关键词: 迁移学习, 卷积神经网络, 光伏电池板, 中等分辨率遥感影像, 场景尺度

Abstract: The acquisition of spatial information of photovoltaic (PV) panels is of great significance to the monitoring of resources and environment and the assessment of energy structure. It has been proved that it is feasible to extract spatial information from remote sensing images based on scene units. The traditional PV power plants recognition depends on the middle and low level characteristics of artificial design. It can not overcome the problems of spectral uncertainty and complex object space structure. The algorithms have problems of low robustness and low efficiency. Although some scholars have used the depth model to classify the image scene, most of the algorithms are based on a few standard remote sensing image scene databases, which do not take into account the influence of the image quality, the scene boundary and the scale on the depth feature of the image scene in the actual application and it restricts the deep application of remote sensing technology in urban structure,economic and social knowledge mining. To this aim, this paper adopts the strategy of transfer learning and model adjustment to identify the scene of PV panels in medium resolution remote sensing images. The results show that the proposed method can extract the multi-level features of the PV panels and achieve good recognition results in the PV panels with complex morphological structure.

Key words: transfer learning, convolutional neural network, photovoltaic panels, medium resolution remote sensing image, scene scale