9:15 AM - 9:30 AM
[STT39-01] Utilization of AI image analysis and multispectral data in emergency surveys of sediment-related disasters using UAVs
Keywords:Unmanned aerial vehicle, Emergency Survey for Sediment-related Disasters, Image analysis by artificial intelligence, Multispectral image, Extraction of disaster locations and situations, SfM-MVS Photogrammetry
1. Introduction
Examination of upgrading of emergency survey methods using UAV is being promoted as a support technique for sediment-related disaster emergency surveys. AI imaging analysis is used to quickly recognize the location and extent of the disaster. And, the utilization of near infrared multispectral data is examined in order to add physical property information to topographic information. As a way to proceed with the examination, starting from a condition that is relatively inexpensive, elemental technologies shall be added according to the purpose and requirements.
2. Utilization of AI Image Analysis in Determining the Location of the Disaster
The utilization of AI image analysis for UAV photographing image was examined. We assumed that cloud services could not be used in the event of a disaster, and decided to construct an edge-type system that does not require a highly functional server. Semantic segmentation is adopted for the algorithm of AI engine. The interface can be operated without expertise on AI.
3. Extracting the location and status of the disaster from UAV movies and still images
We created a support tool to automatically extract disaster information from UAV video. The operation procedure of the automatic extraction tool is shown below. 1)Select the still image to be used for AI image analysis from the moving image. 2)AI image analysis of the selected still image. 3)Automatic classification of image files. 4)Sweep out classified image file list. 5)Location information is associated with the extraction list.
Since the still image extracted from the moving image has the limitation of the practical resolution, the utilization of the still image by SfM or the shooting method which conforms to this was examined in expectation of the improvement in the consequence accuracy of AI image analysis.
4. AI image analysis of orthographic images
AI image analysis for orthographic images was examined. The whole orthographic image is divided into appropriate regions, and AI image analysis is carried out for each divided region. Integrate each region after analysis processing. When the result was verified, the collapsed place and moving soil mass with the spatial extent were able to be excellently extracted. On the other hand, discrimination of microtopography and fine cracks and fractures is not easy. Variations of data to sufficiently learn features of deformation and appropriate setting of annotation are problems.
5. Point cloud data by SfM analysis and LiDAR
With only DSM obtained from SfM-MVS, there are limitations in acquiring informations such as deformed structures, fractures, cracks, etc. peculiar to slope failure and landslide areas. Therefore, this paper compares point cloud data by LiDAR with point cloud data by.SfM examining the combined use of point cloud data by LiDAR, and in addition, evaluates these connectivity. As a result, it was confirmed that both types of point cloud data with high accuracy can be obtained, if sufficient attention is paid to the acquisition of position information, and that there is no special problem in the connection of those data.
6. Utilization of multispectral data
Multispectral data can be used to identify the composition of soil and rock, separate vegetation areas, and evaluate soil moisture (volumetric moisture content). Together with direct information of displacement and deformation of topography, these information become important information of occurrence prediction and expansion grasp of slope failure, landslide, debris flow, river channel blockage, etc.. The feature of the multispectral camera system used in the present examination is that fixed point shoot is possible. Therefore, it is possible to take pictures in complex terrain such as mountainous areas. And, the spatial position of the photographing object position can be linked based on the position informations of UAV.
7. Issues and Development
Assuming an emergency-survey in the event of a disaster, we investigated how to effectively use UAV imaging. As a result, a certain target was achieved. On the other hand, in DSM and two-dimensional image-data analysis methods, there is a limitation in recognition accuracy when microtopography analysis is carried out. In the future, the examination of the analysis and utilization method which combined LiDAR datum will be advanced. And, it is necessary to examine the linkage of physical property information such as multispectral data and resistivity with spatial position information.
Examination of upgrading of emergency survey methods using UAV is being promoted as a support technique for sediment-related disaster emergency surveys. AI imaging analysis is used to quickly recognize the location and extent of the disaster. And, the utilization of near infrared multispectral data is examined in order to add physical property information to topographic information. As a way to proceed with the examination, starting from a condition that is relatively inexpensive, elemental technologies shall be added according to the purpose and requirements.
2. Utilization of AI Image Analysis in Determining the Location of the Disaster
The utilization of AI image analysis for UAV photographing image was examined. We assumed that cloud services could not be used in the event of a disaster, and decided to construct an edge-type system that does not require a highly functional server. Semantic segmentation is adopted for the algorithm of AI engine. The interface can be operated without expertise on AI.
3. Extracting the location and status of the disaster from UAV movies and still images
We created a support tool to automatically extract disaster information from UAV video. The operation procedure of the automatic extraction tool is shown below. 1)Select the still image to be used for AI image analysis from the moving image. 2)AI image analysis of the selected still image. 3)Automatic classification of image files. 4)Sweep out classified image file list. 5)Location information is associated with the extraction list.
Since the still image extracted from the moving image has the limitation of the practical resolution, the utilization of the still image by SfM or the shooting method which conforms to this was examined in expectation of the improvement in the consequence accuracy of AI image analysis.
4. AI image analysis of orthographic images
AI image analysis for orthographic images was examined. The whole orthographic image is divided into appropriate regions, and AI image analysis is carried out for each divided region. Integrate each region after analysis processing. When the result was verified, the collapsed place and moving soil mass with the spatial extent were able to be excellently extracted. On the other hand, discrimination of microtopography and fine cracks and fractures is not easy. Variations of data to sufficiently learn features of deformation and appropriate setting of annotation are problems.
5. Point cloud data by SfM analysis and LiDAR
With only DSM obtained from SfM-MVS, there are limitations in acquiring informations such as deformed structures, fractures, cracks, etc. peculiar to slope failure and landslide areas. Therefore, this paper compares point cloud data by LiDAR with point cloud data by.SfM examining the combined use of point cloud data by LiDAR, and in addition, evaluates these connectivity. As a result, it was confirmed that both types of point cloud data with high accuracy can be obtained, if sufficient attention is paid to the acquisition of position information, and that there is no special problem in the connection of those data.
6. Utilization of multispectral data
Multispectral data can be used to identify the composition of soil and rock, separate vegetation areas, and evaluate soil moisture (volumetric moisture content). Together with direct information of displacement and deformation of topography, these information become important information of occurrence prediction and expansion grasp of slope failure, landslide, debris flow, river channel blockage, etc.. The feature of the multispectral camera system used in the present examination is that fixed point shoot is possible. Therefore, it is possible to take pictures in complex terrain such as mountainous areas. And, the spatial position of the photographing object position can be linked based on the position informations of UAV.
7. Issues and Development
Assuming an emergency-survey in the event of a disaster, we investigated how to effectively use UAV imaging. As a result, a certain target was achieved. On the other hand, in DSM and two-dimensional image-data analysis methods, there is a limitation in recognition accuracy when microtopography analysis is carried out. In the future, the examination of the analysis and utilization method which combined LiDAR datum will be advanced. And, it is necessary to examine the linkage of physical property information such as multispectral data and resistivity with spatial position information.