Fast Tailings Pond Mapping Exploiting Large Scene Remote Sensing Images by Coupling Scene Classification and Sematic Segmentation Models

In the process of extracting tailings ponds from large scene remote sensing images, semantic segmentation models usually perform calculations on all small-size images segmented by sliding window method. However, some these do not have ponds, and their only affect model accuracy, but also speed. For this problem, we proposed a fast pond extraction method (Scene-Classification-Sematic-Segmentation, SC-SS) that couples classification models. The can map rapidly accurately in images. There were two parts method: model, model. Among them, adopted lightweight network MobileNetv2. With help network, scenes containing be quickly screened out interference without reduced. used U-Net to finely segment objects scenes. addition, encoder was replaced VGG16 with stronger feature ability, which improves model’s accuracy. paper, Google Earth Luanping County create dataset dataset, based datasets, training testing completed. According experimental results, accuracy (Intersection Over Union, IOU) SC-SS 93.48%. IOU 15.12% higher than while time shortened 35.72%. This research is great importance dynamic observation scale.