REAL-TIME MONITORING AND EARLY WARNING SYSTEM FOR UNMANNED AERIAL VEHICLE INTEGRATED WITH PIPELINE MONITORING USING DEEP LEARNING WITH YOLOV5

Because they are the most popular way to transport hydrocarbons fluids, gases, and fossil fuels globally, pipelines are crucial to the oil and gas sector. Maintaining infrastructure stability and error-free output from these pipelines calls for constant monitoring. A reliable monitoring system that can identify small leaks that numerous sensors could miss is still desperately needed, despite the fact that pipeline monitoring has come a long way. The unmanned aerial vehicle's (UAV) capacity to collect extremely useful data has increased its applicability in a number of fields. Precise object tracking has been crucial for UAVs used for surveillance in the pipeline. Problems like occlusion and shifts in object representation are typical during UAV-based pipeline monitoring. It is difficult to keep track of the objects in this kind of environment. In this work, we proposed the real-time monitoring and early warning system for UAV integrated with pipeline monitoring using deep learning with YOLOv5. Firstly, we introduced the hybrid DenseNet with stochastic dual elemental algorithm for feature extraction and optimization which extracts more deep features from the gathered pipeline images and optimizes the feature to selects best of them. We then integrate YOLOv5 with deep recurrent neural network (improved YOLOv5) for pipeline threat events detection which ensures the real-time monitoring and early warning like cracks and leakages. To validate the performance of proposed and existing systems using the publicly available sources and the results show their effectiveness. The maximum detection accuracy is achieved by an improved YOLOv5 is 97.856% which is 12.92% higher than the existing state-of-art systems.