Cloud contamination is a common phenomenon in the optical remote sensing field, which limits their application in land surface studies and causes the waste of satellite images. This letter presents a new framework for removing thin clouds from visible images based on multiscale dark channel prior (MDCP). The cloud removal of cloudy images (target images) is carried out with the assistance of a different temporal cloudless image (reference image) from the perspective view of multiscale transform (MST). In order to make it more suitable for the application of thin cloud removal, two improvements are made to this traditional fusion method. For one thing, a dark channel prior module is integrated into the low-frequency component of the target image in the framework of MST. For another, we choose the weighted average for high-frequency components and sparse representation (SR) for low-frequency components as fusion rules. After the fusion process, the modified Laplacian sharpening whose model is optimized is carried out. The performance of MDCP was evaluated with both simulated and real cloudy images. Experimental results show that the proposed MDCP has a good performance.

When it comes to multisource collaboration issues, previous registration-based studies always attempt to build reliable connections between two sources. However, these pixel-oriented algorithms often highly rely on mapping functions or require other auxiliary data. In this letter, a target-oriented multisource association model (ToMsAM) is proposed. It is a new perspective of multiinformation integration driven by practical applications in remote sensing. The two main parts in ToMsAM gradually use the similarity information in pixel and feature level. First, to cover the comprehensive prior information, a triple-unit-graph is built. Then, a feature-constraint deep neural networks for graph representations (FC_DNGRs) model is adopted to automatically learn the compressed representation. So far, the heterogeneous samples are unified in a new feature space. These new abstractions associated with the source domain can solve the problem of insufficient samples in the target domain. Real remote sensing images acquired by synthetic aperture radar (SAR) and optical satellites are utilized to conduct experiments on airplane and vessel detection tasks. Experimental results show that the representation obtained by ToMsAM outperformed the baseline representations. The support vector machine (SVM) trained by associated features can complete the detection tasks even in the case of limited quantity of prior samples in the target domain.

We present a deep sensor fusion method based on frustum point single shot multibox detector (PointSSD) for autonomous driving scenarios. The proposed method solves the problem of precision degradation in frustum PointNets (F-PointNet) caused by relying heavily on 2D detection and making insufficient use of RGB information. The method mainly consists of two subnetworks: pyramid segmentation network (PSNet) and PointSSD. The proposed PSNet uses a novel architecture capable of performing semantic segmentation on RGB information to generate high quality image semantic information. Using these image semantic information, point cloud semantic information is obtained through projection and is then fused with raw 3D spatial features by deep fusion. The fusion results are processed by PointSSD, which is proposed for classification and bounding box regression. Evaluated on the KITTI dataset, our method is superior to other methods in 3D classification and 3D localization. In addition, our method guarantees robustness to 2D false detections.

Long-wave infrared (LWIR) images have important applications in retrieving land surface temperature. However, LWIR images are often inevitably suffered from stripe noise, a special type of spatial domain fixed pattern noise. This paper proposes a novel spectral separation algorithm for LWIR image destriping. Specifically, since the mid-wave infrared (MWIR) bands contain both radiant and reflective energy, we use MWIR as reference images to remove the stripe noise in the LWIR bands. We formulate the spectral separation problem as a convex optimization problem, where the difference between LWIR and the radiant component of MWIR, and the difference between the visible and near infrared (VNIR) image and the reflective component of MWIR are regularized to exploit the similarities between the corresponding bands. The obtained radiant component is then utilized to recover the LWIR band. Experimental results using Gaofen-5 datasets demonstrate that the proposed algorithm has good performance in removing the stripe noise.

Remote sensing images contaminated by clouds cannot be used for target recognition, image classification, and other applications, which leads to a lot of remote sensing data being wasted. We propose a low-rank and sparse constrained dark channel prior for cloud removal in remote sensing image sequence (LRSC-DCP). The sparse and low-rank constraints are used to find the position of thick clouds and remove thick clouds in cloud-contaminated images respectively. The remaining thin clouds can be eliminated by the dark channel prior. We compare the algorithm proposed in this article with the dark channel prior. Experimental results prove that the proposed method has better performance.

The cloud-contaminated phenomenon in the field of remote sensing has a serious impact on image processing so that a large number of images are unusable. To achieve cloud removal for single visible image, we propose a novel method based on modified dark channel prior with multiple scale (MDCPMS). In the structure of multiple scale, the cloudy image is firstly decomposed into high-frequency and low-frequency components. The former is uniformly amplified to enhance its weak contour, and the latter is processed by modified dark channel prior (DCP), whose estimation of atmospheric light is optimized for better cloud removal. Finally, the cloud-removed image is obtained through multi-scale reconstruction. Experimental results show that the proposed MDCPMS obtains a significant performance with slightest color distortion and is closest to the corresponding real image, compared with DCP and nonlocal method.