DETECTION AND LOCALIZATION OF LEAKAGES AT OIL PRODUCTION FACILITIES USING COMPUTER VISION

The process of oil production, transportation and refining is accompanied by the risk of environmental disasters associated with leaks and spills. Therefore, timely detection and elimination of such incidents is an important task for oil producing enterprises. Traditional monitoring methods based on visual inspection of objects by operators are labor-intensive, require significant time and do not always allow timely detection of leaks. The purpose of the work is to identify the most effective model of an artificial neural network and develop a program for automated detection of oil spills and oil leaks at oil producing facilities, assess the accuracy and completeness of the results. In this case, images can be obtained using stationary cameras, from unmanned aerial vehicles or satellites. In this work, images obtained from unmanned aerial vehicles are used. 1118 labeled images for oil spill detection and 698 images for oil leak detection were prepared for training and testing sets. A neural network of the YOLO model is proposed to solve the problem of leak detection and localization. The main advantage of this model is its accuracy and speed with minimal computational costs. A software application with a web interface was developed, trained and tested on various sets of real images. 70 epochs were used for training. The results of assessing the accuracy and completeness of oil spill detection using the proposed artificial neural network model were 82% and 79%, respectively. The estimates of the accuracy and completeness of the oil leak detection results were 70% and 80%, respectively. Thus, the obtained results confirm the effectiveness and applicability of the proposed approach and model in real oil production conditions. The use of this system will allow operators to significantly reduce the time and effort spent on manual monitoring of objects, increase the speed of response to emergency situations, which will help minimize environmental damage and reduce production risks.