Search

Deep learning architectures are applied in the solution of many problems and give very successful results compared to other methods. One of these problems is the Super Resolution problem. In this study, we tried to solve the problem of super resolution by using different deep learning architectures to obtain higher resolution images. The models used in this study are focused on the images scaled up by factors of 2, 3 and 4. As a result of the experimental studies, the model success is increased as the network depth and samples are increased. Instead of a shallow model with more number of parameters, a deep model with lower number of parameters offers more successful results.

25th Signal Processing and Communications Applications Conference (SIU) -- MAY 15-18, 2017 -- Antalya, TURKEY, WOS: 000413813100169, Production of high quality wheat has a great importance especially in the solution of nutrition problems. It is necessary to make decomposition for specifying the quality. Here, high quality and unclassified wheat recognition are realized. The most distinctive feature between high quality and poor quality wheat is the shape difference. In this study, Bag of Contour Fragments (BCF) was used as a shape based descriptor. In contrast to general shape descriptors, BCF represents shape by dividing into small parts rather than entire one. These small fragments are then coded and classified by the Support Vector Machine (SVM) according to their histograms. In the study, a performance of % 70 accuracy rate was achieved., Turk Telekom, Arcelik A S, Aselsan, ARGENIT, HAVELSAN, NETAS, Adresgezgini, IEEE Turkey Sect, AVCR Informat Technologies, Cisco, i2i Syst, Integrated Syst & Syst Design, ENOVAS, FiGES Engn, MS Spektral, Istanbul Teknik Univ

Sinter tesisleri, yüksek fırın kullanılarak demir çelik üretimi yapılan entegre demir çelik fabrikalarının önemli bir parçasıdır. Sinter tesislerinin ana bölümlerinden biri olan sinter makinesinde toz cevherin sinterleşmesi sağlanarak yüksek fırında kullanılmaya elverişli hale getirilir. Bu çalışmada, sinter makinesinin verimini artırmak amacıyla, makine yönetiminin operatörlerden alınıp otomatik yapılması için bir örüntü tanıma sistemi tasarlanmıştır. Tasarlanan bu sistemde sınıflandırıcı olarak doğrusal ayırma sınıflandırıcısı, destek vektör makineleri ve yapay sinir ağı kullanılmıştır. Bu sınıflandırıcılar 4212 adetlik bağımsız veri kümesi üzerinde test edilerek birbiri ile karşılaştırılmıştır., Sinter plant is one of the important unit of the integrated plants, which made iron and steel production via blast furnace. In the sinter machine, which is the main part of the sinter plant fine ore has been provided sintering and rendered suitable for using in the blast furnace. In this study, aimed to increasing sinter machine production a pattern recognition system has been developed for automatic control instead of operator's control. Linear discriminant classifier, support vector machines and artificial neural network is used in designed system. These classifiers were compared to each individual data set by testing over 4212 units.

The probabilistic neural network (PNN) is a special type of radial basis neural network used mainly for classification problems. Due to the size of the network after training, this type of network is usually used for problems with a small-sized training dataset. In this paper, a new training algorithm is presented for use with large training databases. Application to the handwritten digit database shows that the reduced PNN performs better than the standard PNN for all of the studied cases with a big gain in size and processing speed. This new type of neural network can be used easily for problems with large training databases like biometrics and data mining applications. An extension of the network is possible for new training samples and/or classes without retraining.

2012 20th Signal Processing and Communications Applications Conference, SIU 2012 -- 18 April 2012 through 20 April 2012 -- Fethiye, Mugla -- 90786 There are a considerable number of feature extraction methods to be used for the classification of electromyographic (EMG) signals. These features are obtained from the raw EMG data by time domain and time-frequency domain transformations. Time-frequency domain originated features involve a high computational cost. Hence, for the EMG controlled electromechanical prostheses to be readily usable, time domain features are utilized. Previous studies revealed that for a better utilization of the EMG controlled prostheses, the complete signal processing period should be less than 300 ms. In this study, the classification performances of the features in the time domain will be compared. ANFIS neural network has been preferred as the classification structure in line with the wide experience in the literature. The purpose of this study is to put forward a conceptual viewpoint related to the choice of features to be classified in the work towards EMG controlled prostheses development. © 2012 IEEE.

IEEE 17th Signal Processing and Communications Applications Conference -- APR 09-11, 2009 -- Antalya, TURKEY WOS: 000273935600091 In this studs,, phoneme-based isolated Turkish word recognition with Common Vector Approach (CVA) has been performed. CVA has been used to classify phonemes. The phoneme sequence obtained from the classification is decoded into the word using redundant hash addressing (RHA). The phoneme-based speech recognition is more suitable than the word-based speech recognition for implementing applications that use different words in their dictionaries. For that reason in this study the CVA is evaluated to see whether it could be used in phoneme-based word recognition or not. In the experimental study we obtained the word recognition rates 70-80% from random v selected words in METU database. It might be possible to obtain higher recognition rates by improving the CVA and by using different word decoding techniques. IEEE

Although many approaches and algorithms about biometric recognition techniques have been developed and proposed in the literature, relationships about biometrics have not been studied in the field so far. In this study, we have analysised the existence of any relationship between biometric features and we have tried to obtain a biometric feature of a person from another biometric feature of the same person. Consequently, we have designed and introduced a new and intelligent system using a novel approach based on artificial neural networks for generating faces from fingerprints with % 5.68 mean absolute percent errors. Experimental results have demonstrated that it is possible to generate faces from fingerprints without knowing any information about faces. In addition it is shown that fingerprints and faces are related to each other closely. In spite of the proposed system is initial study and it is still under development, the results are very encouraging and promising.