Analysis Of Images Obtained By Unmanned Aerial Vehicle By Deep Learning Methods

In artificial intelligence applications, many sub-methods such as machine learning, artificial neural networks, classification, clustering algorithms are used. One of these methods is deep learning. Deep learning is an advanced machine learning class. Using the Deep Learning method, video analysis, image classification, speech recognition and natural language processing are very successful. The data and experiences to be provided by the projects covering Deep Learning and Unmanned Aerial Vehicles will increase the number of qualified studies on these issues and contribute to the development of high value-added products in these technologies. In this study, a control software that evaluates image data from unmanned aerial vehicles and makes various inferences (classification, positioning, marking) is created. By using the method of retraining the last layers of the pre-trained artificial neural network models with our data set, it has been tried to reduce the training time and increase the success. In these studies, 2 pre-educated models were used and as a result of training of these models, as a result of 190 thousand steps of training, 25.39 and 27.87 mAP values were reached.