No abstract

No abstract

In recent decades, it is easy to obtain remote sensing images which have been successfully applied to various applications, such as urban planning, hazard monitoring, etc. In particular, high resolution (HR) remote sensing (RS) images can better monitor our living environment from a broader spatial perspective. However, raw remote sensing images provide no labeling information to train a classifier, which usually is exploited to generate remote sensing maps. Based on our previous work, in the paper, an automatic classification system is proposed to classify high resolution urban RS images using deep neural networks, in particular, convolutional neural networks and fully convolutional networks. The labeling information is assigned on the context of both social media photos and HR remote sensing images by significantly reducing the cost of manual labeling without the necessity of remote sensing experts. The experiments carried out on high resolution remote sensing images acquired in the city Frankfurt taken by the Jilin-1 satellites confirm the effectiveness of the proposed strategy compared to the state of the art.

In real applications, there is a lack of labeled data to train a proper deep neural network (DNN) model for map generation of remote sensing images. The aim of newly acquired data in spaceborne or airborne platforms is often to consistently observe the Earth for new tasks in the applications such as disaster monitoring, climate change, disease control. To fulfill the tasks, the corresponding classification maps should be obtained traditionally based on the assumption that a classification model should be learnt by the labeled data for the same task from the same scene or at least from the historical labeled remote sensing image pixels provided by domain experts in the same areas by the same sensor, which is denoted as labeled target data. In the paper, a universal deep semantic segmentation lifecycle is proposed against the assumption aforementioned, i.e., there is no need to have the labeled data for the same/similar task from the same locations and sensors to define a proper DNN model. In particular, a general labeled dataset is generated through a feature binding strategy in terms of real-world existed remote sensing images, which is named RSImageNet. In addition, a special training strategy is proposed by using the RSImageNet dataset to train a universal deep semantic segmentation model with a balanced constraint for the loss function. Without the labeled target data from the area observed, we gain an average overall accuracy of 77.32% in the range of 67.28-94.63% on 6 real world datasets by taking advantage of the proposed universal deep semantic segmentation lifecycle and the generated RSImageNet dataset.

No abstract