Search

Cloud detection in satellite imagery is one the most important problems of satellite meteorology. The accuracy of cloud detection significantly determines the quality of other hydrometeorological products. The paper presents an algorithm for detecting clouds in satellite images, which is based on a convolutional neural network with a modified U-Net architecture. Multispectral satellite imagery from the MSU-GS instrument operating onboard Arktika-M No 1 satellite are used as input data. The algorithm accuracy was estimated through machine learning metrics and comparison with reference masks compiled via manual decryption of the satellite images by an experienced image interpreter. In addition, the results are compared with similar products based on data of SEVIRI and VIIRS instruments. The accuracy of a cloud mask obtained following the suggested algorithm is 92% compared to a reference mask for sun-illuminated areas and 89% for dark areas. [ABSTRACT FROM AUTHOR]

Gene Expression Programming (GEP) is a powerful evolutionary algorithm with simple, linear and compact chromosomes, which has been applied in many fields to solve a large variety of complex problems such as logistic regression, function finding and time series prediction. Since online signature data are composed of discrete points, it is difficult to represent by functional forms, resulting in a limited amount of information used in calculating feature values. Hausdorff distance is utilized as a similarity measure to compute the maximum distance between two point sets, which reduces computational complexity compared with other distance measures. The main contributions of this work are: (1) In preprocessing stage, GEP is used to make signature curve continuous and control each parameter to obtain a fitting curve. Curve fitting is to find a suitable function that is the best fitting for a given set of data; (2) In feature extraction stage, curvature and torsion are utilized to construct eight feature sets for characterizing each user's signatures, and then Hausdorff distance is proposed to calculate the distances between feature sets of two signatures to form an eight-dimensional feature vector; (3) In verification stage, combined with Feed-Forward BP Neural Network classifier, distance matrices consisting of feature vectors are trained and tested many times. The best performances can be provided with false rejection rate, false acceptance rate, average error rate and standard deviation. The experimental results implemented on three available online signature databases MCYT-100, SVC2004 and SUSIG indicate the effectiveness and robustness of our proposed method. [ABSTRACT FROM AUTHOR]

This work is devoted to the sentiment classification of Russian sentences into four classes: positive, negative, mixed, and neutral. Unlike most modern works in this area, a mixed-sentiment class of sentences is introduced. Mixed-sentiment sentences contain both positively and negatively inflected speech at the same time. To solve this problem, the following tools were applied: an attention-based LSTM neural network, a dual attention-based GRU neural network, a and BERT neural network with several output layer modifications providing classification into four classes. Experiments comparing the effectiveness of various neural networks have been carried out on three corpora of Russian-language sentences. Two corpora are made up of users reviews: one with clothing reviews and the other with hotel reviews. The third corpus is made up of news articles from Russian publications. The best average weighted F-measure in the experiments of 0.90 is achieved by a BERT model on the clothing review corpus. The best F-measures for positive and negative sentences are noted on the same corpus, amounting to 0.92 and 0.93, respectively. The best classification results for neutral and mixed-sentiment sentences are achieved on the news corpus. For them, the F-measure is 0.72 and 0.58, respectively. As a result of the experiments, a significant superiority of BERT transfer neural networks over the previous generation LSTM and GRU neural networks is shown, which is most pronounced when classifying texts with weakly expressed sentiments. An analysis of the errors shows that "adjacent" sentiment classes (positive/negative and mixed) account for a larger proportion of errors in classification using BERT than "opposite" classes (positive and negative, neutral and mixed). [ABSTRACT FROM AUTHOR]

The present study evaluates the application of the hybrid machine learning methods to detect changes of land use with a focus on agricultural lands through remote sensing data processing. Two spectral images by Landsat 8 were applied to train and test the machine learning model. Feed forward neural network classifier was utilized as the machine learning model in which two evolutionary algorithms including particle swarm optimization and invasive weed optimization were applied for the training process. Moreover, three conventional training methods including Levenberg–Marquardt back propagation (LM), Scaled conjugate gradient backpropagation (SCG) and BFGS quasi-Newton backpropagation (BFG) were used for comparing the robustness and reliability of the evolutionary algorithms. Based on the results in the case study, evolutionary algorithms are not a reliable method for detecting changes through the remote sensing analysis in terms of accuracy and computational complexities. Either BFG or LM is the best method to detect the agricultural lands in the present study. BFG is slightly more robust than the LM method. However, LM might be preferred for applying in the projects due to low computational complexities. [ABSTRACT FROM AUTHOR]

In this report, we are presenting our work on performance analyses of five different neural network classifiers viz. MLP, DL4JMLP, logistic regression, SGD and simple logistic classifier in lung nodule detection using WEKA interface. To the best of our knowledge, this report demonstrates first use of WEKA for comparative performance analyses of neural network classifiers in identifying lung nodules from lung CT-images. A total of 624 handcrafted features from 52 numbers of lung CT-images collected randomly from Lung Image Database Consortium (LIDC) were fed into WEKA to evaluate the performances of the classifiers under four different categories of computation. Performances of the classifiers were observed in terms of 11 important parameters viz. accuracy, kappa statistic, root mean squared error, TPR, FPR, precision, sensitivity, F-measurement, MCC, ROC area and PRC area. Results show 86.53%, 77.77%, 55.55%, 94.44% & 88.88% accuracy as well as 0.91, 0.86, 0.68, 0.91 & 0.93 ROC area for MLP, DL4JMLP, logistic, SGD and simple logistic classifier respectively at tenfold cross-validation by taking 66% of the data set for training and 34% for testing and validation purpose. SGDClassifier has been found the best performing followed by simple logistic classifier for the purpose. [ABSTRACT FROM AUTHOR]

Electrical impedance tomography (EIT) uses non-invasive and non-radiative imaging to detect inhomogeneous electrical properties in tissues. The inverse problem of EIT is a highly nonlinear, ill-posed problem, which causes inaccuracy in target size calculation. We propose a novel approach to discretize the target size and use a neural network (NN) classifier to classify the unknown size in discrete steps. The target size is discretized into distinct steps, and each step can be a unique class. The data is pre-processed with the cumulative distribution transform (CDT) to enhance distinguishability. First, the NN is trained with simulated datasets, divided into time difference (t-d) group and CDT group. After training, the NN classifier is tested by experimental data recorded in a phantom experiment. Linear discriminant analysis (LDA) is performed to assess the distinguishability of classes. There is a significant increase in distance between classes after the CDT pre-processing. The density of the classes has an upward trend with a higher degree of clustering after CDT pre-processing. The CDT data clustering into distinguishable classes is essential to excellent NN classification results. Such an approach is a significant paradigm shift by turning the cumbersome inverse calculation with uncertain accuracy into a classification problem with predetermined step errors. The accuracy and resolution can be further extended by increasing the discretization steps. [ABSTRACT FROM AUTHOR]

Imaging modalities play a major role in early detection and diagnosis of various medical conditions related to the patient. Retinal image segmentation has been taken up for investigation in this research paper to efficiently detect the presence of eye disorder which could be indicators of major onset of conditions like hypertension, cataracts, diabetic retinopathy, age-related macular disorders, etc. A machine learning method for classification of given pixels in the search space into regions containing blood vessels and those that do not contain blood vessels is implemented using a three-stage neural classifier in this paper. Prior to classification, an optimization algorithm namely ant colony optimization derived from nature-inspired phenomena is used to provide an optimal feature vector set to set high standards for the neural network based classification approach. The novelty and merits of the paper lie in back tracing of the segmentation process in which optimization is done first on the preprocessed features followed by classification for segmented output on the optimized features. This results in elimination of redundant feature vectors which tend to occupy much memory as well increase the computational overhead on the process. The entire implemented system is automated by the machine learning process and tested on 30 samples, 15 each on DRIVE and STARE databases. Classification rates of nearly 98% on an average scenario have been achieved for segmentation and 96.5% for abnormality detection. The performances have been compared against Bayesian set models and standalone ANN models. [ABSTRACT FROM AUTHOR]

Exploration (factor, cluster, and decision tree), regression (multiple linear regression), and neural network (regression, classification) models of clustering, approximation and prediction of the stability constants of cation complexes with ionophore antibiotics (nonactin, monactin, dinactin, trinactin, ennatin B, monensin A, and valinomycin) according to the properties of organic solvents (methanol, ethanol, acetonitrile, and nitrobenzene) and cations (Li+, Na+, K+, Rb+, Cs+, Tl+, Ag+, NH4+, Mg2+, Ca2+, Sr2+, Ba2+, and Mn2+) have been developed. It has been shown that neural network performance is better than that of multiple linear regression (the correlation coefficient on the training sample 0.756 compared to 0.697). The neural network classifier and the classification tree has confirmed the clustering of stability of ionophore–cation complexes carried out by the exploratory k-means method by 97.2%. The prognostic capabilities of the constructed multilayer perceptron have been demonstrated. [ABSTRACT FROM AUTHOR]

Models of multiple linear regression and multilayer artificial neural network have been developed for modeling and predicting the stability constants of sodium and potassium coronates basing on the properties of aqueous-organic solvents (water-methanol, water-propan-2-ol, water-acetonitrile, and water-acetone). The values of the coronates stability constants in water-ethanol solvents have been predicted, and the predictions of the models of multiple linear regression and an artificial neural network models have been compared. The contributions of electrostatic, cohesive, and electron-donating interactions to the increase in the stability of the coronates have been quantitatively assessed basing on the models of multiple linear regression and the principle of free energies linearity. Neural network models based on unsupervised (multilayer perceptrons) and supervised (Kohonen networks) learning algorithms have been developed to classify the stability of sodium and potassium coronates. The neural network classifiers have fully confirmed the classification of the coronated stability via the k-means exploration method. [ABSTRACT FROM AUTHOR]

A method for cloud detection using the machine-learning algorithm based on a convolutional neural network is presented. Input data are satellite images received from the MSU-MR multispectral low-resolution scanning unit onboard the Meteor-M No. 2 satellite. The developed method can be an alternative to the traditional algorithms of cloud detection based on the calculation of differential indices and thresholds. The algorithm is verified using the machine-learning metrics, comparing the resulting cloud mask with the reference one obtained by interpreting the satellite image by an experienced meteorologist. It was also compared (for verification) with a similar product based on VIIRS spectroradiometer data. The cloud mask computed using the algorithm allows the automatic thematic processing of satellite images. [ABSTRACT FROM AUTHOR]

Predictive correlation-regression and neural network models for description of the properties of aqueous-organic solvents on the strength of acetic and benzoic acids have been developed. Significant descriptors affecting the dissociation equilibrium of the acids have been found. The features of the solvation (reflected in the electrostatic, cohesion, and electron accepting interactions) on the acids strength have been disclosed. Prediction of the dissociation constants of acetic and benzoic acids has been successfully performed using a three-layer perceptron. Prospects of the neural network modeling for prediction of organic acids strength in aqueous-organic media have been demonstrated. A neural network classifier of acetic and benzoic acids strength based on the aqueous-alcoholic solvents descriptors has been developed and trained. [ABSTRACT FROM AUTHOR]

We present main steps in developing the algorithm of dynamogram classification. They are data processing, procedures of generation and selection of features, constructing of neural network classifier and estimation of its work quality. To estimate the possibility to single out complex defects (subclasses) we analyzed the structure of the input pattern sample with the aid of algorithms of clusterization. [ABSTRACT FROM AUTHOR]

This paper attempts to present a vision-based interface which interacts with computers by hand gesture recognition. This work aims at creating a natural and intuitive application employing both static and dynamic hand gestures. The proposed application can be summarized in three main steps: hands detection in a video, hands tracking and converting hand shapes or trajectories into computer commands. To accomplish this application, a classification phase is paramount whether at the part of hand detection, or at the phase of 'commanding computers'. For this reason, we have proposed to use a wavelet network classifier (WNC) learnt by fast wavelet transform (FWT). To emphasize the robustness of this classifier, we have used a neural network classifier (NNC) version in order to compare the two classifiers' performances aiming at proving the strength of our proposed one. Global rates given by experimental results show the effectiveness of our proposed approaches of hand detection, hand tracking and hand gesture recognition. The comparison of the two classifier's result helps to choose the best classifier, which can improve the performances of our application. [ABSTRACT FROM AUTHOR]

Fuzzy ARTMAP (FAM), which is a supervised model from the adaptive resonance theory (ART) neural network family, is one of the conspicuous neural network classifier. The generalization/performance of FAM is affected by two important factors which are network parameters and presentation order of training data. In this paper we introduce a genetic algorithm to find a better presentation order of training data for FAM. The proposed method which is the combination of genetic algorithm with Fuzzy ARTMAP is called Genetic Ordered Fuzzy ARTMAP (GOFAM). To illustrate the effectiveness of GOFAM, several standard datasets from UCI repository of machine learning databases are experimented. The results are analyzed and compared with those from FAM and Ordered FAM which is used to determine a fixed order of training pattern presentation to FAM. Experimental results demonstrate the performance of GOFAM is much better than performance of Fuzzy ARTMAP and Ordered Fuzzy ARTMAP. In term of network size, GOFAM performs significantly better than FAM and Ordered FAM. [ABSTRACT FROM AUTHOR]

The objective of this work is to develop and implement a medical decision-making system for an automated diagnosis and classification of ultrasound carotid artery images. The proposed method categorizes the subjects into normal, cerebrovascular, and cardiovascular diseases. Two contours are extracted for each and every preprocessed ultrasound carotid artery image. Two types of contour extraction techniques and multilayer back propagation network (MBPN) system have been developed for classifying carotid artery categories. The results obtained show that MBPN system provides higher classification efficiency, with minimum training and testing time. The outputs of decision support system are validated with medical expert to measure the actual efficiency. MBPN system with contour extraction algorithms and preprocessing scheme helps in developing medical decision-making system for ultrasound carotid artery images. It can be used as secondary observer in clinical decision making. [ABSTRACT FROM AUTHOR]

The work develops the approximation approach to solving the inverse MTS problem with the use of neural networks. The inverse problem is considered in model classes of parametrized geoelectric structures, whose electric conductivity is controlled by a few hundreds of macroparameters ( N ∼ 300). An approximate inverse operator of the problem is constructed for each model class as a neural network, whose coefficients are determined in the process of training on a representative sample of standard examples of forward problem solutions. The problem of determination of the model class of geolectric structures corresponding to the presented input MT data is solved with the use of the neural network classifier constructed for the available set of model classes of structures. Regularizing factors and errors of the neural network method are analyzed. The operation of the algorithm is illustrated by examples of the 2-D inversion of synthetic MT data. [ABSTRACT FROM AUTHOR]

In this study, a Multi-Layer BP neural network (MLBP) with dynamic thresholds is employed to build a classifier model. As to the design of the neural network structure, theoretical guidance and plentiful experiments are combined to optimize the hidden layers’ parameters which include the number of hidden layers and their node numbers. The classifier with dynamic thresholds is used to standardize the output for the first time, and it improves the robustness of the model to a high level. Finally, the classifier is applied to forecast box office revenue of a movie before its theatrical release. The comparison results with the MLP method show that the MLBP classifier model achieves more satisfactory results, and it is more reliable and effective to solve the problem. [ABSTRACT FROM AUTHOR]

Previously, we proposed a new method to identify fabric pilling and objectively measure fabric pilling intensity based on the two-dimensional dual-tree complex wavelet reconstruction and neural network classification. Here we further evaluate the robustness of the method. Our results indicate that the pilling identification method is robust to significant variation in the brightness and contrast of the image, rotation of the image, and 2 i ( i is an integer) times dilation of the image. The pilling feature vector developed to characterize the pilling intensity is robust to brightness change but is sensitive to large rotations of the image. As long as all fabric images are adjusted to have the same contrast level and the sample is illuminated from the same direction, the pilling feature vectors are comparable and can be used to classify the pilling intensity. [ABSTRACT FROM AUTHOR]

The mitotic index (MI) is an important measure in cell proliferation studies. Determination of the MI is usually made by light-microscope analysis of slide preparations. The analyst identifies and counts thousands of cells and reports the percentage of mitotic shapes found, among the interphase nuclei. Full automation of this process is an ambitious task, because there can exist very few mitotic shapes among hundreds of nuclei and thousands of artifacts, resulting in a high probability of false positives, i.e. objects erroneously identified as mitosis or nuclei. A semiautomated approach for MI calculation is reported, based on the development of a neural network (NN) for automatic identification of metaphase spreads and stimulated nuclei in digital images of microscope preparations at 10X magnification. After segmentation of the objects on each image, ten different morphometrical, photometrical and textural features are measured on each segmented object. An NN is used to classify the feature vectors into three classes: metaphases, nuclei and artifacts. The system has been able to classify correctly approximately 91% of the objects in each class, in a test set of 191 mitosis, 331 nuclei and 387 artifacts, obtained from 30 different microscope slides. Manual editing of false positives from the metaphase classification results allows the calculation of the MI with an error of 6.5%. [ABSTRACT FROM AUTHOR]

Based on the features extracted from generalized autoregressive (GAR) model parameters of the received waveform, and the use of multilayer perceptron(MLP) neural network classifier, a new digital modulation recognition method is proposed in this paper. Because of the better noise suppression ability of the GAR model and the powerful pattern classification capacity of the MLP neural network classifier, the new method can significantly improve the recognition performance in lower SNR with better robustness. To assess the performance of the new method, computer simulations are also performed. [ABSTRACT FROM AUTHOR]

When a woman diagnosed as having breast cancer has a tumour removed, it is important to try and predict whether she is likely to relapse within, say, the next three years. In this paper, the performance of a neural network classifier trained on a number of prognostic indicators is shown to be better than that of the clinical experts working with the same information. To obtain meaningful statistics with the relatively small dataset available, the network is trained using a modified form of the leave-one-out method. A procedure is also introduced for investigating how much independent information each input parameter contributes. This shows that, in this type of retrospective study, the type of therapy given to the woman does not significantly affect the network's prediction of whether or not she will relapse within three years. Finally, since this problem, in common with many other medical problems, is plagued by a shortage of data, the final section of the paper reports on an investigation of whether or not multi-centre databases might be feasible. [ABSTRACT FROM AUTHOR]