技术领域technical field
本发明涉及一种基于卷积神经网络的多角度遥感影像提取森林高度的方法，属于深度学习以及林业领域，本发明具有较强的泛化性和可行性，可用于实现连续区域森林高度制图的研究中。The invention relates to a method for extracting forest height from multi-angle remote sensing images based on a convolutional neural network, belonging to the field of deep learning and forestry. researching.
背景技术Background technique
森林高度是体现森林垂直结构的重要特征，对碳循环研究至关重要，在森林生物量估算以及动态变化的研究中起着不可替代的作用。传统的森林高度调查采取抽样调查、人工测量的方式，费时费力且获取困难。遥感技术的出现极大地弥补了传统调查存在的不足，激光雷达技术作为新兴的技术，为准确测量森林垂直结构提供了可能。Forest height is an important feature reflecting the vertical structure of forests, which is crucial to the study of carbon cycle and plays an irreplaceable role in the estimation of forest biomass and the study of dynamic changes. The traditional forest height survey adopts the method of sampling survey and manual measurement, which is time-consuming, labor-intensive and difficult to obtain. The emergence of remote sensing technology has greatly compensated for the shortcomings of traditional surveys. As an emerging technology, lidar technology provides the possibility to accurately measure the vertical structure of forests.
基于遥感技术估测森林高度的研究主要分为3类：1，基于光学遥感数据，由于光学遥感信号没有穿透性，无法获取森林垂直结构信息，因此很少用来提取森林高度参数；2，基于激光雷达数据，激光雷达可以提供准确的森林垂直结构测量数据，有大量研究使用激光点云数据或波形数据提取森林高度；3，结合激光雷达数据与光学遥感数据，利用光学遥感数据连续性、容易获取的特点，克服激光雷达数据在估算连续区域的森林高度时存在的局限性，结合这两种数据反演森林高度是目前研究的热点。The research on estimating forest height based on remote sensing technology is mainly divided into three categories: 1. Based on optical remote sensing data, because optical remote sensing signals are not penetrable and cannot obtain forest vertical structure information, they are rarely used to extract forest height parameters; 2. Based on lidar data, lidar can provide accurate forest vertical structure measurement data. There are a lot of studies using laser point cloud data or waveform data to extract forest height; 3. Combine lidar data and optical remote sensing data, use optical remote sensing data continuity, It is easy to obtain and overcomes the limitations of lidar data in estimating forest heights in continuous areas. Combining these two types of data to invert forest heights is a current research hotspot.
近年来深度学习发展迅速，自主学习图像的高层次特征已经成为了可能。卷积神经网络作为深度学习中著名的模型，在图像处理领域表现优异，而遥感影像具有信息丰富、分辨率高、连续成像等特点，并能在一定程度上反射地物的一些信息，因此利用卷积神经网络能对图像高层次特征自主学习的能力，可以实现从遥感影像上提取一些反映树高信息的特征。目前利用卷积神经网路结合遥感影像进行森林高度预测的工作还非常少，因此本发明基于卷积神经网络原理，结合激光雷达树高数据与光学遥感影像设计了一种有效的森林高度预测模型，实现了预测大范围、连续区域的森林高度。Deep learning has developed rapidly in recent years, and it has become possible to autonomously learn high-level features of images. As a well-known model in deep learning, convolutional neural network performs well in the field of image processing, while remote sensing images have the characteristics of rich information, high resolution, continuous imaging, etc., and can reflect some information of ground objects to a certain extent. The ability of convolutional neural network to learn the high-level features of images independently can realize the extraction of some features reflecting tree height information from remote sensing images. At present, there is very little work on forest height prediction using convolutional neural network combined with remote sensing images. Therefore, the present invention designs an effective forest height prediction model based on the principle of convolutional neural network and combining lidar tree height data and optical remote sensing images. , to achieve the prediction of forest height in a large-scale, continuous area.
发明内容SUMMARY OF THE INVENTION
本发明目的在于克服激光雷达数据只能估测小范围、离散区域森林高度的局限性，为森林高度尺度外推研究提供一种新的、有效的方法。本发明使用的多角度遥感影像为资源三号遥感影像，包括全色正视、全色前视、全色后视。为实现上述目的，本发明采用如下的技术方案：The purpose of the invention is to overcome the limitation that the lidar data can only estimate the forest height in a small range and discrete area, and provide a new and effective method for the extrapolation research of the forest height scale. The multi-angle remote sensing image used in the present invention is ZY-3 remote sensing image, including panchromatic front view, panchromatic front view, and panchromatic rear view. For achieving the above object, the present invention adopts the following technical scheme:
一种基于卷积神经网络的多角度遥感影像森林高度提取方法，是使用激光雷达估算的森林高度和资源三号多角度遥感影像训练卷积神经网络模型实现预测大范围连续区域的森林高度，该设计方法依次包括以下步骤：A method for extracting forest height from multi-angle remote sensing images based on convolutional neural network is to use the forest height estimated by lidar and the multi-angle remote sensing image of No. The design method includes the following steps in sequence:
步骤1：对资源三号多角度遥感影像进行正射校正以及重采样；具体步骤包括：Step 1: Perform orthorectification and resampling on the multi-angle remote sensing image of ZY-3; the specific steps include:
步骤1.1：获取研究区域的数字高程模型(DEM)即ASTER GDEM的30m分辨率数据；Step 1.1: Obtain the digital elevation model (DEM) of the study area, that is, the 30m resolution data of ASTER GDEM;
步骤1.2：对获取的多幅DEM影像镶嵌拼接，生成一幅合成DEM影像；Step 1.2: mosaic and stitch the acquired DEM images to generate a synthetic DEM image;
步骤1.3：利用Arcgis软件以及合成DEM影像对资源三号多角度遥感影像进行正射校正；Step 1.3: Orthorectify the multi-angle remote sensing image of ZY-3 using ArcGIS software and synthetic DEM image;
步骤1.4：使用三次卷积插值法对多角度遥感影像进行重采样，使角度不同的影像重采样到相同的栅格分辨率，即栅格分辨率为L米；Step 1.4: Use the cubic convolution interpolation method to resample the multi-angle remote sensing images, so that the images with different angles are resampled to the same grid resolution, that is, the grid resolution is L meters;
步骤2：将位于该研究区域的基于激光雷达数据提取的森林高度数据以及对应光斑中心的经纬度坐标、光斑的唯一标识ID全部保存到同一个shapefile格式的文件中；Step 2: Save the forest height data extracted from the lidar data in the research area, the latitude and longitude coordinates of the corresponding spot center, and the unique ID of the spot in the same shapefile format;
步骤3：分别对角度不同的影像裁剪，裁剪出多幅固定大小的影像，制作训练样本集；具体步骤包括：Step 3: Crop images with different angles respectively, crop out multiple fixed-size images, and create a training sample set; the specific steps include:
步骤3.1：激光雷达光斑直径为H米与多角度影像的栅格分辨率为L米的商值作为要裁剪的影像大小，即影像大小为n×n个像元；Step 3.1: The quotient of the lidar spot diameter of H meters and the grid resolution of the multi-angle image of L meters is used as the size of the image to be cropped, that is, the size of the image is n×n pixels;
步骤3.2：读取shapefile文件，将光斑点的地理坐标转化为落在多角度遥感影像上的像元坐标，以光斑点所在的像元为中心分别裁剪不同角度的遥感影像，裁剪的影像大小为n×n个像元并以该光斑的ID命名，重复步骤3.2，直到所有落在影像上的光斑点均裁剪完毕；Step 3.2: Read the shapefile file, convert the geographic coordinates of the light spot into the pixel coordinates falling on the multi-angle remote sensing image, and cut the remote sensing images of different angles with the pixel where the light spot is located as the center. The size of the cropped image is n×n pixels are named after the ID of the spot, repeat step 3.2, until all spots falling on the image are cropped;
步骤3.3：调整角度不同的影像张数相同，即如果在不同角度影像中有同一命名ID，则保留，否则舍弃，最终得到不同角度的影像张数一致且命名一一对应；Step 3.3: Adjust the number of images with different angles to be the same, that is, if there is the same named ID in the images of different angles, keep it, otherwise discard it, and finally get the same number of images with different angles and the names correspond one-to-one;
步骤4：构造卷积神经网络模型，将上述步骤得到的样本分为训练样本和验证样本，训练样本用于训练模型，验证样本用于验证模型性能；具体步骤包括：Step 4: construct a convolutional neural network model, divide the samples obtained in the above steps into training samples and verification samples, the training samples are used for training the model, and the verification samples are used for verifying the performance of the model; the specific steps include:
步骤4.1：将命名相同角度不同的影像叠加处理，生成新的样本集；Step 4.1: Superimpose images with the same name and different angles to generate a new sample set;
步骤4.2：将新样本集进行分配，约2/3样本用于训练，剩下的1/3用于验证；Step 4.2: Allocate the new sample set, about 2/3 of the samples are used for training, and the remaining 1/3 is used for verification;
步骤4.3：构造卷积神经网络模型，训练样本与对应光斑ID的森林高度数据构成数据对作为网络的输入，输出为预测的森林高度，训练网络并保存模型；Step 4.3: Construct a convolutional neural network model, the training sample and the forest height data corresponding to the spot ID form a data pair as the input of the network, the output is the predicted forest height, train the network and save the model;
步骤4.4：提取步骤4.3保存的模型，验证样本与对应光斑ID的森林高度数据构成数据对作为模型的输入，评估模型性能，如果模型性能未达到预期目标，则调整网络参数或结构，转到步骤4.3；Step 4.4: Extract the model saved in Step 4.3, verify that the sample and the forest height data corresponding to the spot ID form a data pair as the input of the model, evaluate the model performance, if the model performance does not reach the expected target, adjust the network parameters or structure, go to step 4.3;
步骤5：采取滑动裁剪的方式连续无缝地裁剪多角度遥感影像，提取保存的模型，制作研究区域的栅格分辨率为H米的森林高度分布图；具体步骤包括：Step 5: Continuously and seamlessly crop the multi-angle remote sensing images by sliding cropping, extract the saved model, and make a forest height distribution map with a grid resolution of H meters in the study area; the specific steps include:
步骤5.1：对步骤1.5得到的具有相同栅格分辨率的多角度遥感影像分别截取重合区域影像；Step 5.1: intercept the overlapping area images from the multi-angle remote sensing images with the same grid resolution obtained in step 1.5;
步骤5.2：分别对不同角度的重合区域影像采取无缝滑动裁剪的方式裁剪出多幅n×n个像元大小的影像；Step 5.2: Cut out multiple images of n×n pixel size by means of seamless sliding cropping for the overlapping area images of different angles respectively;
步骤5.3：将命名相同角度不同的影像叠加处理，生成新的样本集；Step 5.3: Superimpose images with the same name and different angles to generate a new sample set;
步骤5.4：提取步骤4.4最终保存的模型，输入新的样本集，输出森林高度数据；Step 5.4: Extract the model finally saved in Step 4.4, input a new sample set, and output forest height data;
步骤5.5：生成与重合区域形状、大小、位置相同且分辨率为H米的栅格影像，森林高度数据作为栅格属性逐一对应添加进去；Step 5.5: Generate a raster image with the same shape, size and position as the overlapping area and a resolution of H meters, and the forest height data is added as a raster attribute correspondingly one by one;
步骤5.6：根据栅格森林高度属性值分类显示不同颜色，生成研究区域的栅格分辨率为H米的森林高度分布图。Step 5.6: classify and display different colors according to the attribute value of the raster forest height, and generate a forest height distribution map with a grid resolution of H meters in the study area.
附图说明Description of drawings
图1为本发明的基本方法流程示意图；1 is a schematic flow chart of a basic method of the present invention;
图2为实例的全色前视角度的遥感影像图；Fig. 2 is the remote sensing image diagram of the panchromatic front view angle of example;
图3为实例的卷积神经网络模型结构图；Fig. 3 is the convolutional neural network model structure diagram of example;
图4为实例的最终生成的森林高度分布结果图。Fig. 4 is the final generated forest height distribution result graph of the example.
具体实施方式Detailed ways
本发明实例提供一种基于卷积神经网络的多角度遥感影像森林高度提取方法，下面结合相关附图对本发明进行解释和阐述。An example of the present invention provides a method for extracting forest height from a multi-angle remote sensing image based on a convolutional neural network. The present invention will be explained and described below with reference to the relevant drawings.
本发明实例使用的数据集为2017年某地区的资源三号多角度遥感影像，包括全色正视、全色前视以及全色后视，选择TensorFlow深度学习框架构造卷积神经网络模型，训练模型用于生成该研究区域的森林高度分布图。本发明实例的具体实施方案为：The data set used in the example of the present invention is the multi-angle remote sensing image of Resource No. 3 in a certain area in 2017, including panchromatic front view, panchromatic front view, and panchromatic rear view. The TensorFlow deep learning framework is selected to construct a convolutional neural network model and train the model. The forest height distribution map used to generate the study area. The specific embodiment of the example of the present invention is:
步骤1：对资源三号多角度遥感影像进行正射校正以及重采样；具体步骤包括：Step 1: Perform orthorectification and resampling on the multi-angle remote sensing image of ZY-3; the specific steps include:
步骤1.1：获取研究区域的数字高程模型(DEM)即ASTER GDEM的30m分辨率数据；Step 1.1: Obtain the digital elevation model (DEM) of the study area, that is, the 30m resolution data of ASTER GDEM;
步骤1.2：对获取的4幅DEM影像镶嵌拼接，生成一幅合成DEM影像；Step 1.2: Mosaic and stitch the acquired 4 DEM images to generate a synthetic DEM image;
步骤1.3：利用Arcgis软件以及合成DEM影像对资源三号多角度遥感影像进行正射校正；Step 1.3: Orthorectify the multi-angle remote sensing image of ZY-3 using ArcGIS software and synthetic DEM image;
步骤1.4：使用三次卷积插值法对多角度遥感影像进行重采样，使角度不同的影像重采样到相同的栅格分辨率，即栅格分辨率为2.3米，全色前视的遥感影像图如图2所示；Step 1.4: Use the cubic convolution interpolation method to resample the multi-angle remote sensing images, so that the images with different angles are resampled to the same grid resolution, that is, the grid resolution is 2.3 meters, and the full-color front-view remote sensing image map as shown in picture 2;
步骤2：将位于该研究区域的基于激光雷达数据提取的森林高度数据以及对应光斑中心的经纬度坐标、光斑的唯一标识ID全部保存到同一个shapefile格式的文件中；Step 2: Save the forest height data extracted from the lidar data in the research area, the latitude and longitude coordinates of the corresponding spot center, and the unique ID of the spot in the same shapefile format;
步骤3：分别对角度不同的影像裁剪，裁剪出多幅固定大小的影像，制作训练样本集；具体步骤包括：Step 3: Crop images with different angles respectively, crop out multiple fixed-size images, and create a training sample set; the specific steps include:
步骤3.1：激光雷达光斑直径为30米与多角度影像的栅格分辨率为2.3米的商值作为要裁剪的影像大小，即影像大小为13×13个像元；Step 3.1: The quotient of the lidar spot diameter of 30 meters and the grid resolution of the multi-angle image of 2.3 meters is used as the size of the image to be cropped, that is, the image size is 13×13 pixels;
步骤3.2：读取shapefile文件，将光斑点的地理坐标转化为落在多角度遥感影像上的像元坐标，以光斑点所在的像元为中心分别裁剪不同角度的遥感影像，裁剪的影像大小为13×13个像元并以该光斑的ID命名，重复步骤3.2，直到所有落在影像上的光斑点均裁剪完毕；Step 3.2: Read the shapefile file, convert the geographic coordinates of the light spot into the pixel coordinates falling on the multi-angle remote sensing image, and cut the remote sensing images of different angles with the pixel where the light spot is located as the center. The size of the cropped image is 13×13 pixels are named after the ID of the spot, and repeat step 3.2 until all spots falling on the image are cropped;
步骤3.3：调整角度不同的影像张数相同，即如果在不同角度影像中有同一命名ID，则保留，否则舍弃，最终得到不同角度的影像张数一致且命名一一对应；Step 3.3: Adjust the number of images with different angles to be the same, that is, if there is the same named ID in the images of different angles, keep it, otherwise discard it, and finally get the same number of images with different angles and the names correspond one-to-one;
步骤4：构造卷积神经网络模型，将上述步骤得到的样本分为训练样本和验证样本，训练样本用于训练模型，验证样本用于验证模型性能；具体步骤包括：Step 4: construct a convolutional neural network model, divide the samples obtained in the above steps into training samples and verification samples, the training samples are used for training the model, and the verification samples are used for verifying the performance of the model; the specific steps include:
步骤4.1：将命名相同角度不同的影像叠加处理，生成新的样本集；Step 4.1: Superimpose images with the same name and different angles to generate a new sample set;
步骤4.2：将新样本集进行分配，约2/3样本用于训练，剩下的1/3用于验证；Step 4.2: Allocate the new sample set, about 2/3 of the samples are used for training, and the remaining 1/3 is used for verification;
步骤4.3：构造卷积神经网络模型，训练样本与对应光斑ID的森林高度数据构成数据对作为网络的输入，输出为预测的森林高度，训练网络并保存模型；Step 4.3: Construct a convolutional neural network model, the training sample and the forest height data corresponding to the spot ID form a data pair as the input of the network, the output is the predicted forest height, train the network and save the model;
步骤4.4：提取步骤4.3保存的模型，验证样本与对应光斑ID的森林高度数据构成数据对作为模型的输入，评估模型性能，如果模型性能未达到预期目标，则调整网络参数或结构，转到步骤4.3，最终的卷积神经网络结构图如图3所示；Step 4.4: Extract the model saved in Step 4.3, verify that the sample and the forest height data corresponding to the spot ID form a data pair as the input of the model, evaluate the model performance, if the model performance does not reach the expected target, adjust the network parameters or structure, go to step 4.3, the final convolutional neural network structure diagram is shown in Figure 3;
步骤5：采取滑动裁剪的方式连续无缝地裁剪多角度遥感影像，提取保存的模型，制作研究区域的栅格分辨率为30米的森林高度分布图；具体步骤包括：Step 5: Continuously and seamlessly crop multi-angle remote sensing images by means of sliding cropping, extract the saved model, and make a forest height distribution map with a grid resolution of 30 meters in the study area; the specific steps include:
步骤5.1：对步骤1.5得到的具有相同栅格分辨率的多角度遥感影像分别截取重合区域影像；Step 5.1: intercept the overlapping area images from the multi-angle remote sensing images with the same grid resolution obtained in step 1.5;
步骤5.2：分别对不同角度的重合区域影像采取无缝滑动裁剪的方式裁剪出多幅13×13个像元大小的影像；Step 5.2: Cut out multiple 13×13 pixel images by seamless sliding cropping for the overlapping area images of different angles;
步骤5.3：将命名相同角度不同的影像叠加处理，生成新的样本集；Step 5.3: Superimpose images with the same name and different angles to generate a new sample set;
步骤5.4：提取步骤4.4最终保存的模型，输入新的样本集，输出森林高度数据；Step 5.4: Extract the model finally saved in Step 4.4, input a new sample set, and output forest height data;
步骤5.5：生成与重合区域形状、大小、位置相同且分辨率为30米的栅格影像，森林高度数据作为栅格属性逐一对应添加进去；Step 5.5: Generate a raster image with the same shape, size and position as the overlapping area and a resolution of 30 meters, and the forest height data is added as a raster attribute correspondingly one by one;
步骤5.6：根据栅格森林高度属性值分类显示不同颜色，生成研究区域的栅格分辨率为30米的森林高度分布图，如图4所示。Step 5.6: Display different colors according to the raster forest height attribute value, and generate a forest height distribution map with a grid resolution of 30 meters in the study area, as shown in Figure 4.
以上实例仅用于描述本发明，而非限制本发明所描述的技术方案。因此，一切不脱离本发明精神和范围的技术方案及其改进，均应涵盖在本发明的权利要求范围中。The above examples are only used to describe the present invention, but do not limit the technical solutions described in the present invention. Therefore, all technical solutions and improvements that do not depart from the spirit and scope of the present invention should be covered by the scope of the claims of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910336776.3ACN110070488B (en) | 2019-04-25 | 2019-04-25 | Multi-angle remote sensing image forest height extraction method based on convolutional neural network |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910336776.3ACN110070488B (en) | 2019-04-25 | 2019-04-25 | Multi-angle remote sensing image forest height extraction method based on convolutional neural network |
| Publication Number | Publication Date |
|---|---|
| CN110070488Atrue CN110070488A (en) | 2019-07-30 |
| CN110070488B CN110070488B (en) | 2023-01-03 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910336776.3AActiveCN110070488B (en) | 2019-04-25 | 2019-04-25 | Multi-angle remote sensing image forest height extraction method based on convolutional neural network |
| Country | Link |
|---|---|
| CN (1) | CN110070488B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113673596A (en)* | 2021-08-20 | 2021-11-19 | 自然资源部国土卫星遥感应用中心 | Remote sensing image target detection sample generation method based on traversal source target |
| CN113920438A (en)* | 2021-12-14 | 2022-01-11 | 武汉大学 | Troubleshooting method for hidden dangers of trees near power lines combined with ICESat-2 and Jilin-1 images |
| CN114037911A (en)* | 2022-01-06 | 2022-02-11 | 武汉大学 | A large-scale remote sensing inversion method of forest height considering ecological zoning |
| CN114972989A (en)* | 2022-05-18 | 2022-08-30 | 中国矿业大学(北京) | Single remote sensing image height information estimation method based on deep learning algorithm |
| CN115100630A (en)* | 2022-07-04 | 2022-09-23 | 小米汽车科技有限公司 | Obstacle detection method, obstacle detection device, vehicle, medium, and chip |
| CN117435848A (en)* | 2023-12-06 | 2024-01-23 | 天津师范大学 | Large-scale forest height inversion method and system based on satellite multi-angle index |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070291994A1 (en)* | 2002-05-03 | 2007-12-20 | Imagetree Corp. | Remote sensing and probabilistic sampling based forest inventory method |
| CN103760565A (en)* | 2014-02-10 | 2014-04-30 | 中国科学院南京地理与湖泊研究所 | Regional scale forest canopy height remote sensing retrieval method |
| CN105866792A (en)* | 2016-05-31 | 2016-08-17 | 中国科学院遥感与数字地球研究所 | Novel satellite-borne laser radar tree height extraction method |
| CN108038448A (en)* | 2017-12-13 | 2018-05-15 | 河南理工大学 | Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070291994A1 (en)* | 2002-05-03 | 2007-12-20 | Imagetree Corp. | Remote sensing and probabilistic sampling based forest inventory method |
| CN103760565A (en)* | 2014-02-10 | 2014-04-30 | 中国科学院南京地理与湖泊研究所 | Regional scale forest canopy height remote sensing retrieval method |
| CN105866792A (en)* | 2016-05-31 | 2016-08-17 | 中国科学院遥感与数字地球研究所 | Novel satellite-borne laser radar tree height extraction method |
| CN108038448A (en)* | 2017-12-13 | 2018-05-15 | 河南理工大学 | Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113673596A (en)* | 2021-08-20 | 2021-11-19 | 自然资源部国土卫星遥感应用中心 | Remote sensing image target detection sample generation method based on traversal source target |
| CN113920438A (en)* | 2021-12-14 | 2022-01-11 | 武汉大学 | Troubleshooting method for hidden dangers of trees near power lines combined with ICESat-2 and Jilin-1 images |
| CN113920438B (en)* | 2021-12-14 | 2022-03-04 | 武汉大学 | Troubleshooting method for hidden dangers of trees near power lines combined with ICESat-2 and Jilin-1 images |
| CN114037911A (en)* | 2022-01-06 | 2022-02-11 | 武汉大学 | A large-scale remote sensing inversion method of forest height considering ecological zoning |
| CN114037911B (en)* | 2022-01-06 | 2022-04-15 | 武汉大学 | Large-scale forest height remote sensing inversion method considering ecological zoning |
| CN114972989A (en)* | 2022-05-18 | 2022-08-30 | 中国矿业大学(北京) | Single remote sensing image height information estimation method based on deep learning algorithm |
| CN115100630A (en)* | 2022-07-04 | 2022-09-23 | 小米汽车科技有限公司 | Obstacle detection method, obstacle detection device, vehicle, medium, and chip |
| CN117435848A (en)* | 2023-12-06 | 2024-01-23 | 天津师范大学 | Large-scale forest height inversion method and system based on satellite multi-angle index |
| CN117435848B (en)* | 2023-12-06 | 2024-03-12 | 天津师范大学 | Satellite multi-angle index-based large-scale forest height inversion method and system |
| Publication number | Publication date |
|---|---|
| CN110070488B (en) | 2023-01-03 |
| Publication | Publication Date | Title |
|---|---|---|
| CN110070488B (en) | Multi-angle remote sensing image forest height extraction method based on convolutional neural network | |
| CN101777189B (en) | Method for measuring image and inspecting quantity under light detection and ranging (LiDAR) three-dimensional environment | |
| Balsa-Barreiro et al. | Generation of visually aesthetic and detailed 3D models of historical cities by using laser scanning and digital photogrammetry | |
| US10297074B2 (en) | Three-dimensional modeling from optical capture | |
| CN103729883B (en) | A kind of three-dimensional environment information gathering and reconfiguration system and method | |
| CN103606188B (en) | Geography information based on imaging point cloud acquisition method as required | |
| CN109872401B (en) | Unmanned aerial vehicle video augmented reality implementation method | |
| CN112926468B (en) | Tidal flat elevation automatic extraction method | |
| CN118691776B (en) | A 3D real scene modeling and dynamic updating method based on multi-source data fusion | |
| CN103077559B (en) | Based on the fruit ear three-dimensional rebuilding method of sequence image | |
| CN112819066A (en) | Res-UNet single tree species classification technology | |
| CN109801371B (en) | A Cesium-based Network 3D Electronic Map Construction Method | |
| CN102298793A (en) | Method for obtaining three-dimensional figure data of cultural relic | |
| CN106097433A (en) | Object industry and the stacking method of Image model and system | |
| CN115861527A (en) | Method and device for constructing live-action three-dimensional model, electronic equipment and storage medium | |
| CN108388995A (en) | A kind of method for building up of road asset management system and establish system | |
| CN103196429B (en) | Method for quickly obtaining and measuring orthophotoquad of city skyline contour line facade | |
| CN111667569A (en) | An accurate measurement method of 3D real earthwork visualization based on Rhino and Grasshopper | |
| CN103335606A (en) | Three dimensional laser scanning and surveying and mapping method | |
| CN108871235A (en) | The information acquisition method and plant leaf information acquisition system of plant leaf | |
| CN103245331B (en) | Control method of large-scale aerial survey indoor and outdoor integrated system | |
| CN112509133A (en) | Three-dimensional reservoir high-definition live-action display method based on GIS | |
| CN104573239A (en) | High spatial resolution remote sense image-based tidal flat DEM (Digital Elevation Model) optimization method | |
| Dhruwa et al. | Generation of 3-D Large-Scale Maps Using Lidar Point Cloud Data | |
| CN117036511B (en) | Calibration methods, devices, computer equipment and storage media for multiple types of sensors |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |