Search

In this paper, we present a web-based optical character recognition (OCR) system that converts images of Ottoman documents printed with naskh font into text using CNN+RNN-based deep neural network models. For training, three datasets - original, synthetic, and hybrid - were prepared and three different OCR models were created. The original data set consists of 1,000 pages and the synthetic data set consists of 23,000 pages. Hybrid data set contains both. The trained models were compared with Tesseract's Arabic and Persian, Google Docs' Arabic, Abby FineReader's Arabic, and Miletos OCR model/tools with a 21-page test set. The comparison was made with 3 different texts (raw, normalized, and joined) and using 3 different criteria (character, ligature, and word recognition). The Osmanlica.com Hybrid model produced significantly better results than the others with 88.86% raw, 96.12% normalized, and 97.37% joined accuracy in character recognition; 80.48% raw, 91.60% normalized, and 97.37% joined accuracy in ligature recognition; and 44.08% raw and 66.45% normalized accuracy in word recognition. To investigate the effects of the characteristics of the alphabet on OCR, character, ligature, and word frequency analyses of Ottoman was performed. In this analysis, the characters in the alphabet were grouped according to distinctive features such as connectedness, letter body, position and number of dots, type of character, and source language; and frequencies and recognition accuracies were examined for each group. OCR results are also reported for each character. [ABSTRACT FROM AUTHOR]

Intelligent transportation systems use parameters such as traffic flow, density and speed to manage city traffic. This paper presents a novel prediction model for traffic speed prediction consisting of nine stages. In the presented model, real vehicle data were passed through data filtering and map matching processes, density-based clusters were created, cluster features were generated, instant traffic state was displayed, and traffic speed prediction was performed using with the artificial neural network RNN model. In previous studies, while traffic speed prediction can be performed on a specific road with stationary data sources or on different days with distributed GPS records, in the developed model, characteristics of the interested road are created by obtaining density-based vehicle cluster features, short-term and data-driven speed prediction is made within the changeable structure of traffic. Speed prediction was tested on Eskisehir and Istanbul roads belonging to Ankara province, the error rates were determined for speed prediction using the RNN variant LSTM and GRU methods, Eskisehir road LSTM-GRU error rates were measured as 8,595-8,656 and İstanbul road error rates as 7,331-7,955, respectively. The developed model for the changeable nature of traffic has yielded successful results in near real time. It is considered that the proposed model will offer different and new solutions in the prediction of traffic parameters, accelerate the processes and assist to the users more accurate and faster services. [ABSTRACT FROM AUTHOR]

In this paper, a new hyperspectral image classification method based on 2-dimensional complex Gabor filtering and deep convolutional neural networks is proposed. Specifically, as a deep learning model, convolutional neural network is aimed to extract distinctive high-level features. Deep-learned and Gabor feature extraction methodologies are simultaneously performed on the input hyperspectral samples. Gabor features are calculated by implementing complex Gabor filtering only on the first three principal components of the hyperspectral image. The proposed hybrid model uses Gabor transform to obtain local image features, such as edges, corners and texture. The Gabor features of the images are calculated at multiple orientations and frequencies. Then, deep features and Gabor features are fused to obtain a more robust and discriminative feature vector. Hybrid feature vector is used as input to a softmax classifier for hyperspectral image classification. The parameters of the proposed deep learning architecture are optimized using a small training set. Thus, the over-fitting problem of the proposed convolutional neural network has been reduced to some extent. Experiments performed on two popular hyperspectral datasets show that the proposed method can achieve better classification performance than some conventional methods. Classification results demonstrates that the proposed hybrid model is an efficient method for feature extraction and classification of hyperspectral images. [ABSTRACT FROM AUTHOR]

In mass customization using resources efficiently with changing customer demands and manufacturing conditions is vital, since in today's competitive market conditions, this efficiency provides cost, time and labor savings to companies. In production environment, Process Planning (PP) and Scheduling are two functions that provide efficient usage of resources. Isolation and long time gap between these functions are main problems affecting the effiency of production. In this study, to solve the problems of productivity, it is aimed to build an integrated system which is able do PP and Scheduling in parallel and respond quickly to fluctuations in job floor. In existing integration models aiming to eliminate this isolation and time gap, it has been observed that if the search and solution space expands, the computational time increases rapidly. Therefore, in this research, a hybrid optimization approach, which can find the optimal solution rapidly is considered and a hybrid model combining both Genetic Algorithm (GA) and Artificial Neural Network (ANN) is proposed. To improve GA performance and increase the effiency of searching, clustering activities are carried out for building new cromosome structures. To increase population diversity, effective genetic operator schemes and efficient genetic represantations are used. In the integration module, 3 different GA structures created within the scope of our research are compared and the algorithm formed by clustering method shows better performance than the others. In this paper by using ANN method, a new system trained by data obtained from Scheduling is generated and this system is able to quickly respond the changes in shop floor and provide new schedules instantly. In rescheduling module, ANN's performance measures provide evidence to accuracy of Heuristic solution generated by Integration module. [ABSTRACT FROM AUTHOR]

In this study, Turkey's Carbonization Index (v) depending on the basic energy indicator has determined as an analytical equation. In this paper, artificial neural networks (ANN) which is quite often in recent years and the possibility of applying to many fields, are used to obtain analytical expressions. Carbonization Index is defined the amount of CO2 emissions per energy consumption. This parameter is an important for Turkey because of has signed the Kyoto protocol, February 2009. According to the results, accuracy of the carbonization index formulated empirical was obtained as R²= 1. It is expected that this study will be helpful in demonstrating carbonization index for policy makers. [ABSTRACT FROM AUTHOR]

In this paper, the experimental studies of different flow density values have been carried on a designed and implemented flow meter being known as ultrasound working based on high frequency sound wave technology. The calibration of the flow meter has been realized using the results obtained from the study. The transit-time method has been used as the basis of the flow measure. For calibration of the flow meter, the experimental study has been carried on twenty different flow densities and the differences between the measured and the exact values have been observed. Then, to minimize these differences, two different calibration methods being 'linearequation' and 'artificial-neural-network' methods have been developed and the results have been obtained. Using the results, the two mentioned methods have been compared and those effects on the accuracy of the flow meter have been discussed. It has been shown that between these two methods the accuracy of the 'neuralnetwork' method is higher than that of the 'linear-equation' method. [ABSTRACT FROM AUTHOR]