CN113591668B - Wide area unknown dam automatic detection method using deep learning and space analysis - Google Patents
Wide area unknown dam automatic detection method using deep learning and space analysis Download PDFInfo
- Publication number
- CN113591668B CN113591668B CN202110844610.XA CN202110844610A CN113591668B CN 113591668 B CN113591668 B CN 113591668B CN 202110844610 A CN202110844610 A CN 202110844610A CN 113591668 B CN113591668 B CN 113591668B
- Authority
- CN
- China
- Prior art keywords
- dam
- candidate
- detection
- area
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 90
- 238000004458 analytical method Methods 0.000 title claims abstract description 18
- 238000013135 deep learning Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000000034 method Methods 0.000 claims abstract description 22
- 239000002352 surface water Substances 0.000 claims abstract description 7
- 238000013136 deep learning model Methods 0.000 claims abstract description 4
- 238000013434 data augmentation Methods 0.000 claims description 10
- 230000003416 augmentation Effects 0.000 claims description 7
- 238000009825 accumulation Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000003672 processing method Methods 0.000 claims description 3
- 238000012732 spatial analysis Methods 0.000 claims 6
- 238000012850 discrimination method Methods 0.000 claims 1
- 238000012360 testing method Methods 0.000 abstract description 6
- 238000012795 verification Methods 0.000 description 4
- 238000012876 topography Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 101100096995 Caenorhabditis elegans strm-1 gene Proteins 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 239000004566 building material Substances 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000007499 fusion processing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000013138 pruning Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to an automatic detection method for a wide area unknown dam by using deep learning and space analysis, which comprises the following steps: candidate region extraction-surface water area constraint and administrative boundary dataset intersection constraint to obtain more accurate candidate regions; training a deep learning model, namely training three target recognition models, and fusing a detection result by an NMS algorithm and a length threshold; geographic analysis and comprehensive discrimination-comprehensive terrain constraint, deleting false detection frames by intersecting principle and other target open data sets, and further improving accuracy in dam detection. The automatic detection method provided by the invention has good performance on a test data set, the detection result is manually checked, the accuracy is 80.0%, the recall rate is 91.1%, and 39 new dams which are not on any data set are found. The result shows that the method can automatically, quickly and reliably detect the space position of the dam in the unknown area, and provides a flow thought for detecting the space position of other remote sensing targets.
Description
Technical Field
The invention relates to an automatic detection method for a wide-area unknown dam by using deep learning and space analysis.
Background
In recent years, the adoption and implementation of dams worldwide has steadily increased to accommodate environmental and hydraulic construction needs. The dam is used as an important water conservancy facility, and the data quality influences the river basin change and ecological evaluation from local to large scale. One of the key problems is how to quickly obtain the spatial position of the dam. Some organizations and institutions construct dam datasets by way of ground surveys, compilation of information on government websites or the internet, and fusion of existing database attributes. However, this can lead to loss of time, labor and money. Therefore, it is very urgent to develop a method for automatically detecting the spatial position of the dam.
Many scholars have performed some constructive work in the wide area identification of large scale targets such as airports, solar photovoltaic power stations, missile characterization sites, and the like. These students acquire target candidate areas using a geographic analysis method, and further accuracy improvement is performed by using scene classification and a target recognition model, so that the acquisition of wide-area spatial position data of these targets is not difficult nowadays. However, the dam is small in scale and various in type, the current research is still based on a single image or a slightly large-range spliced image, and the problem of wide-area target identification still exists. Therefore, in order to improve the accuracy and wide area detection capability of the existing method, we have focused on developing an automatic detection method for a wide area unknown dam.
Disclosure of Invention
The invention aims to solve the technical problems that: the method for automatically detecting the wide-area unknown dam is provided for overcoming the defects in the prior art.
In order to solve the technical problems, the invention provides an automatic detection method for a wide area unknown dam by using deep learning and space analysis, which comprises the following steps:
step 1, extracting candidate areas, namely acquiring candidate area point positions detected by a dam, wherein the data processing method comprises the following steps:
1.1 Area constraint, deleting the area with smaller area in the global surface water data set, and reserving the area with larger area as a preliminary candidate area;
1.2 Deleting polygons intersecting with the coastline in the preliminary candidate region to obtain a candidate region of the dam, sampling point positions of the boundary of the candidate region, and obtaining detection point positions of the dam;
step 2, training a deep learning model, namely training three target recognition models, wherein the first model is a dam detection model based on YOLO-v3, the second model is a dam detection model based on YOLO-v5l, and the third model is a dam detection model based on YOLO-v5x, and the training process is as follows:
2.1 Manually and randomly selecting a plurality of images containing dams in a target region range, and marking multiple polygons to obtain marked dam images;
2.2 Data augmentation, performing data augmentation on the selected image containing the dam as a training sample, including geometric augmentation and color augmentation;
2.3 Training a dam detection model of YOLO-v3 by using a training sample, and identifying an image at a dam detection point position by using the trained dam detection model of YOLO-v3 to obtain a first initial detection result;
2.4 Training the dam detection models based on the YOLO-v5l and the YOLO-v5x respectively by using training samples, and identifying images at the dam detection points by using the trained dam detection models based on the YOLO-v5l and the YOLO-v5x respectively to obtain a second initial detection result and a third initial detection result respectively;
2.5 Fusing and combining the first, second and third initial detection results by using an NMS algorithm and a length threshold value to obtain a comprehensive initial detection result of model detection;
step 3, geographic analysis and comprehensive discrimination, namely discriminating the comprehensive initial detection result by applying three geographic constraints in the comprehensive discrimination, wherein the method comprises the following steps:
3.1 The terrain fluctuation degree of the candidate area is judged based on the threshold values of the DEM, the slope direction, the terrain roughness index and the confluence accumulation amount, if the terrain fluctuation degree is low, the candidate area is a false detection candidate frame, the false detection candidate frame is deleted, and the residual detection result is obtained;
3.2 If the candidate area is not in the intersecting range of the road line and the river network, the candidate area is a false candidate frame; if the types of the lines of the intersection areas are all main roads, the candidate areas are road error candidate frames; the candidate area intersected with the coastline is a harbor or coastline error candidate frame; deleting the error candidate frames and obtaining the residual detection result;
3.3 Other target data, deleting the false candidate frame by using target position information in other open data sets, wherein the open data set information comprises: global land utilization data, farmland and bridge data of the Open Street Map; if the land use type in the candidate frame does not contain the impermeable surface, bare land and farmland data, the candidate frame is a false candidate frame; if the candidate frame contains farmland position information and bridge position information with larger area, the candidate frame is a false candidate frame; and eliminating the error candidate frames and obtaining a final detection result.
The automatic detection method of the wide-area unknown dam provided by the invention has good performance on a test data set, the detection result is manually checked, the accuracy is 80.0%, the recall rate is 91.1%, and 39 new dams which are not on any data set are found. The result shows that the method can automatically, quickly and reliably detect the space position of the dam in the unknown area, and provides a flow thought for detecting the space position of other remote sensing targets.
Drawings
The invention is further described below with reference to the accompanying drawings.
Fig. 1 is a general flow chart of an example of the present invention.
FIG. 2 shows the results of the detection of the spatial position of a dam after three models of the three test areas Qinsen, kannechuan and Okinawa are fused.
FIG. 3 is a view of a lost dam image after three models of the present invention are fused.
FIG. 4 is an explanatory diagram of constraints in the geographic analysis and comprehensive discrimination of the present invention.
FIG. 5 is a diagram showing the number of erroneous candidate boxes deleted under various constraints in the geographic analysis and comprehensive discrimination of the present invention.
Fig. 6 is a graph of the results of detection of the spatial position of a dam over a wide area for an example of the present invention.
Detailed Description
The technical route and operation steps of the present invention will be more apparent from the following detailed description of the present invention with reference to the accompanying drawings.
The surface water data set in the selection of the candidate area is a Joint Research Centre global surface water grid data set, and consists of 7 wave bands, wherein the maximum water body range data set is selected, and the spatial resolution is 30 meters. The raster data set is converted into a vector surface data set through the pruning of the area threshold. The administrative boundary data employed are vector face data of the GADM organization.
In the training of three target recognition models, 526 points in the Grand open data set are randomly selected by the dam sample, and google images are downloaded and data enhancement is carried out. Image data in dam detection is derived from Google Earth 19-level slices (resolution about 0.23 m), and the slices are downloaded at a size of 0.005 degrees or 0.005 degrees and spliced into a complete image, wherein the image of the size can fully cover the image around the potential dam point.
The data set required in the geographic analysis and comprehensive discrimination comprises a DEM data set (STRM 1 and AW 3D), a FROM-GLC10 global land utilization data set with spatial resolution of 30 meters and a road, bridge and farmland vector data set with spatial resolution of 10 meters and an Open Street Map. The newly discovered dam obtained in the visual verification is not present with the current dam dataset. The dam dataset is derived from: open acquisition databases of Grand-v1.1, grand-v1.3, GOOD-2015, GOOD-2020, OSM, FAO, GNIS, etc., and open data sets of dams provided by the national basic geographic information center of Japan.
The wide area unknown dam automatic detection method using deep learning and space analysis in the embodiment mainly comprises the following steps of
Step 1, extracting candidate areas, namely acquiring candidate area point positions detected by a dam, wherein the data processing method comprises the following steps:
a1, area constraint, including vectorization of the JRC global surface raster data set and deletion of potential blocks below an area threshold, and obtaining a preliminary candidate region.
A2, intersecting constraint, namely replacing coastline data with administrative boundary data, deleting polygons intersecting with the boundary line in the preliminary candidate region, acquiring the candidate region of the dam, and setting proper point position sampling point intervals to convert the point positions of the dam to be detected.
In the step, the extraction of the candidate area is based on a JRC global surface water data set, the area threshold is 135 square kilometers, the point location sampling interval is 30 meters, the area, the relative area (the ratio of the area of the candidate area to the JRC surface water area) of the candidate area after the extraction of the candidate area and the number of to-be-detected point locations are shown in the following table:
through the area threshold and the intersection constraint processing, the area of the candidate region can be effectively reduced by 55% -95%, and the total number of points to be detected is 12530.
Step 2, training a deep learning model, namely training three target recognition models, wherein the first model is a dam detection model based on YOLO-v3, the second model is a dam detection model based on YOLO-v5l, and the third model is a dam detection model based on YOLO-v5x, and the training process is as follows:
b1, labeling samples, namely randomly selecting dam point positions for labeling in a Japanese range based on a Grand dam data set. The dam is mostly ladder-shaped and thin strip-shaped in the remote sensing image, if the dam is marked by a positive rectangle, the marking frame can contain too much surrounding ground object information, the reinforced marking frame can be enlarged, the background information is increased, and the training of the model is not facilitated. Therefore, the method selects the polygon to mark the dam image selected randomly, and can effectively reduce the information quantity of the surrounding ground objects contained in the dam sample after data augmentation.
B2, data augmentation is limited by the number of marked dam samples, and sample data augmentation is needed to be carried out for improving the generalization capability of the model and training the model with stronger performance. The data augmentation comprises geometric augmentation such as rotation, mirroring, scaling, mosaic and the like, and color augmentation such as brightness increase and decrease, contrast enhancement and the like.
B3, training a dam detection model based on YOLO-v3, wherein YOLO-v3 is one of target detectors with excellent performance at present, and good balance is achieved in speed and precision. And then the model is used for identifying the images of all the points in the candidate area, so as to generate an initial result frame (1) containing correct or incorrect identification.
B4, training based on the YOLO-v5l and YOLLO-v5x dam target recognition models, wherein the YOLO-v5 model is one of the most advanced target detectors at present, and the speed and the accuracy are further improved. Then, the model is used for identifying the images of all the points in the candidate area to generate an initial result frame (2) (3) containing correct or incorrect identification;
in the step, the randomly selected 526 dam images are marked with the polygonal sample, so that the background information in the marking frame can be effectively reduced. The data augmentation uses augmentation methods such as brightness, contrast, rotation, mirroring, scaling, mosaic and the like, and after the data augmentation, the total number of the data is 34190 samples. Dam object model training based on yolo-v 3. In training, the parameters used are respectively: training an optimizer: adam; training steps: 100 times; training strategies: the dark frame layer was frozen and trained 50 times before training, all frame layers were trained and trained 50 times after training until training was completed. Training is based on the Yolo-v5l and Yolo-v5x dam object recognition models. In training, the parameters used are respectively: training an optimizer: momentum; initial learning rate 0.01; training strategies: training is continued after the training is performed for 3 times before the preheating training is performed until the training is completed. In addition, 173 dams are selected as test data of a dam target recognition model, the 173 dams are labeled and data amplified, 11008 test samples are generated, and test results of the three models are shown in the table:
the detection time of the three models on an average image is about 2 seconds, recall rate and accuracy are above 85%, and the detection speed and accuracy are high.
And B5, fusing and combining the initial results (1), (2) and (3) by using an NMS algorithm and a length threshold value to obtain a comprehensive initial detection result of the model detection. The idea of NMS is to search for local maxima, suppress non-maxima elements, and merge multiple candidate boxes of the same potential region into the highest scoring candidate box. Meanwhile, the size of the candidate frame is limited by utilizing the length threshold value, so that the larger error candidate frame can be effectively deleted.
In this step, the threshold of IoU in the NMS algorithm fusion process is set to 0.001, the score is set to 0.6, and the thresholds of length and width are set to 0.006 degrees. The steps of the NMS are as follows: (1) Ordering multiple candidate boxes of the same potential area according to confidence; (2) extracting the candidate frame with the highest confidence coefficient; (3) calculating IoU values for it and the remaining candidate boxes; (4) deleting IoU candidate boxes greater than the threshold value 0.6; (5) Repeating the steps 1-4 until IoU between the candidate frames is smaller than the threshold value, and the rest candidate frames are potential areas. The recall rate, the accuracy rate and the number of identified dams of the three models are shown in the following table, and compared with single model identification, the recall rate of the invention is obviously improved:
the results of the detection of 12530 images by the three models are shown in the table above, and the fusion results of the three models are shown in fig. 2, wherein the detection results of the Qinsen, kanagawa and Okinawa models respectively show that 115 dams are detected, and 39 dams are newly found and are not recorded in any dam data set. Wherein 8 dams were lost, as shown in fig. 3, the absence of dams 2, 4, 5, 6 in fig. 3 is due to the relatively small size of the dams in the whole image, and the absence of training samples for dams 1, 3, 7, 8.
Step 3, geographic analysis and comprehensive discrimination, namely discriminating the preliminary detection result by applying three geographic constraints in the comprehensive discrimination, and if the detection result meets the three geographic constraints, determining that the candidate frame is a position point of a dam; otherwise, the point location is not a dam, and constraint conditions of geographic analysis and comprehensive discrimination are as follows:
c1, if the dam is arranged in the detection frame, the relief of the topography around the dam is large, and the specific characteristic values are as follows: DEM, slope direction, topography rough index and confluence accumulation amount, threshold constraint is carried out based on the characteristic values, if topography fluctuation degree is low, the candidate area is a false detection non-dam candidate frame, then the false detection candidate frames are deleted, and the residual detection result is obtained, as shown in (a) of fig. 4, the potential candidate area of the dam can be screened out by effectively utilizing the height difference at two sides of the dam. The DEM is selected from AW3D (currently, the data with highest DEM precision and 30 m spatial resolution) and STRM1 (the precision in lakes is higher and 30 m spatial resolution), slope direction, rough terrain index and convergence accumulation are all data obtained after DEM hydrologic analysis, comprehensive judgment can be carried out from multiple angles, and potential candidate frames of the dam can be effectively screened out.
And C2, intersecting rules, including intersection with roads and coastline intersection. The intersection of the road line and the river network may be a dam or a bridge where two conditions need to be satisfied, and if the candidate region is not within the intersection range, the candidate region is a false candidate frame. On the other hand, if the types of the lines of the intersection area are primary and secondary roads, the candidate area is a road error candidate frame; further, the coastline data may be replaced with boundary line data, which may be a harbor or coastline error candidate box intersecting the coastline. Then, these erroneous candidate frames are deleted and the remaining detection results are obtained, as in the case of (b) group in fig. 4, where the right is the case of intersecting with the boundary line data, and the left is the case of intersecting with the road in the candidate frames. The line data is derived from the line data in the OSM open data set, the coastline data is derived from the GADM by adopting the boundary line data, and the coastline data are vector data sets. By using the intersecting principle, false candidate frames in roads and coastlines can be effectively removed.
And C3, deleting the error candidate frame by using other target data and target position information in other open data sets. The open dataset information includes: global land utilization data (raster data set From-GLC10, spatial resolution 10 meters), open Street Map Open data set farm and bridge data. According to the main use and building materials of the dam, if the land use type in the candidate frame does not contain the water impermeable surface, bare land and farmland data, the candidate frame is a false candidate frame; in addition, if the candidate frame contains other target information, such as farmland position information and bridge position information with larger area, the candidate frame is occupied by farmland and bridges, and no redundant space exists in a redundant dam. Therefore, the false candidate frames can be effectively removed, and the detection result of comprehensive judgment can be obtained, and the constraint effect of each method is shown in fig. 5.
In this step, specific threshold conditions in the three steps are shown in the following table, and the area ratio represents the ratio of the area of the farmland in the candidate frame to the area of the candidate frame, the range difference is the difference between the maximum value and the minimum value, and the standard deviation can reflect the discrete degree of a group of data and the average value:
verification example:
the following description will proceed with this example for verifying the accuracy and reliability of the method of the present invention.
The reliability verification of the invention is carried out by visually checking the number of dams in all candidate areas, counting the number of dams, and calculating the recall rate and the accuracy rate as shown in the following table:
by manual verification, 123 dams exist in the 11530 scene image, and 223 candidate frames exist. The invention can correctly identify 119 dams, the recall rate is 91.1%, and the accuracy rate reaches 80.00%. The dam space position information detected in the three areas is shown in fig. 6, which contains 115 pieces of detected dam position information and 39 pieces of newly found dam position information.
The invention realizes the method for quickly and automatically detecting the space position of the wide area dam, although the invention has error and omission of results, the overall recall rate reaches 91.1 percent, the accuracy rate reaches 80.0 percent, in the detection process, the detection speed of three models is 2 seconds/scene, the total detection speed of 12530 scenic images is 21 hours, and the speed is further increased along with the continuous change of the data structure and the periodical update of hardware equipment.
In addition to the embodiments described above, other embodiments of the invention are possible. All technical schemes formed by equivalent substitution or equivalent transformation fall within the protection scope of the invention.
Claims (6)
1. An automatic wide area unknown dam detection method using deep learning and spatial analysis, comprising the steps of:
step 1, extracting candidate areas, namely acquiring candidate area point positions detected by a dam, wherein the data processing method comprises the following steps:
1.1 Area constraint, deleting the area with the area smaller than the area threshold value in the global surface water data set, and reserving the area with the area not smaller than the area threshold value as a preliminary candidate area, wherein the area threshold value in the step 1 is 135 square kilometers;
1.2 Deleting polygons intersecting with the coastline in the preliminary candidate region to obtain a candidate region of the dam, sampling point positions of the boundary of the candidate region, and obtaining detection point positions of the dam;
step 2, training a deep learning model, namely training three target recognition models, wherein the first model is a dam detection model based on YOLO-v3, the second model is a dam detection model based on YOLO-v5l, and the third model is a dam detection model based on YOLO-v5x, and the training process is as follows:
2.1 Manually and randomly selecting a plurality of images containing dams in a target region range, and marking multiple polygons to obtain marked dam images;
2.2 Data augmentation, performing data augmentation on the selected image containing the dam as a training sample, including geometric augmentation and color augmentation;
2.3 Training a dam detection model of YOLO-v3 by using a training sample, and identifying an image at a dam detection point position by using the trained dam detection model of YOLO-v3 to obtain a first initial detection result;
2.4 Training the dam detection models based on the YOLO-v5l and the YOLO-v5x respectively by using training samples, and identifying images at the dam detection points by using the trained dam detection models based on the YOLO-v5l and the YOLO-v5x respectively to obtain a second initial detection result and a third initial detection result respectively;
2.5 Fusing and combining the first, second and third initial detection results by using an NMS algorithm and a length threshold value to obtain a comprehensive initial detection result of model detection;
step 3, geographic analysis and comprehensive discrimination, namely discriminating the comprehensive initial detection result by applying three geographic constraints in the comprehensive discrimination, wherein the method comprises the following steps:
3.1 The terrain fluctuation degree of the candidate area is judged based on the threshold values of the DEM, the slope direction, the terrain roughness index and the confluence accumulation amount, if the terrain fluctuation degree is low, the candidate area is a false detection candidate frame, the false detection candidate frame is deleted, and the residual detection result is obtained;
3.2 If the candidate area is not in the intersecting range of the road line and the river network, the candidate area is a false candidate frame; if the types of the lines of the intersection areas are all main roads, the candidate areas are road error candidate frames; the candidate area intersected with the coastline is a harbor or coastline error candidate frame; deleting the error candidate frames and obtaining the residual detection result;
3.3 Other target data, deleting the false candidate frame by using target position information in other open data sets, wherein the open data set information comprises: global land utilization data, farmland and bridge data of the Open Street Map; if the land use type in the candidate frame does not contain the impermeable surface, bare land and farmland data, the candidate frame is a false candidate frame; if the candidate frame contains farmland position information and bridge position information with the area ratio of farmland being more than 0.7, the candidate frame is a false candidate frame; and eliminating the error candidate frames and obtaining a final detection result.
2. The method for automatic wide area unknown dam detection using deep learning and spatial analysis according to claim 1, wherein: in step 2.2, the geometric enhancement includes: rotation, mirroring and scaling; the color enhancement includes: brightness increase and decrease and contrast enhancement.
3. The method for automatic wide area unknown dam detection using deep learning and spatial analysis according to claim 1, wherein: in step 2.5, the NMS algorithm has a IoU threshold of 0.001 and a length threshold of 0.006 degrees by 0.006 degrees.
4. The method for automatic wide area unknown dam detection using deep learning and spatial analysis according to claim 1, wherein: in step 3, in the space position of the detection dam, the scale is 0.005 degrees of the candidate point.
5. The method for automatic wide area unknown dam detection using deep learning and spatial analysis according to claim 1, wherein: in the step 3.1, the standard deviation threshold of the DEM is 2.0, the range threshold of the slope direction is 170.0 degrees, the standard deviation threshold of the terrain roughness index is 3.9, and the standard deviation threshold of the confluence accumulation amount is 20.0 under the condition of the terrain fluctuation constraint condition in the geographic analysis and comprehensive judgment method.
6. The method for automatic wide area unknown dam detection using deep learning and spatial analysis according to claim 1, wherein: in step 3.2, the road types are primary and secondary, based on the intersecting constraint conditions in the geographic analysis and comprehensive discrimination method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110844610.XA CN113591668B (en) | 2021-07-26 | 2021-07-26 | Wide area unknown dam automatic detection method using deep learning and space analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110844610.XA CN113591668B (en) | 2021-07-26 | 2021-07-26 | Wide area unknown dam automatic detection method using deep learning and space analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113591668A CN113591668A (en) | 2021-11-02 |
| CN113591668B true CN113591668B (en) | 2023-11-21 |
Family
ID=78250025
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110844610.XA Active CN113591668B (en) | 2021-07-26 | 2021-07-26 | Wide area unknown dam automatic detection method using deep learning and space analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113591668B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117809172A (en) * | 2023-11-20 | 2024-04-02 | 北京师范大学 | Dam identification method and device based on remote sensing picture |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480341A (en) * | 2017-07-21 | 2017-12-15 | 河海大学 | A kind of dam safety comprehensive method based on deep learning |
| CN108805983A (en) * | 2018-06-11 | 2018-11-13 | 西南交通大学 | " more rule unifications " regional spatial planning data GIS- gravitational field analysis models |
| WO2018208939A1 (en) * | 2017-05-09 | 2018-11-15 | Neurala, Inc. | Systems and methods to enable continual, memory-bounded learning in artificial intelligence and deep learning continuously operating applications across networked compute edges |
| CN109977812A (en) * | 2019-03-12 | 2019-07-05 | 南京邮电大学 | A kind of Vehicular video object detection method based on deep learning |
| CN110084093A (en) * | 2019-02-20 | 2019-08-02 | 北京航空航天大学 | The method and device of object detection and recognition in remote sensing images based on deep learning |
| CN110321811A (en) * | 2019-06-17 | 2019-10-11 | 中国工程物理研究院电子工程研究所 | Depth is against the object detection method in the unmanned plane video of intensified learning |
| CN111582093A (en) * | 2020-04-27 | 2020-08-25 | 北京工业大学 | Automatic small target detection method in high-resolution image based on computer vision and deep learning |
| WO2021026217A1 (en) * | 2019-08-06 | 2021-02-11 | Saudi Arabian Oil Company | Method and system for land encroachment detection and surveillance |
| CN112395998A (en) * | 2020-11-19 | 2021-02-23 | 南京大学 | Verification method for airport space position in global open database |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11461915B2 (en) * | 2020-01-06 | 2022-10-04 | Qualcomm Incorporated | Object size estimation using camera map and/or radar information |
-
2021
- 2021-07-26 CN CN202110844610.XA patent/CN113591668B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018208939A1 (en) * | 2017-05-09 | 2018-11-15 | Neurala, Inc. | Systems and methods to enable continual, memory-bounded learning in artificial intelligence and deep learning continuously operating applications across networked compute edges |
| CN107480341A (en) * | 2017-07-21 | 2017-12-15 | 河海大学 | A kind of dam safety comprehensive method based on deep learning |
| CN108805983A (en) * | 2018-06-11 | 2018-11-13 | 西南交通大学 | " more rule unifications " regional spatial planning data GIS- gravitational field analysis models |
| CN110084093A (en) * | 2019-02-20 | 2019-08-02 | 北京航空航天大学 | The method and device of object detection and recognition in remote sensing images based on deep learning |
| CN109977812A (en) * | 2019-03-12 | 2019-07-05 | 南京邮电大学 | A kind of Vehicular video object detection method based on deep learning |
| CN110321811A (en) * | 2019-06-17 | 2019-10-11 | 中国工程物理研究院电子工程研究所 | Depth is against the object detection method in the unmanned plane video of intensified learning |
| WO2021026217A1 (en) * | 2019-08-06 | 2021-02-11 | Saudi Arabian Oil Company | Method and system for land encroachment detection and surveillance |
| CN111582093A (en) * | 2020-04-27 | 2020-08-25 | 北京工业大学 | Automatic small target detection method in high-resolution image based on computer vision and deep learning |
| CN112395998A (en) * | 2020-11-19 | 2021-02-23 | 南京大学 | Verification method for airport space position in global open database |
Non-Patent Citations (5)
| Title |
|---|
| "Deep Learning based Detection of potholes in Indian roads using YOLO";Dharneeshkar J等;《2020 International Conference on Inventive Computation Technologies》;全文 * |
| "基于深度学习的光学遥感图像多标签分类技术研究";刁莹煜;《中国优秀硕士学位论文全文数据库》;全文 * |
| "时间序列遥感技术支持的国土空间开发监测研究";孙越凡;《中国优秀硕士学位论文全文数据库》;全文 * |
| 农业信息成像感知与深度学习应用研究进展;孙红;李松;李民赞;刘豪杰;乔浪;张瑶;;农业机械学报(05);全文 * |
| 基于深度学习的遥感影像基础设施目标检测研究;王港;陈金勇;高峰;吴金亮;;无线电工程(03);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113591668A (en) | 2021-11-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Degerli et al. | Using the remote sensing method to simulate the land change in the year 2030 | |
| CN106485254A (en) | A kind of electric transmission line channel topographical features change extraction method | |
| CN109086649A (en) | Satellite remote sensing images identifying water boy method | |
| CN112906455A (en) | Coastal zone ecological system remote sensing identification method | |
| CN113591668B (en) | Wide area unknown dam automatic detection method using deep learning and space analysis | |
| CN114067245A (en) | Method and system for identifying hidden danger of external environment of railway | |
| Zhang et al. | Building footprint and height information extraction from airborne LiDAR and aerial imagery | |
| Stanislawski et al. | Automated road breaching to enhance extraction of natural drainage networks from elevation models through deep learning | |
| Le Bris et al. | Change detection in a topographic building database using submetric satellite images | |
| Schaffer et al. | Challenges and possible approaches for using GIS as a tool in historical geography landscape research: A meta-analysis review | |
| CN114663640B (en) | Submarine geographic entity demarcation and classification method based on landform and structural characteristics | |
| Boueshagh et al. | Estimating water level in the Urmia Lake using satellite data: a machine learning approach | |
| Miandad et al. | Landslide mapping using multiscale LiDAR digital elevation models | |
| CN103218814A (en) | Self-adoption water submerging optimization segmentation method for defects in radiographic inspection | |
| CN114005027A (en) | Urban single tree detection system and method based on unmanned aerial vehicle image | |
| Dall'Osso et al. | A novel approach (the CRATER method) for assessing tsunami vulnerability at the regional scale using ASTER imagery | |
| Islam | Bank erosion and movement of river channel: A study of Padma and Jamuna Rivers in Bangladesh using remote sensing and GIS | |
| CN114972991B (en) | Automatic recognition method and system for collapsing sentry | |
| Verrou et al. | Quantitative comparison of temporal changes in coastline utilizing Google Earth, Landsat and Sentinel-2 historical time series case study: the Mytikas coast, Aitoloakarnania Prefecture, Greece | |
| CN117809172A (en) | Dam identification method and device based on remote sensing picture | |
| Taruna et al. | Detection of Shoreline Changes With Automated Water Extraction Index (AWEI) In Galur District, Kulon Progo Regency | |
| Brydsten et al. | Digital elevation models for site investigation programme in Oskarshamn. Site description version 1.2 | |
| Arda et al. | Monitoring the change of coastline with remote sensing and GIS: a case study from Izmit and Gemlik Gulfs, Turkey | |
| MacFaden et al. | A tool for the automated detection of damaged transportation infrastructure | |
| CN118229842A (en) | Topography change repairing and measuring method, server and storage medium |
Legal Events
| 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 |