The laser altimeter loaded on the GaoFen-7(GF-7) satellite is designed to record the full waveform data and footprint image, which can obtain high-precision elevation control points for stereo image. The footprint camera equipped on the GF-7 laser altimetry system can capture the energy distribution at the time of laser emission and the image of the ground object where the laser falls, which can be used to judge whether the laser is affected by the cloud. At the same time, the centroid of laser spot on the footprint image can be extracted to monitor the change of laser pointing stability. In this manuscript, a data quality analysis scheme of laser altimetry based on footprint image is presented. Firstly, the cloud detection of footprint image is realized based on deep learning. The fusion result of the model is about 5% better than that of the traditional cloud detection algorithm, which can quickly and accurately determine whether the laser spot is affected by cloud. Secondly, according to the characteristics of footprint image, a threshold constrained ellipse fitting method for extracting the centroid of laser spot is proposed to monitor the pointing stability of long-period lasers. Based on the above method, the change of laser spot centroid since GF-7 satellite was put into operation is analyzed, and the conclusions obtained have certain reference significance for the quality control of satellite laser altimetry data and the analysis of pointing angle stability.

Reliable and prompt information on forest above-ground biomass (AGB) and tree diameter at breast height (DBH)are crucial for sustainable forest management. Remote sensing technology, especially the Light Detection and Ranging (LiDAR) technology, has been proven to estimate important tree variables effectively. This study proposes predicting DBH and AGB from tree height and other LiDAR data extracted metrics. In the suggested DBH prediction, we developed a nonlinear estimation equation using the total tree height. As for the AGB prediction approach, we used regression methods such as multiple linear regression (MLR), random forest (RF) and support vector machine for regression (SVR). We conducted the study for the Gudao forest area dominated by Robinia Pseudoacacia trees, located in the Yellow River Delta (YRD), China. For our developed approaches, we used Unmanned Aerial Vehicle (UAV) and Backpack LiDAR point cloud datasets obtained in June 2017, and three field data measurements gathered in June 2017 and 2019 and October 2019, all from the same study area. The results demonstrate that: ① The LiDAR data individual tree segmentation (ITS) from which we extracted individual tree information like tree location and tree height, was carried out with an overall accuracy F=0.91; ② We used the ITS height data from the field stand in 2019 as a fit and developed a nonlinear DBH estimation equation with Root Mean Square Error (RMSE)=3.61cm, later validated by the 2017 dataset; ③ Forest AGB at stand level was estimated with the MLR, RF and also SVR regression methods, and results show that the SVR method gave higher accuracy with R²=0.82 compared to the R²=0.72 of RF and the R²=0.70 of the MLR. Calculated AGB at plot level using the 2017 LiDAR data was used to validate both models’ accuracy. Combining the UAV LiDAR data and the Backpack LiDAR significantly improved the overall ITS. The UAV LiDAR ability to provide high accuracy tree height abstraction, the DBH of the regression equation and other extracted LiDAR metrics showed high accuracy in estimating the forest AGB. This study shows that being cost-free is not the only advantage of free available software. In the performance of ITS and the LiDAR, metrics extraction proved to be as good as the commercially available software.

A fast local reference frame (LRF) construction method based on the signed surface variation is proposed, which can adapt to the real-time applications such as self-driving, face recognition, object detection. The z-axis of the LRF is generated based on the concavity of the local surface of keypoint. The x-axis is constructed by the weighted vector sum of a set of projection vectors of the local neighborhoods around keypoint. The performance of the proposed LRF is evaluated on six standard datasets and compared with six state-of-the-art LRF construction methods (e.g., BOARD, FLARE, SHOT, RoPS and TOLDI). Experimental results validate the high repeatability, robustness, universality and time efficiency of our method.

A method for land-cover classification was proposed based on the fusion of features generated from waveform data and point cloud respectively. It aims to partially overcome the ineffectiveness of many traditional classifiers caused by the fact that point cloud is lacking spectral information. The whole flowchart of the method is as follows: Firstly, Gaussian decomposition was applied to fit an echo full-waveform. The parameters associated with the Gaussian function were optimized by LM (Levenberg-Marquard) algorithm. Six and thirteen features were generated to describe the waveform characteristics and the local geometry of point cloud, respectively. Secondly, a random forest was selected as the classifier to which the generated features were input. Relief-F was used to rank the weights of all the features generated. Finally, features were input to the classifier one by one according to the weights calculated from feature ranking, where classification accuracies were evaluated. The experimental results show that the effectiveness of the fusion of features generated from waveform and point cloud for LiDAR data classification, with 95.4% overall accuracy, 0.90 kappa coefficient, which outperform the results obtained by a single class of features, no matter whether they were generated from point cloud or waveform data.

Automatic and accurate classification is a fundamental problem to the analysis and modeling of LiDAR (Light Detection and Ranging) data. Recently, convolutional neural network (ConvNet or CNN) has achieved remarkable performance in image recognition and computer vision. While significant efforts have also been made to develop various deep networks for satellite image scene classification, it still needs to further investigate suitable deep learning network frameworks for 3D dense mobile laser scanning (MLS) data. In this paper, we present a simple deep CNN for multiple object classification based on multi-scale context representation. For the pointwise classification, we first extracted the neighboring points within spatial context and transformed them into a three-channel image for each point. Then, the classification task can be treated as the image recognition using CNN. The proposed CNN architecture adopted common convolution, maximum pooling and rectified linear unit (ReLU) layers, which combined multiple deeper network layers. After being trained and tested on approximately seven million labeled MLS points, the deep CNN model can classify accurately into nine classes. Comparing with the widely used ResNet algorithm, this model performs better precision and recall rates, and less processing time, which indicated the significant potential of deep-learning-based methods in MLS data classification.

Airborne Light Detection And Ranging (LiDAR) can provide high-quality three-dimensional information for the safety inspection of electricity corridors. However, the robust extraction of transmission lines from airborne point cloud data is still greatly challenging. Therefore, this paper proposes a robust transmission line extraction method based on model fitting from airborne point cloud data. First, the candidate power line generation method based on height information is used to reduce the computational complexity at the subsequent steps and the false positives in the extracted results. Then, on the basis of the block-and-slice-constraint Euclidean clustering, a linear structure recognition method based on RANdom SAmple Consensus (RANSAC) is proposed to produce the initial individual transmission line components. Finally, a robust nonlinear least square-based fitting method is developed for the individual transmission line to generate the parameters of its mathematical model for further optimizing the extraction. Experiments were performed on LiDAR point cloud data captured from the helicopter and Unmanned Aerial Vehicle (UAV) platform. Results indicate that the proposed method can efficiently extract the different types of transmission lines along electricity corridors, with the average precision of approximately 98.1%, the average recall of approximately 95.9%, and the average quality of approximately 94.2%, respectively.

Point cloud registration is a fundamental task in both remote sensing, photogrammetry, and computer vision, which is to align multiple point clouds to the same coordinate frame. Especially in LiDAR odometry, by conducting the transformation between two adjacent scans, the pose of the platform can be estimated. To be specific, the goal is to recover the relative six-degree-of-freedom (6 DoF) pose between the source point cloud and the target point cloud. In this paper, we explore the use of robust estimators in the phase correlation when registering two point clouds, enabling a 6 DoF pose estimation between point clouds in a sub-voxel accuracy. The estimator is a rule for calculating an estimate of a given quantity based on observed data. A robust estimator is an estimation rule that is insensitive to nonnormality and can estimate parameters of a given objective function from noisy observations. The proposed registration method is theoretically insensitive to noise and outliers than correspondence-based methods. Three core steps are involved in the method: transforming point clouds from the spatial domain to the frequency domain, decoupling of rotations and translations, and using robust estimators to estimate phase shifts. Since the estimation of transformation parameters lies in the calculation of phase shifts, robust estimators play a vital role in shift estimation accuracy. In this paper, we have tested the performance of six different robust estimators and provide comparisons and discussions on the contributions of robust estimators in the 3D phase correlation. Different point clouds from two urban scenarios and one indoor scene are tested. Results validate the proposed method can reach performance that predominant rotation and translation errors reaching less than 0.5° and 0.5m, respectively. Moreover, the performance of various tested robust estimators is compared and discussed.

ZiYuan3-03 (ZY3-03) satellite was launched on July 25, 2020, equipped with China’s second-generation laser altimeter for earth observation. In order to preliminarily evaluate the in-orbit performance of the ZY3-03 laser altimeter, the pointing bias calibration based on terrain matching method was adopted. Three tracks of laser data were employed for the ZY3-03 laser altimeter calibration test. Three groups of pointing parameters were obtained respectively, and the mean value of pointing is considered as the optimal calibration result. After calibration, ZY3-03 laser pointing accuracy is greatly improved by the method, and its pointing accuracy is approximately 12.7 arcsec. The first-track laser data on the Black Sea surface is used to evaluate the relative elevation accuracy of ZY3-03 laser altimeter after pointing bias calibration, which is improved from 0.33m to 0.19m after calibration. Meanwhile, the absolute elevation accuracy of ZY3-03 laser altimeter after pointing bias calibration is evaluated by the Ground Control Points (GCPs) measured by RTK (Real-Time Kinematic), which is better than 0.5m in the flat terrain.