Search

Fire radiative power (FRP) is a key indicator for evaluating the intensity of wildfires, unlike traditional real-time fire lines or combustion areas that only provide binary information, and its accurate prediction is more important for firefighting actions and environmental pollution assessment. To this end, we used a combination of data from geostationary satellites and polar orbit satellites to correct the FRP data. Incorporating various factors that affect wildfire spread, such as meteorological conditions, topography, vegetation indexes, and population density, we constructed a comprehensive California wildfire spread dataset, covering information since 2017. Then, we established a deep learning framework that integrates various modules to analyze multimodal data for the accurate prediction of FRP imagery. We investigated the impact of input sequence length and loss function design on model predictive performance, leading to subsequent model optimization. Furthermore, our model has demonstrated acceptable performance in transfer learning and multistep prediction, emphasizing its application value in wildfire prediction and management. It can provide more detailed information about wildfire spread, showcasing the powerful capability of deep learning to process multimodal data and its potential in the emerging field of real-time FRP prediction.

Polarimetric decomposition is an effective way for synthetic aperture radar (SAR) data interpretation. As for hybrid compact polarimetric (HCP) SAR, there are mainly two different types of decomposition methods, including wave-dichotomy-theorem-based and model-based. However, the existing methods are not universal because they cannot estimate volume scattering component accurately in different scenes. In order to solve this problem, a general two-stage model-based three-component HCP decomposition method (GTM) is proposed by involving Arii volume scattering model. In the first stage, a method is designed to identify the dominant scattering mechanism (surface, dihedral, and volume scattering mechanism) of each pixel according to the characteristics of the three scattering models. Then, the GTM method is solved when the three basic scattering mechanisms are dominant, respectively, in the second stage. Two simulated HCP datasets are used to verify the performance of the decomposition method. The results show that GTM can more accurately identify the dominant scattering mechanism in different regions than existing methods, and it can obtain closer results to the quad-polarimetric method.

Accurate delineation of cardiac structures in CTA images remains challenging, due to the similar appearance of different components, the highly variable shape and intensity among individuals, and especially the vicinity of heart volumes and backgrounds (including fat tissues, great vessels, and livers). Therefore, we proposed an accurate heart volume segmentation method with landmark-based registration (LMR) and 3D fully convolutional network (3D-FCN). First, we defined uniformity constrained landmarks in the training images and then trained a regression forest model (RFM) to detect these landmarks in the testing image. Second, the registration between the landmarks in each training image and the test image was performed by a shallow neural network, which guided the label propagation from atlases to the test image. After the label fusion with majority voting, we finally constructed a 3D-FCN to further refine the boundary voxels with low voting values. In 22 cardiac CTA images, we compared our method with multi-atlas segmentation, active shape model, DeepMedic, LMR + DeepMedic, and the separately implemented LMR and 3D-FCN. The results demonstrated the superiority of our method for the segmentation of heart volumes, with the average accuracy, dice coefficient, and mean Hausdorff distance as 96.25%, 93.98%, and 2.12 voxels, respectively. Furthermore, the training efficiency of the constructed 3D-FCN was higher than that of the DeepMedic through the comparison of training time and convergence of loss. The proposed heart segmentation could provide an accurate region of interest (ROI) for not only the four heart chambers but also the major trunks of vessels and coronary arteries.

Due to the tortuosity and the complexity of cerebral vasculature and the similar intensity distribution with the background, it remains challenging to accurately segment cerebral vessels from magnetic resonance angiography (MRA). The previous rule-based methods have limitations when applied to accurate clinical diagnosis, such as the under-segmentation on complex vessels, the dependence on domain knowledge, and the lack of quantification estimation. In this paper, we proposed a semi-supervised cerebrovascular segmentation method with a hierarchical convolutional neural network (H-CNN) that transfers the exquisite model/feature design in rule-based methods to solve the mapping from MRA images to cerebral vessels. First, we generated the tube-level labels of cerebral vessels with centerlines and estimated radii. Second, we constructed and trained an H-CNN with the MRA images and corresponding tube-level labels. Third, the stopping criterion of the proposed H-CNN was determined by the comprehensive index (CI) that was defined based on partially annotated voxel-level ground truth. The comparison of our H-CNN with the vesselness, bi-Gaussian, optimally oriented flux, vessel enhancing diffusion, hybrid diffusion with continuous switch, Mimics software, convolutional neutral network2D (CNN)2D, and CNN3D were conducted on six testing images. The mean sensitivity, accuracy, and the CI of our H-CNN are 94.69%, 97.85%, and 2.99%, respectively, outperforming the other methods. The curved planar reformation also visualized the performance of H-CNN for cerebrovascular segmentation. Given only the tube-level labels, the proposed H-CNN method accomplished the voxel-level vessel segmentation via the hierarchical update of CNN. The H-CNN can potentially to be applied for the accurate diagnosis of cerebrovascular diseases and other medical image segmentation with only partially correct labels.

Micro-computed tomography (micro-CT) provides an in vivo high-resolution preclinical imaging for murine kidneys. However, due to the relatively low dosage of X-rays, accurate and efficient segmentation of murine kidneys in micro-CT imaging remains challenging. In this paper, we proposed an efficient kidney segmentation method in micro-CT images based on multi-atlas registration (MAR) and random forests (RFs). First, we constructed a probability map of kidneys by the MAR and obtained an initial shape estimation of kidneys. We acquired the transformations of MAR based on high-contrast organs and then mapped kidneys to the new micro-CT images. Second, we extracted multiple features (including intensity, texture, and context features) from the voxels with lower probabilities and fed these features to a RF classifier. The role of RF is to fine-tune kidney boundaries after MAR. Finally, combining the initial shape with high probabilities and the fine-tuning of RF, we obtained the final segmentation of kidneys. The experiments were conducted on datasets acquired by micro-CT imaging of mice with and without the administration of contrast agent (Dataset1 and Dataset2). The results demonstrated the proposed MAR-RF outperformed the level sets, statistical-atlas registration, active shape model, and other supervised learning methods, with the Dice coefficients of 0.9766 and 0.9255, and the mean surface distances of 1.25 and 0.98 mm on Dataset1 and Dataset2, respectively. The training and prediction time of our MAR-RF were only 37.04% and 17.68% of the compared method, respectively. The proposed method has great potential for applications in other segmentation tasks of computer-aided diagnosis.

Several dammed lakes caused by landslides and rock avalanches appeared after Wenchuan earthquake. Tangjiashan dammed lake was the largest one among all of them, which resulted in obvious threats to people around. In this paper, multitemporal high-resolution airborne synthetic aperture radar (SAR) images are used to map the waterline of Tangjiashan dammed lake and its change continuously, which is meaningful for flooding assessment and prediction. An approach combining different techniques is proposed for dammed lake segmentation. In this approach, subpatched intensity thresholding segmentation, morphological operators, and local Gaussian distribution fitting energy (LGDF)-based active contours model (ACM) are used in combination to get the final waterline. Flood-affected area is detected and its size is calculated for each segmental result. In addition, the variations of the flood-affected area are also obtained based on the difference of the above-mentioned segmental results. The proposed approach is tested on real SAR imagery acquired from three different dates in May, 2008 with high-temporal resolution in the same area containing Tangjiashan dammed lake. Experimental results demonstrate the effectiveness of the proposed approach. [ABSTRACT FROM PUBLISHER]

Nonnegative eigenvalue decomposition (NNED), which insists and guarantees that each decomposed scattering component corresponds to a physically realizable scatterer, is powerful for polarimetric synthetic aperture radar (SAR) images analysis. Previous NNED is mainly illustrated under the reflection symmetric condition. In this paper, the coherency matrix approach is derived to implement the NNED for the nonreflection symmetry scattering case. We explicitly show the diversifications of the decomposition results between NNED with and without reflection symmetry assumptions, and quantitatively analyze the differences between them using the E-SAR polarimetric data acquired over the Oberpfaffenhofen area in Germany. [ABSTRACT FROM PUBLISHER]

The purpose of this letter is to show that the azimuth scanning mode of a synthetic aperture radar can be applied to deriving ocean wave spectra and determining the wave propagation direction for the first time, which reveals its enormous potential and great future in ocean observation. The improved Doppler beam sharpening imaging algorithm is used to produce a sequence of individual subimages of ocean waves in the same scan region from different aspect angles with a high revisit rate. These subimages have an inherent property that they are formed at different discretely delayed times. Therefore, wave propagation direction can be determined from a pair of wave images in different scans. Several different methods are applied to the real airborne radar wave data, including the methods of scan sum (taking the standard Fourier spectrum of the scan-summed image), spectral sum, spectral phase shift, and cross-correlation function of subimages. The processing results demonstrate the effectiveness of the algorithms. [ABSTRACT FROM PUBLISHER]