No abstract

No abstract

The deep neural network has made the most advanced breakthrough in almost all 2D image tasks, so we consider the application of deep learning in 3D images. Point cloud data, as the most basic and important form of representation of 3D images, can accurately and intuitively show the real world. The authors propose a new network based on feature fusion to improve the point cloud classification and segmentation tasks. Our network mainly consists of three parts: global feature extractor, local feature extractor and adaptive feature fusion module. A multi‐scale transformation network is devised to guarantee the invariance of the transformation of the global feature, and a residual block is introduced to alleviate the problem of gradient disappearance to enhance the global feature extractor. Based on the edge convolution and multi‐layer perceptron, a local feature extractor is constructed. Finally, an adaptive feature‐fusion module is proposed to complete the fusion of global features and local features. Extensive experiments on point cloud classification and segmentation tasks are carried out to verify the effectiveness of the proposed method. The classification accuracy of the ModelNet40 is 93.6%, which is 4.4% higher than that of the PointNet. Similarly, the segmentation accuracy on the ShapeNet is 85.6%, which is higher than other methods.

This paper provides a relative radiometric calibration method based on the linear CCD imaging the same region of nonuniform scene, which makes full use of the ability of yaw angle control to ensure all the linear CCD detectors imaging the same scene. Firstly, when it is needed to perform the satellite relative radiometric calibration task, the initial drift angle will be calculated, according to which the yaw angle can be adjusted to ensure on-orbit satellite performing the calibration imaging mode, and in this mode the linear CCD and the satellite motion are in the approximate direction. Secondly, in calibration imaging process the yaw angle will be continuously adjusted to control the push-broom direction, and the linear CCD camera can be sequentially on the same region of non-uniform scene, which can obtain the remote-sensing image observing the same region with all detectors. Finally, after obtaining the same region image, histogram matching method is used to establish the high-precision nonlinear relative radiometric calibration model, and this method overcomes the nonlinear response problem caused by the camera photon noise, the dark current noise. This method needs neither the on-orbit calibration device, nor the ground uniform scaling field, and the general earth observation scene can meet the requirements. This method does not need a lot of on-orbit imaging data for statistical analysis compared with the statistical method, and each track is scaled to meet the conditions for calibration imaging, which avoids the unreliable problem of the calibration source itself caused by the unstable differences between the different tracks.

Structural similarity (SSIM) is one image quality assessment metric that focuses on the statistic information in the spatial domain. It cannot reflect the small details of the contrast and the changing of texture, which can be perceived by human visual system，because SSIM cannot detect the distortion image with aliasing and blur effectively. This paper proposes a new image quality assessment metric called structural similarity based on global phase coherence (GPC-SSIM), which considers both the structural information in the spatial domain and the phase characteristics in the frequency domain. Through experiments, as the level of blur and aliasing of an image gets more and more serious, the dynamic range of the results obtained through SSIM is 0.6~1, while the ones through the new assessment index GPC_SSIM is 0~1. Thus GPC-SSIM is more sensitive to the blur and aliasing of image and can give more accurate assessment results for various kinds of degraded images than SSIM.Keywords-image quality assessment; phase information; structural similarity (SSIM); global phase coherence (GPC)