A method for fast detection of wind farms from remote sensing images using deep learning and geospatial analysis

Abstract The wind power industry is increasing worldwide every year. Thus, obtaining timely and detailed information on wind farms’ number and spatial distribution is critical for quantitatively estimating wind energy utilization and for planning the construction of new wind farms. Therefore, this study proposes a method for quickly identifying wind farms in a large-scale area. Given that wind farms mainly comprise individual objects such as wind turbines and substations, we labeled sample images of wind turbines and substations on a global scale. Then, these sample images are used to train a target recognition model and an object classification model and detect the specific locations of wind turbines and substations in the study area. Additionally, we deeply analyzed the location features of the wind turbines and further improved the recognition accuracy based on these known features using geographic constraints. Based on the location information of wind turbines and substations, a clustering model organizes them effectively into complete wind farms. A comprehensive evaluation of the clustering model verifies its scientific validity and reliability. Specifically, this framework was systematically tested throughout Vietnam with remarkable results, using high-resolution historical images provided by Google Earth. Indeed, our framework achieved 90.45% recall and 95.73% accuracy for wind turbines and 81.37% recall and 78.96% accuracy for substations. Finally, we successfully obtained the spatial location and distribution of 15 completed wind farms, demonstrating that the proposed scheme can quickly and accurately identify wind power plants in a large-scale area, which supports wind power management and energy utilization planning.