In this study, breast cancer classification as benign or malignant was made using images obtained by histopathological procedures, one of the medical imaging techniques. First of all, different noise types and several intensities were added to the images in the used data set. Then, the noise in images was removed by applying the Wavelet Transform (WT) process to noisy images. The performance rates in the denoising process were found out by evaluating Peak Signal to Noise Rate (PSNR) values of the images. The Gaussian noise type gave better results than other noise types considering PSNR values. The best PSNR values were carried out with the Gaussian noise type. After that, the denoised images were classified by Convolution Neural Network (CNN), one of the deep learning techniques. In this classification process, the proposed CNN model and the VggNet-16 model were used. According to the classification result, better results were obtained with the proposed CNN model than VggNet-16. The best performance (86.9%) was obtained from the data set created Gaussian noise with 0.3 noise intensity.
31 December 2021 in Düzce Üniversitesi Bilim ve Teknoloji Dergisi
Runge Kutta (RUN) is an up-to-date and well-founded metaheuristic algorithm. The RUN algorithm aims to find the global best in solving problems by going beyond the traps of metaphors. For this purpose, enhanced solution quality mechanism is used to avoid local optimum solutions and increase the convergence speed. Although the RUN algorithm offers promising solutions, it is seen that this algorithm has shortcomings, especially in solving high dimensional multimodal problems. In this study, the solution candidates that guide the search process in the RUN algorithm are developed using the Fitness-Distance Balance (FDB) method. Thus, using the FDB-based RUN algorithm, the global optimum value of many optimization problems will be obtained in the future. CEC 2020 which has current benchmark problems was used to test the performance of the developed FDB-RUN algorithm. 10 different unconstrained benchmark problems taken from CEC 2020 were designed by arranging them in 30/50/100 dimensions. Experimental studies were carried out using the designed benchmark problems and analyzed with Friedman and Wilcoxon statistical test methods. According to the results of the analysis, it was seen that the FDB-RUN variations showed a superior performance compared to the base algorithm (RUN) in all experimental studies. In particular, it has been shown to provide more effective results for the continuous optimization of high-dimensional problems.
29 May 2021 in Düzce Üniversitesi Bilim ve Teknoloji Dergisi
There has been a significant increase in the use of deep learning algorithms in recent years. Convolutional neural network (CNN), one of the deep learning models, is frequently used in applications to distinguish important objects such as humans and vehicles from other objects, especially in image processing. After the ImageNet Large Scale Visual Recognition Competition (ILSVRC) in 2012, the use of ESA in applications is becoming quite common. With the development of image processing hardware, the image processing process is significantly reduced. Thanks to these developments, the performance of studies on deep learning is increasing. In this study, a system based on deep learning has been developed to detect and classify objects (human, car and motorcycle / bicycle) from images captured by drones. Two datasets, the image set of Stanford University and the drone image set created at Afyon Kocatepe University (AKÜ), are used to train and test the deep neural network with the transfer learning method. Training and testing processes are carried out using a total of 3841 images, 2591 from the Stanford dataset and 1250 from the AKÜ dataset. The precision, recall and f1 score values are evaluated according to the process of determining and classifying human, car and motorcycle / bicycle classes using GoogleNet, VggNet and ResNet50 deep learning algorithms. According to this evaluation result, high performance results are obtained with 0.916 precision, 0.895 recall and 0.906 f1 score value in the ResNet50 model.
31 December 2021 in Düzce Üniversitesi Bilim ve Teknoloji Dergisi
In this study, the performance of Slime-Mould-Algorithm (SMA), a current Meta-Heuristic Search algorithm, is improved. In order to model the search process lifecycle process more effectively in the SMA algorithm, the solution candidates guiding the search process were determined using the fitness-distance balance (FDB) method. Although the performance of the SMA algorithm is accepted, it is seen that the performance of the FDB-SMA algorithm developed thanks to the applied FDB method is much better. CEC 2020, which has current benchmark problems, was used to test the performance of the developed FDB-SMA algorithm. 10 different unconstrained comparison problems taken from CEC 2020 are designed by arranging them in 30-50-100 dimensions. Experimental studies were carried out using the designed comparison problems and analyzed with Friedman and Wilcoxon statistical test methods. According to the results of the analysis, it has been seen that the FDB-SMA variations outperform the basic algorithm (SMA) in all experimental studies.