Browsing by Subject "neural network"

Now showing 1 - 20 of 20

Access status: Open Access ,
Aerial SLAM algorithm with global alignment using satellite imagery
(Wydawnictwa AGH, 2023) Karbowski, Jakub; Bartoszewski, Bartosz; Kokot, Daria; Kwiecień, Michał Jan; Turlej, Tymoteusz
Simultaneous localization and mapping (SLAM) algorithms can be used in aerial vehicles with a down-facing camera for navigation and terrain mapping. However, they suffer from an incremental position error. Each neighbouring image pair match introduces an alignment error which accumulates to a global position error proportional to the map size. By using a reference satellite image map this error can be reduced to a constant factor. This work proposes a novel image alignment algorithm called Embed Match which can be applied to any SLAM algorithm. Embed Match uses a neural network to represent a map region as a field of embedding vectors which are a semantic representation of the area. By comparing the embedding fields of a satellite image and the camera image, an alignment function is defined. Maximising this function leads to image alignment. The proposed algorithm shows a radically different approach from keypoint-based alignment methods which usually struggle with matching drastically different images, such as satellite and real camera images.
Access status: Open Access ,
Applying neural network in computing filling coefficient of four-stroke internal combustion engine
(2011) Bera, Piotr
Neural networks consist of many simple elements operating in parallel. In supervised training they are capable of finding their own solution to a particular problem, given only examples of proper behavior. It is a very useful method of solving complex, non-linear problems. The following article discusses the usage of artificial neural network to compute the value of filling coefficient of four-stroke internal combustion engines as the function of crankshaft rotational speed and throttle opening angle. The paper presents the idea of a static, two-layer feedforward network trained with the basic backpropagation algorithm in which the weights and biases are updated in the direction of the negative gradient. The article discusses network architecture and data structure, training parameters and result analysis.
Access status: Open Access ,
Automatyzacja procesu badania neuronowego systemu wnioskującego opartego na programie Statistica w praktycznym zastosowaniu
(Wydawnictwa AGH, 2009) Grabska-Chrząstowska, Joanna; Lazar, Wojciech
The paper presents the use of an automated system choosing neural system network parameters in order to classify patients into two groups. Categorisation of spirometric tests was chosen for practical testing of the created software. The results of the system were compared to the earlier published attempt of an empirical choice of network system parameters.
Access status: Open Access ,
Building Semantic Segmentation Using UNet Convolutional Network on SpaceNet Public Data Sets for Monitoring Surrounding Area of Chan Chan (Peru)
(Wydawnictwa AGH, 2024) Chicchon, Miguel; Malinverni, Eva Savina; Sanità, Marsia; Pierdicca, Roberto; Colosi, Francesca; Trujillo, Francisco James León
The amount of damage to cultural heritage sites is increasing rapidly every year. This is due to inadequate heritage management and uncontrolled urban growth as well as unpredictable seismic and atmospheric events that manifest themselves in a continuously deteriorating ecosystem. Thus, applications of artificial intelligence (AI) in remote-sensing (RS) techniques (machine-learning and deep-learning algorithms) for monitoring archaeological sites have increased in recent years. This research involves the surrounding area of the archaeological site of Chan Chan in Peru in particular. An approach that is based on the use of AI algorithms for building footprint segmentation and change-detection analysis by means of RS images is proposed. It involves a UNet convolutional network based on an EfficientNet B0 to B7 encoder. The network was trained on two public data sets from SpaceNet that were based on WV2 and WV3 satellite images: SpaceNet V1 (Rio), and SpaceNet V2 (Shanghai). In the pre-processing phase, the images from the two data sets have been equalized in order to improve their quality and avoid overfitting. The building segmentation has been performed on HRV images of the study area that were downloaded from Google Earth Pro. The value that was achieved in the IoU metric was around 70% in both experiments. The purpose of this proposed methodology is to assist scientists in drafting monitoring and conservation protocols based on already-recorded data in order to prevent future disasters and hazards.
Access status: Restricted ,
Design and implementation of a neural network of object pose regression
(Data obrony: 2019-10-25) Słabuszewska, Anna
Wydział Informatyki, Elektroniki i Telekomunikacji
Access status: Restricted ,
Interpretacja kominów gazowych z obszaru zdjęcia sejsmicznego Trzciana-Cierpisz-Zaczernie 3D
(Data obrony: 2018-01-29) Rogocz, Patryk
Wydział Geologii, Geofizyki i Ochrony Środowiska
Projekt inżynierski polegał na identyfikacji kominów gazowych na zdjęciu sejsmicznym 3D Trzciana-Cierpisz-Zaczernie na podstawie różnych technik interpretacyjnych. Kominy gazowe odznaczają się na obrazie sejsmicznym poprzez pionowe strefy zakłóceń sygnału. Osłabienie amplitud refleksów sejsmicznych lub ich całkowite wygaszenie wiąże się z wpływem migrującego gazu na propagację fali sprężystej w ośrodku. Na podstawie analizy atrybutów sejsmicznych oraz klasyfikacji wykonanej przez sieci neuronowe wytypowano potencjalne miejsca występowania kominów gazowych na obszarze zdjęcia sejsmicznego 3D Trzciana-Cierpisz-Zaczernie. Charakter zapisu sejsmicznego utworów miocenu oraz wpływ nasunięcia karpackiego powodował wyznaczenie przez sieci neuronowe nieprawidłowych lokalizacji kominów gazowych. Główne obszary w których następuje migracja gazu to strefy nad wyniesieniami podłoża miocenu.
Access status: Restricted ,
Klasyfikacja danych tekstowych przy pomocy rekurencyjnych sieci neuronowych
(Data obrony: 2018-01-23) Łyko, Tomasz
Wydział Informatyki, Elektroniki i Telekomunikacji
Access status: Restricted ,
Klasyfikacja obrazów SAR za pomocą sieci neuronowej
(Data obrony: 2014-09-25) Podsiadło, Iwona
Wydział Geologii, Geofizyki i Ochrony Środowiska
The aim of this work was to create a classifier polarimetric signatures using neural networks. To solve the problem has been applied Hamming network and neural networks for pre-processing the data. Environment in which the solution was implemented Matlab 7.1. Hamming classifier has proved to be an adequate tool to solve the problem in the case of coherent polarimetric signatures and for the signatures of coherent network volume did not give satisfactory results.
Access status: Open Access ,
Machine learning methods for diagnosing the causes of die-casting defects
(Wydawnictwa AGH, 2023) Okuniewska, Alicja; Perzyk, Marcin; Kozłowski, Jacek
The research was focused on analyzing the causes of high-pressure die-casting defects, more specifically on casting leakage, which is considered perhaps the most important and common defect. The real data used for modelling was obtained from a high-pressure die-casting foundry that manufactures aluminum cylinder blocks for the world's leading automotive brands. This paper compares and summarizes the results of applying advanced modelling using artificial neural networks, regression trees, and support vector machines methods to select artificial neural networks as the most effective method to perform a multidimensional optimization of process parameters to diagnose the causes of die-casting defects and to indicate the future research scope in this area. The developed system enables the prediction of the level of defects in castings with satisfactory accuracy and is therefore a highly relevant reference for process engineers of high-pressure foundries. This article indicates exactly which process parameters significantly influence the formation of a defect in a casting.
Access status: Open Access ,
Model sieci neuronowej zliczającej obiekty w obrazie
(Wydawnictwa AGH, 2007) Wołoszyn, Paweł
W pracy przedstawiono model sieci neuronowej zbudowany przy użyciu agentowego systemu dynamicznego, który naśladuje niektóre cechy biologicznych komórek nerwowych. Zadaniem systemu jest zliczanie obiektów występujących w prezentowanym sieci obrazie. Rezultaty eksperymentów symulacyjnych wskazują, że sieć jest zdolna realizować postawione zadanie z pewnymi ograniczeniami naśladującymi błędy popełniane przez człowieka.
Access status: Open Access ,
Modelowanie przemysłowego procesu mielenia rudy z wykorzystaniem energetycznych wskaźników oceny
(2006) Trybalski, Kazimierz; Krawczykowski, Damian
The costs analysis of grinding and classification center in one of KGHM »Polska Miedź« SA ore enrichment plants was conducted in the paper, what identified the highest energy consumption of grinding process. The energetic-technological factors evaluating grinding and classification processes were then proposed and calculated. On their basis the examples of models were constructed, which were regressive ones and neural networks forms, taking into consideration dependencies between process evaluation factors and energetic-technological data of investigated process. The comparison of given models was carried out.
Access status: Open Access ,
Próba neuronowego modelowania zawartości radioaktywnego kobaltu w zależności od składu chemicznego wody w reaktorze jądrowym
(Wydawnictwa AGH, 2008) Grabska-Chrząstowska, Joanna
W pracy przedstawiono wykorzystanie sieci neuronowych do tworzenia modelu zależności zawartości pierwiastka promieniotwórczego $^{60}Co$ od zawartości pięciu metali w wodzie reaktora jądrowego. Otrzymano bardzo obiecujące modele procesu, rokujące nadzieje na badanie czułości modelu na zmianę parametrów wejściowych. Równocześnie wykazano ogromną rolę historii pomiarów przy tworzeniu modelu.
Access status: Open Access ,
Próba zastosowania sieci neuronowych do prognozowania osiadań powierzchni terenu powstałych na skutek eksploatacji górniczej
(2006) Pawluś, Dorota
This paper presents an application of neural networks for the prediction of a surface subsidence. The main advantage of the artificial neural network approach is that there is no need to assume the type of functional relation and there is no need to have an accurate knowledge of material properties in the area of interest. Only the geometry of the neural network has to be chosen and the learning procedure has to be successfully completed. There are several types of neural network geometry. The multi-layer feed-forward networks were used for modeling the surface subsidence trough. Neural networks need to learn in order to produce useful results. There are two different kinds of learning: unsupervised learning and supervised learning. The supervised learning has been used. The networks were used as a solution to following problem. There was given excavated quadrangular area which was described by the following factors: the cordinates of vertices of a worked area, the seam thickness, the depth of the opening. We want to predicate the final subsidence of any point P(x,y). The neural networks could be used for computing the surface subsidence. The author will intend to use networks for computing the other factors of the surface deformations.
Access status: Open Access ,
Prognozowanie osiadań powierzchni terenu przy użyciu sieci neuronowych
(2007) Pawluś, Dorota
This paper presents an application of neural networks for the prediction of a surface subsidence. The main advantage of the artificial neural network approach is that there is no need to assume the type of functional relation and there is no need to have an accurate knowledge of material properties in the area of interest. Only the geometry of the neural network has to be chosen and the learning procedure has to be successfully completed. The networks were used as a solution to following problem. There was given excavated quadrangular area which was described by the following factors: the coordinates of vertices of a worked area, the seam thickness, the depth of the opening, an angle of the mining influence and the subsidence factor. We want to predict the final subsidence of any point of surface. The multi-layer feed-forward networks were used for modeling the surface subsidence trough. The supervised learning has been used. Figures 4 and 5 present the final subsidences of the points lying on two lines. The neural networks could be used for computing the surface subsidence. The author will intend to use networks for computing the other factors of the surface deformations.
Access status: Open Access ,
Przydatność różnych typów sieci neuronowych w klasyfikacji gleb
(Wydawnictwa AGH, 2006) Gruszczyński, Stanisław
The application of three neural networks algorithm in task soils classification, on the basis of features obtained from analog cartographic documentation, is presented. The MLP (Multi-Layer Perceptron) type net and PNN (Probabilistic Neural Network) give the best classification results among examined algorithms. The PNN and SOM (Self-Organizing Map) combination of net operational results gives more deep classification relations within sphere this study, based among others on fuzzy relationships visualization between complexes in analyzed area.
Access status: Restricted ,
Recurrent neural network synthesis for control
(Data obrony: 2017-01-20) Solecki, Ignacy
Wydział Elektrotechniki, Automatyki, Informatyki i Inżynierii Biomedycznej
Access status: Open Access ,
Rozpoznawanie kształtów w sekwencjach wizyjnych z zastosowaniem algorytmu wstecznej propagacji błędów
(Wydawnictwa AGH, 2008) Głowacz, Adam; Głowacz, Witold
A new approach to automatic shape recognition is presented. This approach is based on backpropagation neural network. Investigations of the shape recognition were carried out for film sequences. Results of investigations with application of backpropagation neural network show that shape recognition efficiency is very high.
Access status: Restricted ,
Sztuczne sieci neuronowe jako narzędzie wspierające diagnostykę nowotworów piersi
(Data obrony: 2018-01-24) Smoła, Ewelina
Wydział Elektrotechniki, Automatyki, Informatyki i Inżynierii Biomedycznej
Access status: Open Access ,
Time stability of computer generated control in real-time
(Wydawnictwa AGH, 2013) Zwonarz, Wojciech
W pracy przedstawiono analizę zachowania czasowych reżimów manipulatora sterowanego z poziomu komputera klasy PC. Komputer zarządzany jest przez system operacyjny Windows XP, wzbogacony o pakiet obliczeniowy Matlab/Simulink z zainstalowanym przybornikiem RT-CON. Badany manipulator to ramię stanfordzkie o trzech stopniach swobody, napędzane bezprzekładniowymi silnikami o wysokich momentach obrotowych. Przeprowadzono analizę wpływu zjawiska jitteru na zachowanie przez manipulator powtarzalności wymuszeń okresowych. Eksperyment przeprowadzono dla sterowania generowanego przez regulatory PD oraz PD wzbogacony o zmodyfikowaną siecią neuronową. Badano również zachowanie systemu obciążonego dodatkowymi zadaniami.
Access status: Restricted ,
Wykrywanie anomalii w dużych zbiorach danych za pomocą sieci neuronowych
(Data obrony: 2018-01-23) Piwowarczyk, Wojciech
Wydział Informatyki, Elektroniki i Telekomunikacji