Deep learning models have become very popular and widespread in recent years. With the developing technology, the amount of data kept in databases is increasing day by day. In this thesis, it is aimed to process and classify these data kept in data sets by using deep learning-based methods. In this thesis study, 3 different applications were carried out.
In the first application, a data set consisting of images of human movements was used. A data set was created by taking a certain number of image samples from the videos in this data set. An ESA-based model has been developed for the classification of human movements in the created data set. In the second application, using brain MR images, these images were classified as tumor and tumor-free images. In this application, a hybrid model is proposed. In the last application, a data set consisting of chest X-Ray images was used. Successful results were obtained in the proposed hybrid method to classify the images in this data set as pneumonia or normal. In the classification of brain MR images and chest X-Ray images, Resnet50, one of the ESA architectures, was used as the basis in the proposed hybrid model. Publicly available datasets were used in all three studies.
Successful results have been obtained in improved deep learning-based applications. These proposed methods were seen to be more successful than the pre-trained models used in the study. In addition, the applications were compared with similar studies in the literature. |