Global Relation-Aware-Based Oil Detection Method for Water Surface of Catchment Wells in Hydropower Stations

The oil in hydropower station catchment wells is a source of water pollution which can cause the downstream river to become polluted. Timely detection effectively prevent expansion leakage and has important significance for protecting sources. However, poor environment insufficient light on surface make difficult, real-time performance poor. To address these problems, this paper proposes well method based global relation-aware attention mechanism. By embedding mechanism backbone network Yolov5s, main features are highlighted minor information suppressed at spatial channel levels, improving accuracy. Additionally, problem partial loss detail dataset caused by harsh wells, such as dim limited area, single-scale retinex histogram equalization used improve grayscale contrast images, enhancing details images suppressing noise. experimental results show that accuracy proposed achieves 94.1% 89% detecting engine turbine pollution, respectively. Compared with Faster R-CNN, SSD, FSSD algorithms, our reduces problems missing false detection, certain reference wells.