Semantic segmentation is a fundamental task in remote sensing image analysis (RSIA). Fully convolutional networks (FCNs) have achieved state-of-the-art performance in the task of semantic segmentation of natural scene images. However, due to distinctive differences between natural scene images and remotely-sensed (RS) images, FCN-based semantic segmentation methods from the field of computer vision cannot achieve promising performances on RS images without modifications. In previous work, we proposed an RS image semantic segmentation framework SDFCNv1, combined with a majority voting postprocessing method. Nevertheless, it still has some drawbacks, such as small receptive field and large number of parameters. In this paper, we propose an improved semantic segmentation framework SDFCNv2 based on SDFCNv1, to conduct optimal semantic segmentation on RS images. We first construct a novel FCN model with hybrid basic convolutional (HBC) blocks and spatial-channel-fusion squeeze-and-excitation (SCFSE) modules, which occupies a larger receptive field and fewer network model parameters. We also put forward a data augmentation method based on spectral-specific stochastic-gamma-transform-based (SSSGT-based) during the model training process to improve generalizability of our model. Besides, we design a mask-weighted voting decision fusion postprocessing algorithm for image segmentation on overlarge RS images. We conducted several comparative experiments on two public datasets and a real surveying and mapping dataset. Extensive experimental results demonstrate that compared with the SDFCNv1 framework, our SDFCNv2 framework can increase the mIoU metric by up to 5.22% while only using about half of parameters.

To increase the observation performance of optical remote sensing satellites, onboard camera systems usually perform multi-slice and multi-band imaging. The resulting multi-band or multi-slice sub-images require correction into an aligned and stitched complete image, to provide users with standard image products. In a linear array sensor, multi-band and multi-slice sensor correction processing is achieved by mapping sub-images on the same virtual linear array based on the determined geometric parameters and the principle of object space consistency. However, a planar array sensor has the same geometric parameters outside the camera; therefore, object space consistency is essentially equivalent to image space consistency, which simplifies coordinate mapping calculations. In this study, we developed a novel sensor correction method based on image space consistency for planar array sensors, and combined it with a method to determine the relative geometric parameters between bands. The internal geometric parameters of multiple bands were determined under connection constraints between bands established by consistent image space pointing. Next, the coordinate mapping model between the virtual image point and physical sub-image point was established, and the angular resolution was introduced for coordinate solving. Based on strict geometric correspondence between the sub-images and virtual images, the whole image with inter-band registration and inter-slice stitching was generated from the sub-images. We experimentally verified the accuracy and effectiveness of our method using real multi-band and simulated multi-slice plane array sensor images from GF-4 satellites. The corrected whole image showed satisfactory inter-band registration accuracy and internal geometric accuracy.