Papers by Asst. Prof. Dr. Selin Üzelaltınbulat
Procedia Computer Science, 2016
Gaze Prediction Based on Convolutional Neural Network
Proceedings of International Conference on Emerging Technologies and Intelligent Systems, 2021
Journal of Food Quality, 2021
This study presents the design of an intelligent system based on deep learning for grading fruits... more This study presents the design of an intelligent system based on deep learning for grading fruits. For this purpose, the recent residual learning-based network “ResNet-50” is designed to sort out fruits, particularly bananas into healthy or defective classes. The design of the system is implemented by using transfer learning that uses the stored knowledge of the deep structure. Datasets of bananas have been collected for the implementation of the deep structure. The simulation results of the designed system have shown a great generalization capability when tested on test (unseen) banana images and obtained high accuracy of 99%. The simulation results of the designed residual learning-based system are compared with the results of other systems used for grading the bananas. Comparative results indicate the efficiency of the designed system. The developed system can be used in food processing industry, in real-life applications where the accuracy, cost, and speed of the intelligent sys...
Spatiotemporal Precipitation Modeling by AI Based Ensemble Approach
This study aimed at time-space estimations of monthly precipitation via a two-stage modeling fram... more This study aimed at time-space estimations of monthly precipitation via a two-stage modeling framework. In temporal modeling as the first stage, three different AI models were applied to observed precipitation data from seven stations located in the Turkish Republic of Northern Cyprus (TRNC). In this way two scenarios were examined, each employing a specific inputs set. Afterwards, the outputs of single AI models were used to generate ensemble techniques to improve the performance of the precipitation predictions by the single AI models. To end this aim, two linear and one nonlinear ensemble techniques were proposed and then, the obtained outcomes were compared. In the second stage, for estimation of the spatial distribution of precipitation over whole region, the results of temporal modeling were used as inputs for the IDW spatial interpolator. The cross-validation was finally applied to evaluate the overall accuracy of the proposed hybrid spatiotemporal modeling approach. The obta...
Stacked Autoencoders Deep Learning Approach for Left Ventricular Localization in Magnetic Resonance Slices
Proceedings of International Conference on Emerging Technologies and Intelligent Systems, 2021
Environmental Earth Sciences, 2019
This study aimed at time-space estimations of monthly precipitation via a two-stage modeling fram... more This study aimed at time-space estimations of monthly precipitation via a two-stage modeling framework. In temporal modeling as the first stage, three different Artificial Intelligence (AI) models were applied to observed precipitation data from 7 gauges located at Northern Cyprus. In this way 2 different input scenarios proposed, by employing different input combinations. Afterwards, the outputs of single AI models were used to generate ensemble techniques to enhance the precision of modeling by the single AI models. For this purpose, 2 linear and 1 non-linear methods of ensembling were designed and afterwards, the results were evaluated. In the second stage, for estimation of the spatial distribution of precipitation over whole region, the results of temporal modeling were used as inputs for the Inverse Distance Weighting (IDW) spatial interpolator. The cross-validation was finally applied to evaluate the overall accuracy of the proposed hybrid spatiotemporal modeling approach. The obtained results in temporal modeling stage demonstrated that the non-linear ensemble technique provided more accurate results. Results of spatial modeling stage indicated that IDW scheme is a good choice for spatial estimation of the precipitation. The overall results show that the combination of temporal and spatial modeling tools could simulate the precipitation appropriately by serving unique features of both tools.
Arabian Journal of Geosciences, 2019
Procedia Computer Science, 2017
Peer-review under responsibility of the scientific committee of the 9th International Conference ... more Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of Soft Computing, Computing with Words and Perception.
Theoretical and Applied Climatology, 2019
The target of the current paper was to examine the performance of three Markovian and seasonal ba... more The target of the current paper was to examine the performance of three Markovian and seasonal based artificial neural network (ANN) models for one-step ahead and three-step ahead prediction of monthly precipitation which is the most important parameter of any hydrological study. The models proposed here are feed forward neural network (FFNN, as a classic ANN-based models), Wavelet-ANN (WANN, as a hybrid model), and Emotional-ANN (EANN, as a modern generation of ANN-based models). The models were used to precipitation prediction of seven stations located in the Northern Cyprus. Two scenarios were examined each having specific inputs set. The scenario 1 was developed for predicting each station's precipitation through its own data at previous time steps, while in scenario 2, the central station's data were also imposed into the models in addition to each station's data, as exogenous inputs. The obtained results showed the better performance of the EANN model in comparison with other models (FFNN and WANN) especially in three-step ahead prediction. The superiorities of the EANN model over other models are due to its ability in dealing with error magnification in multi-step ahead prediction. Also, the results indicated that the performance of the scenario 2 was better than scenario 1, showing improvement of modeling efficiency up to 17% and 26% in calibration and verification steps, respectively.
Procedia Computer Science, 2017
The methodology for estimation of the impact of changes made in legislative base of tax system to... more The methodology for estimation of the impact of changes made in legislative base of tax system to the tax revenues is proposed in the paper. The adjustment of time series of tax revenues and relevant computer simulation of them was conducted by the fuzzy numbers according to the expert evaluation. Double expert estimation methodology was used in order to ensure the maximum reliability and truthfulness of the existing information. The relevant fuzzy regression dependency of gross domestic product (GDP) on the time series of tax revenues, the relevant fuzzy regression dependency on the time series of adjusted tax revenues was arranged.
Procedia Computer Science, 2016
In this paper the design of recognition system for retinal images using neural network is conside... more In this paper the design of recognition system for retinal images using neural network is considered. Retina based recognition is perceived as the most secure method for identification of an identity used to distinguish individuals. The retina recognition stages including retina image acquisition, feature extraction and classification of the features are discussed. The structure of the neural network based retina identification is presented. Training of neural network based recognition system is performed using backpropagation algorithm. The structure of neural networks used for retina recognition and its learning algorithm are described. The implementation of recognition system has been done using MATLAB package.
Deep Learning Based on Residual Networks for Automatic Sorting of Bananas
Hindawi-Artificial Intelligence in Food Quality Improvement, 2021
This study presents the design of an intelligent system based on deep learning for grading fruits... more This study presents the design of an intelligent system based on deep learning for grading fruits. For this purpose, the recent residual learning-based network “ResNet-50” is designed to sort out fruits, particularly bananas into healthy or defective classes. The design of the system is implemented by using transfer learning that uses the stored knowledge of the deep structure. Datasets of bananas have been collected for the implementation of the deep structure. The simulation results of the designed system have shown a great generalization capability when tested on test (unseen) banana images and obtained high accuracy of 99%. The simulation results of the designed residual learning-based system are compared with the results of other systems used for grading the bananas. Comparative results indicate the efficiency of the designed system. The developed system can be used in food processing industry, in real-life applications where the accuracy, cost, and speed of the intelligent system will enhance the production rate and allow meeting the demand of consumers. The system can replace or assist human operators who can exert their energy on the selection of fruits.
Stacked Autoencoders Deep Learning Approach for Left Ventricular Localization in Magnetic Resonance Slices
Lecture Notes in Networks and Systems book series (LNNS,volume 322), 2021
Deep learning (DL) is an effective method for medical object detection. Studies show that deep ne... more Deep learning (DL) is an effective method for medical object detection. Studies show that deep networks can achieve accuracy in medical segmentation and detection tasks. This is due to the depth and training methods of deep networks which allows them to derive different levels of abstractions of input mages. In this paper, the left ventricle detection task is carried out using a deep network called stacked auto-encoder (SAE). The networks take off this task as a binary classification task wherein left and non-left ventricles cropped images are being recognized by the SAE. Once the network recognizes left and non-left ventricles, the whole task starts by initiating a sliding window that moves through the whole magnetic resonance (MR) slice till a left ventricle is detected. Experimentally, the network showed effective detection performance when target images are noisy as it is seen that it can detect left ventricles in target images with up to 10% of salt and pepper noise.
Gaze Prediction Based on Convolutional Neural Network
Lecture Notes in Networks and Systems book series (LNNS,volume 322), 2021
In this paper gaze prediction aims to acquire good performance via “Convolutional Neural Network ... more In this paper gaze prediction aims to acquire good performance via “Convolutional Neural Network (CNN)” based identification. Gaze prediction was proposed to localize the gaze direction of the students in classrooms using captured face images. Such systems can raise the teachers’ awareness of whether students are gazing towards him or not and alert the students to pay attention to their teachers if their gazes were scattered. In this way, the prediction of gaze was proposed firstly to achieve the proposed eyes area detection algorithm based on modeling facial gestures. The inputs of the gaze prediction are the output of the eye detection. Secondly, the localizing gaze of the students had been considered as straight, right, or left. CNN was employed to different training models for the gaze prediction phase. The dataset was obtained from a public database for neural network training.
Spatiotemporal precipitation modeling by artificial intelligence-based ensemble approach
Environmental Earth Science, 2020
This study aimed at time-space estimations of monthly precipitation via a two-stage modeling fram... more This study aimed at time-space estimations of monthly precipitation via a two-stage modeling framework. In temporal modeling as the first stage, three different Artificial Intelligence (AI) models were applied to observed precipitation data from 7 gauges located at Northern Cyprus. In this way 2 different input scenarios proposed, by employing different input combinations. Afterwards, the outputs of single AI models were used to generate ensemble techniques to enhance the precision of modeling by the single AI models. For this purpose, 2 linear and 1 non-linear methods of ensembling were designed and afterwards, the results were evaluated. In the second stage, for estimation of the spatial distribution of precipitation over whole region, the results of temporal modeling were used as inputs for the Inverse Distance Weighting (IDW) spatial interpolator. The cross-validation was finally applied to evaluate the overall accuracy of the proposed hybrid spatiotemporal modeling approach. The obtained results in temporal modeling stage demonstrated that the non-linear ensemble technique provided more accurate results. Results of spatial modeling stage indicated that IDW scheme is a good choice for spatial estimation of the precipitation. The overall results show that the combination of temporal and spatial modeling tools could simulate the precipitation appropriately by serving unique features of both tools.
Emotional artificial neural networks (EANNs) for multi-step ahead prediction of monthly precipitation; case study: northern Cyprus
Theoretical and Applied Climatology, 2019
The target of the current paper was to examine the performance of three Markovian and seasonal ba... more The target of the current paper was to examine the performance of three Markovian and seasonal based artificial neural network (ANN) models for one-step ahead and three-step ahead prediction of monthly precipitation which is the most important parameter of any hydrological study. The models proposed here are feed forward neural network (FFNN, as a classic ANN-based models), Wavelet-ANN (WANN, as a hybrid model), and Emotional-ANN (EANN, as a modern generation of ANN-based models). The models were used to precipitation prediction of seven stations located in the Northern Cyprus. Two scenarios were examined each having specific inputs set. The scenario 1 was developed for predicting each station’s precipitation through its own data at previous time steps, while in scenario 2, the central station’s data were also imposed into the models in addition to each station’s data, as exogenous inputs. The obtained results showed the better performance of the EANN model in comparison with other models (FFNN and WANN) especially in three-step ahead prediction. The superiorities of the EANN model over other models are due to its ability in dealing with error magnification in multi-step ahead prediction. Also, the results indicated that the performance of the scenario 2 was better than scenario 1, showing improvement of modeling efficiency up to 17% and 26% in calibration and verification steps, respectively.
Spatiotemporal Precipitation Modeling by AI Based Ensemble Approach
Advances in Intelligent Systems and Computing book series (AISC,volume 1095), 2019
This study aimed at time-space estimations of monthly precipitation via a two-stage modeling fram... more This study aimed at time-space estimations of monthly precipitation via a two-stage modeling framework. In temporal modeling as the first stage, three different AI models were applied to observed precipitation data from seven stations located in the Turkish Republic of Northern Cyprus (TRNC). In this way two scenarios were examined, each employing a specific inputs set. Afterwards, the outputs of single AI models were used to generate ensemble techniques to improve the performance of the precipitation predictions by the single AI models. To end this aim, two linear and one nonlinear ensemble techniques were proposed and then, the obtained outcomes were compared. In the second stage, for estimation of the spatial distribution of precipitation over whole region, the results of temporal modeling were used as inputs for the IDW spatial interpolator. The cross-validation was finally applied to evaluate the overall accuracy of the proposed hybrid spatiotemporal modeling approach. The obtained results in temporal modeling stage demonstrated that the non-linear ensemble method revealed higher prediction efficiency.
Comparative Analysis of Artificial Intelligence Based Methods for Prediction of Precipitation. Case Study: North Cyprus
Advances in Intelligent Systems and Computing book series (AISC,volume 896), 2018
Prediction of precipitation is important for design, management of water resources systems, plann... more Prediction of precipitation is important for design, management of water resources systems, planning, flood predicting and hydrological events. This study aimed to compare the performance of three different “Artificial Intelligence (AI)” techniques which are “Artificial Neural Network (ANN), Adaptive Neuro Fuzzy Inference System (ANFIS) and Least Square Support Vector Machine (LSSVM)” to estimate monthly rainfall in Kyrenia Station of Turkish Republic of Northern Cyprus (TRNC). The monthly data covering ten years’ precipitation were used for the predictions. The comparative results showed that the LSSVM model can cause a bit more reliable performance in regard to ANN and ANFIS.
Estimation of impact of the changes made to the tax legislation to the tax receipts through fuzzy numbers
Procedia Computer Science, 2017
The methodology for estimation of the impact of changes made in legislative base of tax system to... more The methodology for estimation of the impact of changes made in legislative base of tax system to the tax revenues is proposed in the paper. The adjustment of time series of tax revenues and relevant computer simulation of them was conducted by the fuzzy numbers according to the expert evaluation. Double expert estimation methodology was used in order to ensure the maximum reliability and truthfulness of the existing information. The relevant fuzzy regression dependency of gross domestic product (GDP) on the time series of tax revenues, the relevant fuzzy regression dependency on the time series of adjusted tax revenues was arranged.
Lung tumor segmentation algorithm
Procedia Computer Science, 2017
This paper is a development of an algorithm based medical image processing to segment the lung tu... more This paper is a development of an algorithm based medical image processing to segment the lung tumor in CT images due to the lack of such algorithms and approaches used to detect tumor where most of researches involve machine learning to solve such segmentation problem. The work involves different image processing tools which successfully achieved the required goals when combined and successively applied. The segmentation system comprises of different stages to finally reach its target which is to segment the lung tumor. Image pre-processing takes place first where some enhancement techniques are used to enhance and reduce noise in images. The next stage is where the different parts in the images are seperated to be able to segment the tumor in later stages. In this phase threshold was selected automatically which assures the right selection of all images since the tumor have different gray-levels intensities in each image. Another technique was also used here to remove the tumor from the thresholded image. Finally, the lung tumor is accurately segmented by subtracting the thresholded and the other image.
Uploads
Papers by Asst. Prof. Dr. Selin Üzelaltınbulat