Impervious Surface Mapping from High Resolution Remote Sensing Imagery Using Object Based Deep CNNs

Impervious surface plays an important role in urban planning and sustainable environmental management. High-spatial-resolution (HSR) images containing pure pixels have significant potential for the detailed delineation of land surfaces. However, due to high intra-class variability and low inter-class distance, the mapping and monitoring of impervious surfaces in complex town-rural areas using HSR images remains a challenge. The recent significant progress in deep convolution neural networks (CNNs) makes them an effective strategy for connecting complex image patterns and semantic categories. The fully convolutional network (FCN) model, a variant of CNNs, recently achieved state-of-the-art performance in HSR image classification applications. However, due to the inherent nature of FCN processing, it is challenging for an FCN to precisely capture the boundaries of classification targets. To solve this problem, we propose an object-based deep CNN framework that integrates object-based image analysis (OBIA) with deep CNNs to accurately extract and estimate impervious surfaces. Specifically, two widely used transfer learning technologies were investigated using both fine and coarse segmentation scales. Finally, we compare our method with conventional OBIA classification, FCN-8s and the U-Net approach. Our experimental results show that our method achieves the highest accuracy of greater than 90% for both producers and users. Our findings also suggest that the overall accuracy is related to both the segmentation scale and training strategy. In addition, the proposed object-based deep CNNs with transfer learning technology can be easily applied to extract additional types of complex objects from HSR images. Our approach for the automatic extraction and mapping of impervious surfaces also lays a solid foundation for intelligent monitoring and the management of land use and land cover.