Stadium detection from Google and Yandex images with deep learning method

As a result of the rapid development of technology and the progress in computer hardware, the importance of deep learning is increasing day by day. Recently, many institutions/organizations prefer to use autonomous systems instead of manpower. One of the areas where deep learning algorithms are constantly being developed is models that can detect objects. The aim of the thesis study is to develop a model with satellite images provided by two different companies at the same corner coordinate and scale, and to analyze the effect of the satellite images shared by the companies on the success of the model. In the thesis study, Python 3.6 was used as the software language and Anaconda was used as the software developer. Tensorflow library was used as deep learning library, OpenCV library was used as computer vision library and Tensorflow Object Detection Application Programming Interface (API) was used as object detection model. In this context, the 'Faster R-CNN Inception v2 COCO' and 'Faster R-CNN Resnet101 COCO' models trained with the COCO data set were transferred to the user system, and the 'stadium' images from the satellite images shared by Google and Yandex companies via the open source program SAS Planet application were provided and both models were trained with satellite images belonging to both Google and Yandex companies. Accuracy analysis of the study was made using the F1-Score method. The model that gives the most successful result in the F1-Score accuracy metric is the Faster R-CNN Inception v2 COCO, which is trained with the satellite images of the Yandex company with an accuracy rate of 91%.