Detection of environmental waste out of uav images by using deep learning methods

Since plastic and glass bottles are generally disposable materials, they are thrown away carelessly after use. One of the main reasons of environmental pollution is plastic and glass bottle waste. Furthermore, glass bottles have a huge impact on forest fires. For these reasons, it is necessary to detect plastic and glass bottles and collect them without harming the environment. Unmanned aerial vehicles (UAV) are widely used in various sectors today. The increase in the use of unmanned aerial vehicles and the ease of access -when compared to the past- have paved the way for development in many fields. Many problems have been solved at lower costs by using less manpower. In this thesis, image processing techniques were used to detect plastic and glass bottles in images taken from UAVs. Principally, the images with the objects to be detected for training were determined. Objects in these images were labeled, necessary arrangements were made, then training and test data were obtained. Using these data, training was carried out with the YOLOv3 deep learning algorithm, and object detection test was applied with the obtained data. Images of plastic and glass bottle wastes were used for this test process. As a result, plastic and glass bottles in the nature were detected with a great accuracy rate. Since there is no similar study for the protection of the ecosystem, especially for the detection of plastic and glass bottles, I think that the thesis I have prepared will benefit future studies.