Wyniki wyszukiwania - BazTech

1

Brain tumor classification in MRI imagesusing genetic algorithm appended CNN

Balamurugan Thiyagu, Gnanamanoharan E.

Computer Assisted Methods in Engineering and Science

|

2023

|

Vol. 30, no. 3

305--321

EN

Brain tumors are fatal for majority of the patients, the different nature of the tumorcells requires the use of combined medical measures, and categorizing such tumors isa difficult task for radiologists. The diagnostic structures based on PCs have been offeredas an aid in diagnosing a brain tumor using magnetic resonance imaging (MRI). Generalfunctions are retrieved from the lowest layers of the neural network, and these lowestlayers are responsible for capturing low-level features and patterns in the raw input data,which can be particularly unique to the raw image. To validate this, the EfficientNetB3pre-trained model is utilized to classify three types of brain tumors: glioma, meningioma,and pituitary tumor. Initially, the characteristics of several EfficientNet modules are takenfrom the pre-trained EfficientNetB3 version to locate the brain tumor. Three types of braintumor datasets are used to assess each approach. Compared to the existing deep learningmodels, the concatenated functions of EfficientNetB3 and genetic algorithms give betteraccuracy. Tensor flow 2 and Nesterov-accelerated adaptive moment estimation (Nadam)are also employed to improve the model training process by making it quicker and better.The proposed technique using CNN attains an accuracy of 99.56%, a sensitivity of 98.9%,a specificity of 98.6%, an F-score of 98.9%, a precision of 98.9%, and a recall of 99.54%.

2

Ensemble of classifiers based on CNN for increasing generalization ability in face image recognition

Szmurło Robert, Osowski Stanisław

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2022

|

Vol. 70, nr 3

art. no. e141004

EN

The paper considers the problem of increasing the generalization ability of classification systems by creating an ensemble of classifiers based on the CNN architecture. Different structures of the ensemble will be considered and compared. Deep learning fulfills an important role in the developed system. The numerical descriptors created in the last locally connected convolution layer of CNN flattened to the form of a vector, are subjected to a few different selection mechanisms. Each of them chooses the independent set of features, selected according to the applied assessment techniques. Their results are combined with three classifiers: softmax, support vector machine, and random forest of the decision tree. All of them do simultaneously the same classification task. Their results are integrated into the final verdict of the ensemble. Different forms of arrangement of the ensemble are considered and tested on the recognition of facial images. Two different databases are used in experiments. One was composed of 68 classes of greyscale images and the second of 276 classes of color images. The results of experiments have shown high improvement of class recognition resulting from the application of the properly designed ensemble.

3

Rotation Invariance in Graph Convolutional Networks

Mac Nguyen Anh, Nguyen Hung Son

Annals of Computer Science and Information Systems

|

2021

|

Vol. 25

81--90

EN

Convolution filters in deep convolutional networks display rotation variant behavior. While learned invariant behavior can be partially achieved, this paper shows that current methods of utilizing rotation variant features can be improved by proposing a grid-based graph convolutional network. We demonstrate that Grid-GCN heavily outperforms existing models on rotated images, and through a set of ablation studies, we show how the performance of Grid-GCN implies that there exist more performant methods to utilize fundamentally rotation variant features and we conclude that the inherit nature of spectral graph convolutions is able to learn invariant behavior.

4

Comparing the effectiveness of Convolutional Neural Network and Long Short-Term Memory Network for Disaster Based Social Media Messages - Using Thunderstorm and Cyclone as Case Studies

Singla Annie, Agrawal Rajat, Linh Nguyen Thi Dieu

Annals of Computer Science and Information Systems

|

2021

|

Vol. 28

273--275

EN

We present a framework to ameliorate the classification of disaster-related social media messages. In the present work, we have incorporated the Convolutional Neural Network, and Long Short-Term Memory Network. To demonstrate the applicability and effectiveness of the proposed approach, it is applied to the thunderstorm and cyclone Fani dataset. The results indicate that CNN is better than the LSTM model with an accuracy score of 0.9999 (99.99%) and loss score of 0.0410. The output from the research study is helpful for disaster managers to make effective decisions on time.

5

Analiza modeli splotowych w zadaniach klasyfikacji dźwieków środowiskowych

Piczak Karol J.

Przegląd Telekomunikacyjny + Wiadomości Telekomunikacyjne

|

2020

|

nr 7-8

366-369, CD

PL

Splotowe sieci neuronowe są obecnie popularnym narzędziem wykorzystywanym w rozpoznawaniu dźwięków środowiskowych. Na skuteczność ich działania wpływa wiele potencjalnych czynników. Niniejszy referat przedstawia podsumowanie wyników uzyskanych w rozprawie doktorskiej autora w zakresie analizy wrażliwości modeli splotowych na dobierane wartości hiperparametrów. W szczególności zastosowanie techniki dropout okazuje się mieć znaczący wpływ na funkcjonowanie tego typu modeli.

EN

Convolutional neural networks are a popular tool used in environmental sound recognition tasks. Their performance depends on multiple factors. This paper presents a summarized extract from author’s PhD dissertation on analyzing the sensitivity of convolutional models to hyperparameter values. In particular, dropout happens to play an important role in these kinds of models.

6

Deep learning approach for open switch fault diagnosis in matrix frequency converter

Sabour Mustapha

Przegląd Elektrotechniczny

|

2020

|

R. 96, nr 11

155--160

EN

This document proposes a new method for detecting and locating open circuit faults in a matrix frequency converter (MC) based on the technique of pattern recognition by neural networks. The converter input and output current signals are used for this purpose. For this, a database of current signals under healthy conditions and defective for different operating conditions was established. After transforming these signals into a Concordia lair, a process of deep learning by a convolutional neural network was carried out. To verify the robustness of our proposed approach, a simulation of a MC system with a defective power electronic switch supplying an asynchronous motor controlled by DTC-SVM under different conditions of torque and speed was developed. The diagnostic results demonstrate the feasibility and effectiveness of the proposed method. It made it possible to locate the faulty switch precisely and quickly.

PL

Zaproponowano nową metodę wykrywania i lokalizowania uszkodzeń obwodu otwartego w przekształtniku matrycowym (MC) w oparciu o technikę rozpoznawania wzorców przez sieć neuronową. W tym celu wykorzystywane są sygnały wejściowe i wyjściowe prądu przekształtnika. Utworzono bazę danych sygnałów prądowych w warunkach znamionowych i z uszkodzeniem dla różnych warunków pracy. Po przekształceniu tych sygnałów w środowisku Concordia przeprowadzono proces głębokiego uczenia się przez splotową sieć neuronową. Aby zweryfikować Wiarygodność naszego proponowanego podejścia, opracowano model symulacyjny układu MC z uszkodzonym łącznikiem energoelektronicznym zasilającym silnik asynchroniczny sterowany metodą DTC-SVM z róznymi wartościami momentu i prędkości obrotwej. Wyniki diagnostyczne pokazują wykonalność i skuteczność proponowanej metody.

7

Worm gear condition monitoring and fault detection from thermal images via deep learning method

Karabacak Yunus Emre, Gürsel Özmen Nurhan, Gümüşel Levent

Eksploatacja i Niezawodność

|

2020

|

Vol. 22, no. 3

544--556

EN

Worm gearboxes (WG) are often preferred, because of their high torque, quickly reducing speed capacity and good meshing effectiveness, in many industrial applications. However, WGs may face with some serious problems like high temperature at the speed reducer, gear wearing, pitting, scoring, fractures and damages. In order to prevent any damage, loss of time and money, it is an important issue to detect and classify the faults of WGs and develop the maintenance plans accordingly. The present study addresses the application of the deep learning method, convolutional neural network (CNN), in the field of thermal imaging that were gathered from a test rig operating on different loads and speeds. Deep learning approaches, have proven their powerful capability to exploit faulty information from big data and make intelligently diagnostic decisions. Studies concerning the condition monitoring of WGs in the literature are limited. This is the first study on WGs with infrared thermography rather than vibration and sound measurements which have some deficiencies about hardware requirements, restricted measurement abilities and noisy signals. For comparison, CNN was also trained, with vibration and sound data which were collected from the healthy and faulty WGs. The results of fault diagnosis show that thermal image based CNN model on WG has achieved 100% success rate whereas the vibration performance was 83.3 % and sound performance was 81.7%. As a result, thermal image based CNN model showed a better diagnosing performance than the others for WGs. Moreover, condition monitoring of WGs, can be performed correctly with less measurement costs via thermal imaging methods.

PL

W wielu zastosowaniach przemysłowych preferuje się przekładnie ślimakowe, ze względu na ich wysoki moment obrotowy, możliwość szybkiej redukcji prędkości i dobrą sprawność zazębienia. Jednakże przekładnie tego typu narażone są często na poważne problemy, takie jak wysoka temperatura przy reduktorze prędkości czy też zużycie, pitting (wżery), zatarcie, pęknięcie lub uszkodzenie kół zębatych. Zapobiec takim uszkodzeniom, i związanym z nimi stratom finansowym i czasowym, można poprzez wykrywanie i klasyfikowanie błędów przekładni i odpowiednie opracowanie planów konserwacji. Niniejsze badanie dotyczy zastosowania metody głębokiego uczenia oraz splotowych sieci neuronowych (SSN) do monitoringu stanu przekładni na podstawie termogramów zarejestrowanych na stanowisku testowym pracującym przy różnych obciążeniach i prędkościach. Podejścia oparte na uczeniu głębokim umożliwiają efektywne wykorzystanie informacji o błędach pochodzących z dużych zbiorów danych i podejmowanie trafnych decyzji diagnostycznych. Niewiele z dostępnych publikacji poświęconych jest monitorowaniu stanu przekładni ślimakowych. Niniejsza praca jako pierwsza przedstawia badania przekładni ślimakowej z zastosowaniem termografii zamiast zwyczajowo prowadzonych pomiarów drgań i dźwięku, które mają pewne wady dotyczące wymagań sprzętowych, ograniczonych możliwości pomiarowych i głośności sygnałów. SNN opartą na danych termicznych porównano z siecią, którą uczono na zbiorach danych wibracyjnych i akustycznych pochodzących z prawidłowo działających i uszkodzonych przekładni ślimakowych. Wyniki diagnostyki uszkodzeń pokazują, że model SSN przekładni ślimakowej oparty na obrazie termicznym osiągnął stuprocentową (100%) skuteczność, podczas gdy skuteczność modeli opartych na danych wibracyjnych i akustycznych wyniosła, odpowiednio, 83,3% i 81,7%. Tym samym, model SNN oparty na obrazie termicznym pozwalał na trafniejsze diagnozowanie przekładni ślimakowej niż pozostałe modele. Ponadto zastosowanie metod opartych na termografii pozwala na poprawne monitorowanie stanu przy niższych kosztach pomiaru.

8

Interpolation merge as augmentation technique in the problem of ship classification

Połap Dawid, Włodarczyk-Sielicka Marta

Annals of Computer Science and Information Systems

|

2020

|

Vol. 21

443--446

EN

Quite a common problem during training the classifier is a small number of samples in the training database, which can significantly affect the obtained results. To increase them, data augmentation can be used, which generates new samples based on existing ones, most often using simple transformations. In this paper, we propose a new approach to generate such samples using image processing techniques and discrete interpolation method. The described technique creates a new image sample using at least two others in the same class. To verify the proposed approach, we performed tests using different architectures of convolution neural networks for the ship classification problem.

9

Improving real-time performance of U-nets for machine vision in laser process control

Dolata Przemysław, Reiner Jacek

Annals of Computer Science and Information Systems

|

2019

|

Vol. 18

29--33

EN

Many industrial machine vision problems, particularly real-time control of manufacturing processes such as laser cladding, require robust and fast image processing. The inherent disturbances in images acquired during these processes makes classical segmentation algorithms uncertain. Among many convolutional neural networks introduced recently to solve such difficult problems, U-Net balances simplicity with segmentation accuracy. However, it is too computationally intensive for usage in many real-time processing pipelines. In this work we present a method of identifying the most informative levels of detail in the U-Net. By only processing the image at the selected levels, we reduce the total computation time by 80%, while still preserving adequate quality of segmentation.

10

Counting instances of objects in color images using U-net network on example of honey bees

Westwańska Weronika W., Respondek Jerzy S.

Annals of Computer Science and Information Systems

|

2019

|

Vol. 18

87--90

EN

This article presents a novel approach to segmentation and counting of objects in color digital images. The objects belong to a certain class, which in this case are honey bees. The authors briefly present existing approaches which use Convolutional Neural Networks to solve the problem of image segmentation and object recognition. The focus however is on application of U-Net convolutional neural network in an environment where knowledge about the object of interest is only limited to its rough, single pixel location. The authors provide full access to the details of the code used to implement the algorithms, as well as the data sets used and results obtained. The results show an encouraging low level of counting error at 14.27% for the best experiment.

11

A Deep learning and multimodal ambient sensing framework for human activity recognition

Yachir Ali, Amamra Abdenour, Djamaa Badis, Zerrouki Ali, khierEddine Amour Ahmed

Annals of Computer Science and Information Systems

|

2019

|

Vol. 18

101--105

EN

Human Activity Recognition (HAR) is an important area of research in ambient intelligence for various contexts such as ambient-assisted living. The existing HAR approaches are mostly based either on vision, mobile or wearable sensors. In this paper, we propose a hybrid approach for HAR by combining three types of sensing technologies, namely: smartphone accelerometer, RGB cameras and ambient sensors. Acceleration and video streams are analyzed using multiclass Support Vector Machine (SVM) and Convolutional Neural Networks, respectively. Such an analysis is improved with the ambient sensing data to assign semantics to human activities using description logic rules. For integration, we design and implement a Framework to address human activity recognition pipeline from the data collection phase until activity recognition and visualization. The various use cases and performance evaluations of the proposed approach show clearly its utility and efficiency in several everyday scenarios.

12

Estimating aesthetic value of images with convolutional neural networks

Suchecki M., Trzciński T.

Elektronika : konstrukcje, technologie, zastosowania

|

2018

|

Vol. 59, nr 4

24--29

EN

Evaluating aesthetic value of digital photographs is a challenging task, mainly due to the subjective manner of the process. In this paper, we propose to approach this problem using deep convolutional neural networks. Using a dataset of over 1.7 million photos collected from Flickr, we train and evaluate a deep learning model whose goal is to classify photographs by analysing their aesthetics. The result of this work is a system that can be used e.g. to improve the workflow of professional photographers by preselecting the best photos.

PL

Ocena estetyki zdjęć cyfrowych jest trudnym zadaniem, głównie ze względu na jego subiektywność. W niniejszym artykule proponujemy podejście do problemu za pomocą głebokich splotowych sieci neuronowych. Ze zbioru ponad 1,7 miliona zdjęć z platformy Flickr, trenujemy model, którego celem jest klasyfikacja fotografii poprzez analizę ich estetyki. Rezultatem jest system, który może być używany np. aby usprawnić pracę profesjonalnych fotografów poprzez wstępne wybranie najlepszych zdjęć.

13

Speeding-up convolutional neural networks: A survey

Lebedev V., Lempitsky V.

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2018

|

Vol. 66, nr 6

799--811

EN

Convolutional neural networks (CNN) have become ubiquitous in computer vision as well as several other domains, but the sheer size of the modern CNNs means that for the majority of practical applications, a significant speed up and compression are often required. Speeding-up CNNs therefore have become a very active area of research with multiple diverse research directions pursued by many groups in academia and industry. In this short survey, we cover several research directions for speeding up CNNs that have become popular recently. Specifically, we cover approaches based on tensor decompositions, weight quantization, weight pruning, and teacher-student approaches. We also review CNN architectures designed for optimal speed and briefly consider automatic architecture search.

14

Detection of driver fatigue symptoms using transfer learning

Jakubowski J., Chmielińska J.

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2018

|

Vol. 66, nr 6

869--874

EN

This paper presents the results of the scientific investigations which aimed at developing the detectors of the selected driver fatigue symptoms based on face images. The presented approach assumed using convolutional neural networks and transfer learning technique. In the conducted research the pretrained model of AlexNet was used. The net underwent slight modification of the structure and then the fine-tuning procedure was applied with the use of an appropriate dataset. In this way all detectors of the selected fatigue symptoms were created. The results of conducted computations indicate that it is potentially possible to apply such an approach to the problem of fatigue symptom detection. The values of the overall misclassification rates for the most troublesome symptom are less than 5.5%, which seems to be a quite satisfactory result.

15

Deep learning versus classical neural approach to mammogram recognition

Kurek J., Świderski B., Osowski S., Kruk M., Barhoumi W.

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2018

|

Vol. 66, nr 6

831--840

EN

Automatic recognition of mammographic images in breast cancer is a complex issue due to the confusing appearance of some perfectly normal tissues which look like masses. The existing computer-aided systems suffer from non-satisfactory accuracy of cancer detection. This paper addresses this problem and proposes two alternative techniques of mammogram recognition: the application of a variety of methods for definition of numerical image descriptors in combination with an efficient SVM classifier (so-called classical approach) and application of deep learning in the form of convolutional neural networks, enhanced with additional transformations of input mammographic images. The key point of the first approach is defining the proper numerical image descriptors and selecting the set which is the most class discriminative. To achieve better performance of the classifier, many image descriptors were defined by means of applying different characterization of the images: Hilbert curve representation, Kolmogorov-Smirnov statistics, the maximum subregion principle, percolation theory, fractal texture descriptors as well as application of wavelet and wavelet packets. Thanks to them, better description of the basic image properties has been obtained. In the case of deep learning, the features are automatically extracted as part of convolutional neural network learning. To get better quality of results, additional representations of mammograms, in the form of nonnegative matrix factorization and the self-similarity principle, have been proposed. The methods applied were evaluated based on a large database composed of 10,168 regions of interest in mammographic images taken from the DDSM database. Experimental results prove the advantage of deep learning over traditional approach to image recognition. Our best average accuracy in recognizing abnormal cases (malignant plus benign versus healthy) was 85.83%, with sensitivity of 82.82%, specificity of 86.59% and AUC = 0.919. These results are among the best for this massive database.

16

Application of convolutional neural networks with anatomical knowledge for brain MRI analysis in MS patients

Stasiak B., Tarasiuk P., Michalska I., Tomczyk A.

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2018

|

Vol. 66, nr 6

857--868

EN

In this paper we consider the problem of automatic localization of multiple sclerosis (MS) lesions within brain tissue. We use a machine learning approach based on a convolutional neural network (CNN) which is trained to recognize the lesions in magnetic resonance images (MRI scans) of the patient’s brain. The training images are relatively small fragments clipped from the MRI scans so – in order to provide additional hints on location of a given clip within the brain structures – we include anatomical information in the training/testing process. Our research has shown that indicating the location of the ventricles and other structures, as well as performing brain tissue classification may enhance the results of the automatic localization of the MS-related demyelinating plaques in the MRI scans.

17

Music Genre Recognition Using Convolutional Neural Networks

Matocha M., Zieliński S. K.

Advances in Computer Science Research

|

2018

|

Nr 14

125--142

EN

The aim of this study was to develop a music genre classifier using convolutional neural networks and to compare its performance with a traditional algorithm based on support vector machines. A distinct feature of the proposed approach was to utilize two-channel stereo signals at the input of the convolutional network. The proposed method yielded similar results compared to those obtained with the traditional approach, demonstrating the potential of the proposed method and indicating the need for its further optimization. Using two-channel stereo signals at the input of the algorithm showed no improvements over the baseline method exploiting single-channel recordings, suggesting that monaural signals fed to the convolutional network might be sufficient to undertake the task of music genre recognition. According to the results, the network ‘prioritized’ the temporal changes over the frequency variations of the signals. This observation tentatively implies that the classifiers specifically designed to account for temporal changes might potentially better serve the task of music genre recognition than the convolutional neural networks.

PL

Celem niniejszej pracy było opracowanie klasyfikatora gatunków muzycznych z użyciem splotowych sieci neuronowych i porównanie go z tradycyjnym algorytmem opartym na maszynie wektorów wspierających. Wyróżniającą cechą zaproponowanego podejścia było wykorzystanie dwu-kanałowego dźwięku stereofonicznego na wejściu sieci splotowej. Zaproponowana metoda dała podobne wyniki do rezultatów otrzymanych z użyciem podejścia tradycyjnego, demonstrując potencjał zaproponowanej metody oraz wskazując na potrzebę jej dalszej optymalizacji. Wykorzystanie dwu-kanałowego dźwięku stereofonicznego na wejściu algorytmu nie poprawiło wyników w porównaniu z metodą bazową wykorzystującą nagrania jednokanałowe, sugerując, iż zastosowanie dźwięków monofonicznych na wejściu splotowej sieci neuronowej jest adekwatne do celów rozpoznawania gatunków muzycznych. Zgodnie z uzyskanymi wynikami, sieć potraktowała priorytetowo zmiany czasowe w porównaniu ze zmianami częstotliwościowymi sygnałów. Obserwacja ta pozwala wstępnie przypuszczać że klasyfikatory specjalnie zaprojektowane, by uwzględnić zmiany czasowe, potencjalnie mogłyby lepiej służyć celom rozpoznawania gatunków muzycznych niż neuronowe sieci splotowe.

18

Current trends in the field of steganalysis and guidelines for constructions of new steganalysis schemes

Czaplewski B.

Przegląd Telekomunikacyjny + Wiadomości Telekomunikacyjne

|

2017

|

nr 10

1121--1125

EN

The paper concerns blind steganalysis techniques in the passive steganalysis scenario designed to detect the steganographic cover modification schemes. The goal is to investigate the state-of-art in the field of steganalysis, and, above all, to recognize current trends existing in this field and determine guidelines for constructions of new steganalysis schemes. The intended effects are to examine the possibilities for the development of knowledge in the field of steganography and to set directions for future research.

PL

Artykuł dotyczy niekoherentnych technik steganalizy w scenariuszu pasywnej steganalizy przeznaczonych do detekcji systemów steganograficznych stosujących metodę modyfikacji obrazów cover. Celem jest zbadanie aktualnego stanu wiedzy w dziedzinie steganalizy, a przede wszystkim rozpoznanie aktualnych kierunków w tej dziedzinie i ustalenie wytycznych dla konstrukcji nowych systemów steganalitycznych. Zamierzonymi efektami są zbadanie możliwości rozwoju wiedzy w dziedzinie steganografii i wyznaczenie celów dla przyszłych badań.

19

Geometric transformations embedded into convolutional neural networks

Tarasiuk P., Pryczek M.

Journal of Applied Computer Science

|

2016

|

Vol. 24, nr 3

33--48

EN

This paper presents a novel extension to convolutional neural networks. While CNNs are known for invariance to object translation, changes to the other parameters could make the image recognition tasks difficult - that includes rotations and scaling. Some improvement in this area could be achieved with embedded geometric transformations used inside the CNNs. In order to provide a practical solution, which allows fast propagation and learning of the modified networks, “fast geometric transformations” are introduced.