Browsing by Author "Akagunduz, Erdem"

Now showing 1 - 7 of 7

A Hybrid Framework for Matching Printing Design Files to Product Photos
(2020) Akagunduz, Erdem; Kaplan, Alper; 233834
We propose a real-time image matching framework, which is hybrid in the sense that it uses both hand - crafted features and deep features obtained from a well -tuned deep convolutional network. The matching problem, which we concentrate on, is specific to a certain application, that is, printing design to product photo matching. Printing designs are any kind of template image files, created using a design tool, thus are perfect image signals. For this purpose, we create an image set that includes printing design and corresponding product photo pairs with collaboration of an actual printing facility. Using this image set, we benchmark various hand-crafted (SIFT, SURF, GIST, HoG) and deep features for matching performance. Various segmentation algorithms including deep learning based segmentation methods are applied to select feature regions. Results show that SIFT features selected from deep segmented regions achieves up to 96% product photo to design file matching success in our dataset. We propose a framework in which deep learning is utilized with highest contribution, but without disabling real-time operation using an ordinary desktop computer.
Citation - WoS: 5
Citation - Scopus: 7
Comparison of single channel indices for U-Net based segmentation of vegetation in satellite images
(Spie-int Soc Optical Engineering, 2020) Ulku, Irem; Barmpoutis, Panagiotis; Stathaki, Tania; Akagunduz, Erdem; 233834
Hyper-spectral satellite imagery, consisting of multiple visible or infrared bands, is extremely dense and weighty for deep operations. Regarding problems related to vegetation as, more specifically, tree segmentation, it is difficult to train deep architectures due to lack of large-scale satellite imagery. In this paper, we compare the success of different single channel indices, which are constructed from multiple bands, for the purpose of tree segmentation in a deep convolutional neural network ( CNN) architecture. The utilized indices are either hand-crafted such as excess green index (ExG) and normalized difference vegetation index (NDVI) or reconstructed from the visible bands using feature space transformation methods such as principle component analysis (PCA). For comparison, these features are fed to an identical CNN architecture, which is a standard U-Net-based symmetric encoder-decoder design with hierarchical skip connections and the segmentation success for each single index is recorded. Experimental results show that single bands, which are constructed from the vegetation indices and space transformations, can achieve similar segmentation performances as compared to that of the original multi-channel case.
Citation - WoS: 1
Citation - Scopus: 2
Dynamical system parameter identification using deep recurrent cell networks: Which gated recurrent unit and when?
(Springer London Ltd, 2021) Akagunduz, Erdem; Cifdaloz, Oguzhan; 279762
In this paper, we investigate the parameter identification problem in dynamical systems through a deep learning approach. Focusing mainly on second-order, linear time-invariant dynamical systems, the topic of damping factor identification is studied. By utilizing a six-layer deep neural network with different recurrent cells, namely GRUs, LSTMs or BiLSTMs; and by feeding input/output sequence pairs captured from a dynamical system simulator, we search for an effective deep recurrent architecture in order to resolve the damping factor identification problem. Our study's results show that, although previously not utilized for this task in the literature, bidirectional gated recurrent cells (BiLSTMs) provide better parameter identification results when compared to unidirectional gated recurrent memory cells such as GRUs and LSTM. Thus, indicating that an input/output sequence pair of finite length, collected from a dynamical system and when observed anachronistically, may carry information in both time directions to predict a dynamical systems parameter.
Citation - WoS: 3
Citation - Scopus: 3
Filter design for small target detection on infrared imagery using normalized-cross-correlation layer
(Tubitak Scientific & Technological Research Council Turkey, 2020) Demir, H. Seckin; Akagunduz, Erdem; 233834
In this paper, we introduce a machine learning approach to the problem of infrared small target detection filter design. For this purpose, similar to a convolutional layer of a neural network, the normalized-cross-correlational (NCC) layer, which we utilize for designing a target detection/recognition filter bank, is proposed. By employing the NCC layer in a neural network structure, we introduce a framework, in which supervised training is used to calculate the optimal filter shape and the optimum number of filters required for a specific target detection/recognition task on infrared images. We also propose the mean-absolute-deviation NCC (MAD-NCC) layer, an efficient implementation of the proposed NCC layer, designed especially for FPGA systems, in which square root operations are avoided for real-time computation. As a case study we work on dim-target detection on midwave infrared imagery and obtain the filters that can discriminate a dim target from various types of background clutter, specific to our operational concept.
Citation - WoS: 0
Citation - Scopus: 0
Infrared Target Detection using Shallow CNNs
(Ieee, 2020) Uzun, Engin; Aksoy, Tolga; Akagunduz, Erdem; 233834
Convolutional Neural Networks can solve the target detection problem satisfactorily. However, the proposed solutions generally require deep networks and hence, are inefficient when it comes to utilising them on performance-limited systems. In this paper, we study the infrared target detection problem using a shallow network solution, accordingly its implementation on a performance limited system. Using a dataset comprising real and simulated infrared scenes; it is observed that, when trained with the correct training strategy, shallow networks can provide satisfactory performance, even with scale-invariance capability.
Citation - WoS: 0
Citation - Scopus: 0
Machine Learning-based Silence Detection in Call Center Telephone Conversations
(Ieee, 2019) Iheme, Leonardo O.; Ozan, Sukru; Akagunduz, Erdem; 233834
This study presents the development of a voice activity detection (VAD) system tested on call center telephony data obtained from our local site. The concept of bag of audio words (BoAW) combined with a naive Bayes classifier was applied to achieve the task. It was formulated as a binary classification problem with speech as the positive class and silence/background noise as the negative class. All the processing was performed on the Mel-frequency cepstral coefficients (MFCCs) extracted from the audio recordings. The results which are presented as accuracy score and receiver operating characteristics (ROC) indicate an excellent performance of the developed model. The system is to be deployed within our call center to aid data analysis and improve overall efficiency of the center.
Citation - WoS: 0
Citation - Scopus: 0
Representing Earthquake Accelerogram Records for CNN Utilization
(Ieee, 2020) Cikis, Melis; Tileyoglu, Salih; Akagunduz, Erdem
In this study, a spectrogram based false color representation of earthquake accelergrams is proposed and its usability for both human investigation and its application in convolutional networks are discussed. By using more than forty two thousand earthquake records open to the public, an epicenter clustering algorithm was employed, and it was observed that earthquakes in similar clusters produce similar representations. The prospective purpose of the proposed representation is to estimate the epicenter of an earthquake by processing the accelerograms recorded in a single station using convolutional networks.