrument type, spectral bands, and spatial resolution. In our case, such data are routinely provided by the freely accessible European Sentinel satellite products (e.g., Sentinel-1, and Sentinel-2). Our basic work then included investigations of how some additional processing steps – to be linked with the selected training data – can provide better machine learning results.</p><p>To this end, we analysed and compared three different approaches to find out machine learning strategies for the joint selection and processing of training data for our Earth observation images:</p><ul><li>One can optimize the training data selection by adapting the data selection to the specific instrument, target, and application characteristics [1].</li> <li>As an alternative, one can dynamically generate new training parameters by Generative Adversarial Networks. This is comparable to the role of a sparring partner in boxing [2].</li> <li>One can also use a hybrid semi-supervised approach for Synthetic Aperture Radar images with limited labelled data. The method is split in: polarimetric scattering classification, topic modelling for scattering labels, unsupervised constraint learning, and supervised label prediction with constraints [3].</li> </ul><p>We applied these strategies in the ExtremeEarth sea-ice monitoring project (http://earthanalytics.eu/). As a result, we can demonstrate for which application cases these three strategies will provide a promising alternative to a simple conventional selection of available training data.</p><p>[1] C.O. Dumitru et. al, “Understanding Satellite Images: A Data Mining Module for Sentinel Images”, Big Earth Data, 2020, 4(4), pp. 367-408.</p><p>[2] D. Ao et. al., “Dialectical GAN for SAR Image Translation: From Sentinel-1 to TerraSAR-X”, Remote Sensing, 2018, 10(10), pp. 1-23.</p><p>[3] Z. Huang, et. al., "HDEC-TFA: An Unsupervised Learning Approach for Discovering Physical Scattering Properties of Single-Polarized SAR Images", IEEE Transactions on Geoscience and Remote Sensing, 2020, pp.1-18.</p>

scholarly journals Improved SinGAN Integrated with an Attentional Mechanism for Remote Sensing Image Classification

2021 ◽  
Vol 13 (9) ◽  
pp. 1713
Author(s):  
Songwei Gu ◽  
Rui Zhang ◽  
Hongxia Luo ◽  
Mengyao Li ◽  
Huamei Feng ◽  
...  
Keyword(s):  
Remote Sensing ◽  
Real Life ◽  
Attention Mechanism ◽  
Training Data ◽  
Generative Adversarial Networks ◽  
Natural Image ◽  
Remote Sensing Images ◽  
Training Time ◽  
Adversarial Networks ◽  
Remote Sensing Image Classification

Deep learning is an important research method in the remote sensing field. However, samples of remote sensing images are relatively few in real life, and those with markers are scarce. Many neural networks represented by Generative Adversarial Networks (GANs) can learn from real samples to generate pseudosamples, rather than traditional methods that often require more time and man-power to obtain samples. However, the generated pseudosamples often have poor realism and cannot be reliably used as the basis for various analyses and applications in the field of remote sensing. To address the abovementioned problems, a pseudolabeled sample generation method is proposed in this work and applied to scene classification of remote sensing images. The improved unconditional generative model that can be learned from a single natural image (Improved SinGAN) with an attention mechanism can effectively generate enough pseudolabeled samples from a single remote sensing scene image sample. Pseudosamples generated by the improved SinGAN model have stronger realism and relatively less training time, and the extracted features are easily recognized in the classification network. The improved SinGAN can better identify sub-jects from images with complex ground scenes compared with the original network. This mechanism solves the problem of geographic errors of generated pseudosamples. This study incorporated the generated pseudosamples into training data for the classification experiment. The result showed that the SinGAN model with the integration of the attention mechanism can better guarantee feature extraction of the training data. Thus, the quality of the generated samples is improved and the classification accuracy and stability of the classification network are also enhanced.

Nowcasting heavy precipitation over the Netherlands using a 13-year radar archive: a machine learning approach

2021 ◽  
Author(s):  
Eva van der Kooij ◽  
Marc Schleiss ◽  
Riccardo Taormina ◽  
Francesco Fioranelli ◽  
Dorien Lugt ◽  
...  
Keyword(s):  
Machine Learning ◽  
The Netherlands ◽  
Heavy Rainfall ◽  
Predictive Performance ◽  
Heavy Precipitation ◽  
Early Warning Systems ◽  
Training Data ◽  
Short Term ◽  
Data Set ◽  
Radar Images

<p>Accurate short-term forecasts, also known as nowcasts, of heavy precipitation are desirable for creating early warning systems for extreme weather and its consequences, e.g. urban flooding. In this research, we explore the use of machine learning for short-term prediction of heavy rainfall showers in the Netherlands.</p><p>We assess the performance of a recurrent, convolutional neural network (TrajGRU) with lead times of 0 to 2 hours. The network is trained on a 13-year archive of radar images with 5-min temporal and 1-km spatial resolution from the precipitation radars of the Royal Netherlands Meteorological Institute (KNMI). We aim to train the model to predict the formation and dissipation of dynamic, heavy, localized rain events, a task for which traditional Lagrangian nowcasting methods still come up short.</p><p>We report on different ways to optimize predictive performance for heavy rainfall intensities through several experiments. The large dataset available provides many possible configurations for training. To focus on heavy rainfall intensities, we use different subsets of this dataset through using different conditions for event selection and varying the ratio of light and heavy precipitation events present in the training data set and change the loss function used to train the model.</p><p>To assess the performance of the model, we compare our method to current state-of-the-art Lagrangian nowcasting system from the pySTEPS library, like S-PROG, a deterministic approximation of an ensemble mean forecast. The results of the experiments are used to discuss the pros and cons of machine-learning based methods for precipitation nowcasting and possible ways to further increase performance.</p>

Synthetic Sonic Log Generation With Machine Learning: A Contest Summary From Five Methods

2021 ◽  
Vol 62 (4) ◽  
pp. 393-406
Author(s):  
Yanxiang Yu ◽  
◽  
Chicheng Xu ◽  
Siddharth Misra ◽  
Weichang Li ◽  
...  
Keyword(s):  
Machine Learning ◽  
Test Data ◽  
Short Term Memory ◽  
Rock Physics ◽  
Training Data ◽  
Machine Learning Techniques ◽  
Blind Test ◽  
Data Set ◽  
Benchmark Model ◽  
Sonic Log

Compressional and shear sonic traveltime logs (DTC and DTS, respectively) are crucial for subsurface characterization and seismic-well tie. However, these two logs are often missing or incomplete in many oil and gas wells. Therefore, many petrophysical and geophysical workflows include sonic log synthetization or pseudo-log generation based on multivariate regression or rock physics relations. Started on March 1, 2020, and concluded on May 7, 2020, the SPWLA PDDA SIG hosted a contest aiming to predict the DTC and DTS logs from seven “easy-to-acquire” conventional logs using machine-learning methods (GitHub, 2020). In the contest, a total number of 20,525 data points with half-foot resolution from three wells was collected to train regression models using machine-learning techniques. Each data point had seven features, consisting of the conventional “easy-to-acquire” logs: caliper, neutron porosity, gamma ray (GR), deep resistivity, medium resistivity, photoelectric factor, and bulk density, respectively, as well as two sonic logs (DTC and DTS) as the target. The separate data set of 11,089 samples from a fourth well was then used as the blind test data set. The prediction performance of the model was evaluated using root mean square error (RMSE) as the metric, shown in the equation below: RMSE=sqrt(1/2*1/m* [∑_(i=1)^m▒〖(〖DTC〗_pred^i-〖DTC〗_true^i)〗^2 + 〖(〖DTS〗_pred^i-〖DTS〗_true^i)〗^2 ] In the benchmark model, (Yu et al., 2020), we used a Random Forest regressor and conducted minimal preprocessing to the training data set; an RMSE score of 17.93 was achieved on the test data set. The top five models from the contest, on average, beat the performance of our benchmark model by 27% in the RMSE score. In the paper, we will review these five solutions, including preprocess techniques and different machine-learning models, including neural network, long short-term memory (LSTM), and ensemble trees. We found that data cleaning and clustering were critical for improving the performance in all models.

scholarly journals Using Vehicle Synthesis Generative Adversarial Networks to Improve Vehicle Detection in Remote Sensing Images

2019 ◽  
Vol 8 (9) ◽  
pp. 390 ◽  
Author(s):  
Kun Zheng ◽  
Mengfei Wei ◽  
Guangmin Sun ◽  
Bilal Anas ◽  
Yu Li
Keyword(s):  
Remote Sensing ◽  
Traffic Control ◽  
Data Augmentation ◽  
Vehicle Detection ◽  
Training Data ◽  
Generative Adversarial Networks ◽  
Remote Sensing Images ◽  
Adversarial Networks ◽  
Vehicle Data ◽  
Augmentation Strategy

Vehicle detection based on very high-resolution (VHR) remote sensing images is beneficial in many fields such as military surveillance, traffic control, and social/economic studies. However, intricate details about the vehicle and the surrounding background provided by VHR images require sophisticated analysis based on massive data samples, though the number of reliable labeled training data is limited. In practice, data augmentation is often leveraged to solve this conflict. The traditional data augmentation strategy uses a combination of rotation, scaling, and flipping transformations, etc., and has limited capabilities in capturing the essence of feature distribution and proving data diversity. In this study, we propose a learning method named Vehicle Synthesis Generative Adversarial Networks (VS-GANs) to generate annotated vehicles from remote sensing images. The proposed framework has one generator and two discriminators, which try to synthesize realistic vehicles and learn the background context simultaneously. The method can quickly generate high-quality annotated vehicle data samples and greatly helps in the training of vehicle detectors. Experimental results show that the proposed framework can synthesize vehicles and their background images with variations and different levels of details. Compared with traditional data augmentation methods, the proposed method significantly improves the generalization capability of vehicle detectors. Finally, the contribution of VS-GANs to vehicle detection in VHR remote sensing images was proved in experiments conducted on UCAS-AOD and NWPU VHR-10 datasets using up-to-date target detection frameworks.

scholarly journals Machine Learning Based Risk Prediction for Major Adverse Cardiovascular Events

2021 ◽  
Author(s):  
Michael Schrempf ◽  
Diether Kramer ◽  
Stefanie Jauk ◽  
Sai P. K. Veeranki ◽  
Werner Leodolter ◽  
...  
Keyword(s):  
Machine Learning ◽  
Risk Prediction ◽  
Test Data ◽  
Cardiovascular Events ◽  
Prediction Models ◽  
Early Stage ◽  
Training Data ◽  
Major Adverse Cardiovascular Events ◽  
Future Research ◽  
Patients At Risk

Background: Patients with major adverse cardiovascular events (MACE) such as myocardial infarction or stroke suffer from frequent hospitalizations and have high mortality rates. By identifying patients at risk at an early stage, MACE can be prevented with the right interventions. Objectives: The aim of this study was to develop machine learning-based models for the 5-year risk prediction of MACE. Methods: The data used for modelling included electronic medical records of more than 128,000 patients including 29,262 patients with MACE. A feature selection based on filter and embedded methods resulted in 826 features for modelling. Different machine learning methods were used for modelling on the training data. Results: A random forest model achieved the best calibration and discriminative performance on a separate test data set with an AUROC of 0.88. Conclusion: The developed risk prediction models achieved an excellent performance in the test data. Future research is needed to determine the performance of these models and their clinical benefit in prospective settings.

scholarly journals TEXTURE CLUSTERING OF SATELLITE IMAGES USING SELF-ORGANIZING NEURAL NETWORK

2014 ◽  
pp. 15-21
Author(s):  
M. M. Lukashevich ◽  
R. Kh. Sadykhov
Keyword(s):  
Neural Network ◽  
Remote Sensing ◽  
Satellite Images ◽  
Image Data ◽  
Remote Sensing Image ◽  
Radar Images ◽  
Texture Information ◽  
Winner Takes All ◽  
Data Browsing ◽  
Self Organizing

The goal of this paper is to present a texture clustering system for remote sensing image data. Texture information is useful for image data browsing and retrieval. Authors present the results of self-organizing neural network design for solving the clustering task of gray scale remote sensing image data. The architecture of neural network and the learning algorithms for this network such as: algorithm WTA (Winner Takes All), algorithm CWTA (Winner Takes All with Conscience) and classic Kohonen algorithm WTM (Winner Takes Most - the Winner receives more) are considered. Some experimental results using textures of the Brodatz album, multi-spectral and radar images are also represented.

scholarly journals Improving Remote Sensing Multiple Classification by Data and Ensemble Selection

2021 ◽  
Vol 87 (11) ◽  
pp. 841-852
Author(s):  
S. Boukir ◽  
L. Guo ◽  
N. Chehata
Keyword(s):  
Remote Sensing ◽  
Ensemble Learning ◽  
Remote Sensing Data ◽  
Training Data ◽  
Dramatic Improvement ◽  
Ensemble Classifiers ◽  
Learning Framework ◽  
Ensemble Selection ◽  
Multiple Classification ◽  
Selection Of

In this article, margin theory is exploited to design better ensemble classifiers for remote sensing data. A semi-supervised version of the ensemble margin is at the core of this work. Some major challenges in ensemble learning are investigated using this paradigm in the difficult context of land cover classification: selecting the most informative instances to form an appropriate training set, and selecting the best ensemble members. The main contribution of this work lies in the explicit use of the ensemble margin as a decision method to select training data and base classifiers in an ensemble learning framework. The selection of training data is achieved through an innovative iterative guided bagging algorithm exploiting low-margin instances. The overall classification accuracy is improved by up to 3%, with more dramatic improvement in per-class accuracy (up to 12%). The selection of ensemble base classifiers is achieved by an ordering-based ensemble-selection algorithm relying on an original margin-based criterion that also targets low-margin instances. This method reduces the complexity (ensemble size under 30) but maintains performance.

scholarly journals Optimal Length of R-R Interval Segment Window for Lorenz Plot Detection of Paroxysmal Atrial Fibrillation by Machine Learning

2020 ◽  
Author(s):  
Masaya Kisohara ◽  
Yuto Masuda ◽  
Emi Yuda ◽  
Norihiro Ueda ◽  
Junichiro Hayano
Keyword(s):  
Machine Learning ◽  
Atrial Fibrillation ◽  
Likelihood Ratio ◽  
Test Data ◽  
Cross Validation ◽  
Positive Likelihood Ratio ◽  
Training Data ◽  
Segment Length ◽  
Optimal Length ◽  
X 32

Abstract Background Machine learning of R-R interval Lorenz plot (LP) images is a promising method for the detection of atrial fibrillation (AF) in long-term ECG monitoring, but the optimal length of R-R interval segment window for the LP images is unknown. We examined the performance of LP AF detection by differing the segment length using convolutional neural network (CNN). LP images with a 32 x 32-pixel resolution of non-overlapping R-R interval segments with lengths of 10, 20, 50, 100, 200, and 500 beats were created from 24-h ECG data in 52 patients with chronic AF and 58 non-AF controls as training data and in 53 patients with paroxysmal AF and 52 non-AF controls as test data. For each segment length, classification models were made by 5-fold cross-validation subsets of the training data and its classification performance was examined with the test data. Results In machine learning with the training data, the averages of cross-validation scores were 0.995 and 0.999 for 10 and 20-beat LP images, respectively, and >0.999 for 50 to 500-beat images. The classification of test data showed good performance for all segment lengths with an accuracy from 0.970 to 0.988. Positive likelihood ratio for detecting AF segments, however, showed a convex parabolic curve linear relationship to log segment length with a peak ratio of 111 at 100 beats, while negative likelihood ratio showed monotonous increase with increasing segment length. Conclusions This study suggests that the optimal R-R interval segment window length that maximizes the positive likelihood ratio for detecting paroxysmal AF with 32 x 32-pixel LP image is about 100 beats.

Sign in / Sign up

Export Citation Format

Share Document