CN116503464B - Farmland building height prediction method based on remote sensing image - Google Patents
Farmland building height prediction method based on remote sensing image Download PDFInfo
- Publication number
- CN116503464B CN116503464B CN202310747563.6A CN202310747563A CN116503464B CN 116503464 B CN116503464 B CN 116503464B CN 202310747563 A CN202310747563 A CN 202310747563A CN 116503464 B CN116503464 B CN 116503464B
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- image
- sensing image
- farmland
- data
- 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
- 238000000034 method Methods 0.000 title claims abstract description 53
- 230000003287 optical effect Effects 0.000 claims abstract description 25
- 238000011156 evaluation Methods 0.000 claims abstract description 23
- 238000007781 pre-processing Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000012360 testing method Methods 0.000 claims abstract description 13
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 238000012795 verification Methods 0.000 claims abstract description 10
- 238000011176 pooling Methods 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 3
- 108700041286 delta Proteins 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 2
- 238000007726 management method Methods 0.000 abstract description 5
- 238000005259 measurement Methods 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 43
- 238000005516 engineering process Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000012271 agricultural production Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000009435 building construction Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/02—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness
- G01B11/06—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness for measuring thickness ; e.g. of sheet material
- G01B11/0608—Height gauges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/86—Combinations of radar systems with non-radar systems, e.g. sonar, direction finder
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Remote Sensing (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Radar, Positioning & Navigation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Astronomy & Astrophysics (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Geometry (AREA)
- Computer Networks & Wireless Communication (AREA)
- Image Processing (AREA)
Abstract
The invention provides a farmland building height prediction method based on remote sensing images, which comprises the following steps: s1) a remote sensing optical image and an SAR image of a farmland building are in one-to-one correspondence with a normalized digital surface model nDSM to form a training set and a testing set; s2) carrying out data fusion and data enhancement preprocessing operation on the training set data; s3) constructing a farmland building remote sensing image height estimation network model; s4) constructing a network model loss function; s5) using the test set to carry out verification and evaluation on a farmland building height estimation task under a remote sensing scene on a farmland building remote sensing image height estimation network model; s6) outputting a final farmland building remote sensing image height estimation network model after the evaluation index meets the requirement, and applying the network model to farmland building remote sensing image height estimation. The invention is used for improving the measurement efficiency and the data precision of the farmland height, and can also provide powerful support and help for rural planning and management.
Description
Technical Field
The invention relates to the technical fields of farmland building construction, remote sensing technology, computer image technology and the like, in particular to a farmland building height prediction method based on remote sensing images.
Background
The farmland building height prediction is an important ring of rural planning development design. In rural areas, the height of farmland buildings is often one of the key indicators for agricultural planning, land utilization and agricultural production management. At present, the traditional manual measurement method has the problems of low efficiency, high cost, low data precision and the like.
In recent years, with the increasing development of high-resolution remote sensing technology, the acquisition of buildings by using high-resolution remote sensing images has attracted the research of many students at home and abroad. In addition, with the rise of deep learning tide, the method is rapidly developed in the remote sensing field, and the height of a ground building can be detected through a deep learning technology. By combining the deep learning technology and the remote sensing technology, the building in the image video shot by the remote sensing satellite can be directly detected, and the height of the building can be estimated, so that the workload of manually analyzing the image video of the remote sensing satellite is greatly reduced, and the urban management efficiency is further improved.
At present, when the problem of remote sensing image height estimation is solved by the existing method, three defects exist: firstly, shooting at the same place can be calculated only by a plurality of visual angles, so that timeliness is reduced; secondly, the existing method cannot utilize the remote sensing optical image and the SAR satellite image to carry out altitude estimation; thirdly, the time and space complexity of the existing method is larger, and the method is far away from practical application.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a farmland building height prediction method based on remote sensing images, which is used for improving the measurement efficiency and data precision of farmland heights and simultaneously can provide powerful support and help for rural planning and management.
The farmland building height prediction method based on the remote sensing image is characterized by comprising the following steps of:
s1) a remote sensing optical image and an SAR image of a farmland building are in one-to-one correspondence with a normalized digital surface model nDSM to form a training set and a testing set;
s2) constructing a data preprocessing module, and performing data fusion and data enhancement preprocessing operation on the training set data;
s3) constructing a farmland building remote sensing image height estimation network model based on encoder and decoder architecture;
s4) constructing a network model loss function, including a pixel mean square loss function and a binary cross entropy loss function;
s5) using a test set to carry out verification and evaluation on a farmland building height estimation task under a remote sensing scene on a farmland building remote sensing image height estimation network model to obtain two evaluation indexes of mean square error MSE and accuracy delta 1;
and S6) if the evaluation index in the step S5) does not meet the requirement, repeating the steps S2) to S5), if the evaluation index meets the requirement, outputting a final farmland building remote sensing image height estimation network model, and applying the farmland building remote sensing image height estimation network model to farmland building remote sensing image height estimation.
Preferably, in step S1), the remote sensing optical image includes a high resolution satellite No. 1 with a resolution of 2m, a high resolution satellite No. two with a resolution of 0.8m, a high resolution satellite No. three with a resolution of 0.5m, a SuperView satellite with a resolution of 0.5m, a resource satellite No. one with a resolution of 16m or 30m, a resource satellite No. two with a resolution of 2m, 10m or 16m, a resource satellite No. three with a resolution of 3m, 5m, 10m or 30m, and a space satellite No. one with a resolution of 1m or a space satellite No. two with a resolution of 0.6 m; the SAR image comprises a third high-resolution satellite with resolution of 1m, 3m or 5m-30 m.
Preferably, in step S1), the normalized digital surface model is generated according to stereo images captured by high-resolution satellite No. two and world view satellite, and the ground sampling distance GSD is 7m; both the telemetry optical image and the SAR image were resampled to 512 x 512 pictures at the same resolution of 2.0 m.
Preferably, in step S2), the data fusion operation is to fuse the remote sensing optical image and the SAR image, and the formula is as follows:
；
wherein, I is the fused image, R, G, B is three-channel data of the remote sensing optical image, sar is SAR satellite data, and Cat () is a splicing function; z () is a normalization function, and the specific formula is:
；
wherein the method comprises the steps ofX is original image data, mean () is a Mean function, std () is a variance function.
Preferably, in step S2), the data enhancing operation includes a calculation process through a random flip function, a random rotation function, and a random clipping function.
Preferably, in step S3), the farmland building remote sensing image height estimation network model includes an encoder for downsampling and a decoder for upsampling:
the encoder comprises five feature extraction downsampling blocks, wherein a pooling layer is used for increasing receptive fields at the tail of each block, and the relative positions after pooling are stored, and the encoder comprises 13 convolutions layers and 5 pooling layers;
the decoder comprises five upsampling blocks, wherein the top of each block uses an upsampling layer to amplify the remote sensing image, and the remote sensing image is restored according to the relative position saved in the encoder stage, and the decoder has 5 upsampling layers; the last upsampling block adds a Softmax activation layer to get the same size pixel level height value as the original.
Preferably, in step S4), the network model loss function formula is as follows:
L = L mse + L bce
wherein L is mse Is a pixel mean square loss function, L bce Is a binary cross entropy loss function;
the pixel mean square loss function L mse The formula is:
；
wherein y is ij For the true value, p, on the remote sensing image (i, j) ij H and W are predicted values on the remote sensing image (i, j) and respectively represent the length and the width of the remote sensing image;
the binary cross entropy loss function L bce The formula is:
；
wherein n is the number of samples of the data set, p i For predicted probability values, y i The specific formula is as follows:
；
in the formula, h i Is the actual height of the farmland building on the remote sensing image.
The invention also provides a farmland building height prediction system based on the remote sensing image, which is characterized by comprising a data acquisition module, a data preprocessing module, a farmland building remote sensing image height estimation network model, a verification evaluation module and a prediction output module;
the data acquisition module is used for: the method comprises the steps of correspondingly forming a training set and a testing set by remote sensing optical images and SAR images of a farmland building and a normalized digital surface model nDSM one by one;
the data preprocessing module is used for: the data fusion module fuses the remote sensing optical image and the SAR image, and the formula is as follows:
；
wherein, I is the fused image, R, G, B is three-channel data of the remote sensing optical image, sar is SAR satellite data, and Cat () is a splicing function; z () is a normalization function, and the specific formula is:
；
wherein the method comprises the steps ofX is original image data, mean () is a Mean function, std () is a variance function;
the data enhancement module: the method comprises the steps of performing calculation processing on image data by using a random overturn function, a random rotation function and a random clipping function;
the farmland building remote sensing image height estimation network model comprises the following steps: comprising an encoder for downsampling and a decoder for upsampling;
the verification and evaluation module: the method comprises the steps of verifying and evaluating a farmland building height estimation task under a remote sensing scene for a farmland building remote sensing image height estimation network model until an evaluation index meets the requirement;
the prediction output module: and the method is used for inputting the remote sensing image of the farmland building to be predicted into the network model for estimating the height of the remote sensing image of the farmland building and outputting the estimated height of the remote sensing image of the farmland building.
Further, the encoder comprises five feature extraction downsampling blocks, wherein a pooling layer is used at the tail of each block to increase the receptive field, and the relative position after pooling is stored, and the encoder comprises 13 convolutional layers and 5 pooling layers; the decoder comprises five upsampling blocks, wherein the top of each block uses an upsampling layer to amplify the remote sensing image, and the remote sensing image is restored according to the relative position saved in the encoder stage, and the decoder has 5 upsampling layers; the last upsampling block adds a Softmax activation layer to get the same size pixel level height value as the original.
The invention further provides a computer readable storage medium storing a computer program which when executed by a processor realizes the farmland building height prediction method based on the remote sensing image.
The invention has the beneficial effects that:
1. the invention greatly improves the measurement efficiency and the data precision of the farmland height, and simultaneously provides powerful support and help for rural planning and management;
2. the method fully utilizes the multi-mode satellite image, realizes that the height of the building in the remote sensing image can be estimated by adopting the picture shot by one view angle, greatly reduces the calculated amount and improves the timeliness of data utilization;
3. the invention uses the data preprocessing module to fuse and strengthen the remote sensing optical image data and SAR image data, and fully utilizes the information of SAR images, so that the data characteristics of different modes can be used for network learning;
4. the remote sensing image height estimation network model constructed by the invention adopts the downsampling block as an encoder and the upsampling block as a decoder, so that the calculated amount is reduced, the space-time complexity is reduced, and the accuracy of height estimation is improved compared with the deconvolution calculation of the existing method;
5. the loss function of the network model adopted by the invention comprises a pixel mean square loss function and a binary cross entropy loss function, not only considers the error of the height estimation value and the error of building identification under the remote sensing scene, but also optimizes again to reduce the calculated amount, so that the method is particularly suitable for farmland building height estimation under the remote sensing scene.
Drawings
Fig. 1 is a visual effect diagram of a farmland building height prediction method based on remote sensing images.
Detailed Description
The invention is described in further detail below with reference to the drawings and specific examples, but embodiments of the invention are not limited thereto.
The farmland building height prediction method based on the remote sensing image provided by the invention specifically comprises the following steps:
s1) the remote sensing optical image and the SAR image of the farmland building are in one-to-one correspondence with the normalized digital surface model nDSM to form a training set and a testing set.
In this embodiment, the remote sensing optical image is from a high-resolution satellite No. 2m, a high-resolution satellite No. two with resolution of 0.8m, a high-resolution satellite No. three with resolution of 0.5m, a SuperView satellite with resolution of 0.5m, a resource satellite No. one with resolution of 16m or 30m, a resource satellite No. two with resolution of 2m, 10m or 16m, a resource satellite No. three with resolution of 3m, 5m, 10m or 30m, a space hawk satellite No. one with resolution of 1m or a space hawk satellite No. two with resolution of 0.6 m; the SAR image comes from a high-resolution satellite III with the resolution of 1m, 3m or 5m-30 m; the normalized digital surface model is generated according to the stereo images captured by the high-resolution satellite II and the WorldView satellite, and the ground sampling distance GSD is 7m; all the images above are resampled to 512 x 512 pictures with the same resolution of 2.0 m; the training set selected contained 1773 samples and the test set contained 579 samples.
S2) constructing a data preprocessing module, performing data fusion and data enhancement preprocessing operation on the training set data, wherein the data preprocessing module comprises a data fusion module and a data enhancement module, and the accuracy can be further improved by preprocessing the training set data in the step S1).
Optionally, the data fusion module fuses the remote sensing optical image and the SAR image, and the formula is as follows:
；
wherein, I is the fused image, R, G, B is three-channel data of the remote sensing optical image, sar is SAR satellite data, and Cat () is a splicing function; z () is a normalization function, and the specific formula is:
；
wherein the method comprises the steps ofX is original image data, mean () is a Mean function, std () is a variance function.
Optionally, the data enhancement module includes a random flip function, a random rotation function, a random clipping function.
S3) constructing a farmland building remote sensing image height estimation network model based on the encoder and decoder architecture. The remote sensing image height estimation network model includes an encoder for downsampling and an upsampled decoder.
The encoder comprises five feature extraction downsampling blocks, wherein each block respectively comprises 2, 3 and 3 output convolution layers with the same size, namely the size of a remote sensing image after convolution is unchanged, a pooling layer is used at the tail of each block to increase a receptive field, the relative position after pooling is stored, and the encoder comprises 13 convoluting layers and 5 pooling layers;
the decoder comprises five upsampling blocks, the head of each block uses an upsampling layer to amplify the remote sensing image, the remote sensing image is restored according to the relative position saved in the encoder stage, and the decoder has 5 upsampling layers; the last upsampling block adds a Softmax activation layer to get the same size pixel level height value as the original.
S4) constructing a network model loss function, including a pixel mean square loss function and a binary cross entropy loss function;
in this embodiment, the network model loss function formula is as follows:
L = L mse + L bce
wherein L is mse Is a pixel mean square loss function, L bce Is a binary cross entropy loss function;
specifically, the pixel mean square loss function formula is:
；
wherein y is ij For the true value, p, on the remote sensing image (i, j) ij For predicted values, H andw represents the length and width of the remote sensing image;
the specific formula of the binary cross entropy loss function is as follows:
；
where n is the number of samples of the dataset, p i For predicted probability values, y i The specific formula is as follows:
；
wherein h is i Is the actual height of the farmland building on the remote sensing image.
S5) using the test set to carry out verification and evaluation on the farmland building height estimation task under the remote sensing scene on the whole network model, and obtaining two evaluation indexes of mean square error MSE and accuracy delta 1.
In the step, training equipment adopts 4 NVIDIA GTX 4096 24GB display cards, the learning rate optimized by an Adam algorithm is set to be 0.001, the batch size is set to be 32, the maximum training times are 300, and the attenuation strategy of the learning rate is adjusted according to the fact that the Loss of a test set is not reduced; obtaining initial parameters by training the whole network model: a weight parameter W and a bias parameter B. The weight parameter W and the bias parameter B are the set of all parameters used in the method of the invention.
And S6) if the evaluation index in the step S5) does not meet the requirement, repeating the steps S2) to S5), and if the evaluation index meets the requirement, outputting a final overall network model, and applying the network model to farmland building remote sensing image height estimation.
The results of comparing the method of the present invention with existing methods are shown in the following table, where MSE is the mean square error and delta1 is the accuracy. From the results in the table, it can be seen that the building height estimation accuracy of the method of the present invention is higher than that of the existing method, and the effectiveness of the present invention can be demonstrated.
The invention also provides a remote sensing image height estimation system and a computer readable storage medium, which are used for executing the remote sensing image height estimation method.
Specifically, the farmland building height prediction system based on the remote sensing image comprises a data acquisition module, a data preprocessing module, a farmland building remote sensing image height estimation network model, a verification and evaluation module and a prediction output module;
and a data acquisition module: the method comprises the steps of correspondingly forming a training set and a testing set by remote sensing optical images and SAR images of a farmland building and a normalized digital surface model nDSM one by one;
and a data preprocessing module: the system comprises a data fusion module and a data enhancement module, wherein the data fusion module fuses the remote sensing optical image and the SAR image;
and a data enhancement module: the method comprises the steps of performing calculation processing on image data by using a random overturn function, a random rotation function and a random clipping function;
farmland building remote sensing image height estimation network model: comprising an encoder for downsampling and a decoder for upsampling; in the embodiment, the encoder comprises five feature extraction downsampling blocks, a pooling layer is used at the tail of each block to increase the receptive field, and the relative position after pooling is stored, wherein the encoder comprises 13 convolutions layers and 5 pooling layers; the decoder comprises five upsampling blocks, the head of each block uses an upsampling layer to amplify the remote sensing image, the remote sensing image is restored according to the relative position saved in the encoder stage, and the decoder has 5 upsampling layers; adding a Softmax activation layer to the last up-sampling block to obtain a pixel level height value with the same size as the original image;
and (3) a verification and evaluation module: the method comprises the steps of verifying and evaluating a farmland building height estimation task under a remote sensing scene for a farmland building remote sensing image height estimation network model until an evaluation index meets the requirement;
and a prediction output module: and the method is used for inputting the remote sensing image of the farmland building to be predicted into the network model for estimating the height of the remote sensing image of the farmland building and outputting the estimated height of the remote sensing image of the farmland building.
The remote sensing image height estimation system can estimate the height of a building in a remote sensing image by only adopting pictures shot by one view angle, thereby greatly reducing the calculated amount and improving the timeliness of data utilization; compared with the deconvolution calculation of the existing method, the constructed network model based on the encoder and decoder architecture reduces the calculated amount, reduces the space-time complexity, improves the accuracy of the height estimation and is more convenient for practical application.
Fig. 1 is a visual effect diagram of the present invention, in which fig. 1-1 is a remote sensing image, and fig. 1-2 is a height estimation diagram of a building in the remote sensing image obtained by applying the system and method of the present invention, and the light color in the diagram represents the height of the building. According to fig. 1-2, the height results in the figures can be directly used to identify the building height in the remote sensing image of fig. 1-1.
The foregoing is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various improvements and modifications can be made to the present invention without departing from the technical principles of the present invention, and such improvements and modifications fall within the scope of the appended claims.
Claims (9)
1. A farmland building height prediction method based on remote sensing images is characterized in that: the method comprises the following steps:
s1) a remote sensing optical image and an SAR image of a farmland building are in one-to-one correspondence with a normalized digital surface model nDSM to form a training set and a testing set;
s2) constructing a data preprocessing module, and performing data fusion and data enhancement preprocessing operation on the training set data;
s3) constructing a farmland building remote sensing image height estimation network model based on encoder and decoder architecture;
s4) constructing a network model loss function, including a pixel mean square loss function and a binary cross entropy loss function; the network model loss function formula is as follows:
L＝L mse +L bce
wherein L is mse Is a pixel mean square loss function, L bce Is a binary cross entropy loss function;
the pixel mean square loss function L mse The formula is:
wherein y is ij For the true value, p, on the remote sensing image (i, j) ij H and W are predicted values on the remote sensing image (i, j) and respectively represent the length and the width of the remote sensing image;
the binary cross entropy loss function L bce The formula is:
wherein p is i For predicted probability values, y i The specific formula is as follows:
in the formula, h i The real height of the farmland building on the remote sensing image;
s5) using a test set to carry out verification and evaluation on a farmland building height estimation task under a remote sensing scene on a farmland building remote sensing image height estimation network model to obtain two evaluation indexes of mean square error MSE and accuracy delta 1;
and S6) if the evaluation index in the step S5) does not meet the requirement, repeating the steps S2) to S5), if the evaluation index meets the requirement, outputting a final farmland building remote sensing image height estimation network model, and applying the farmland building remote sensing image height estimation network model to farmland building remote sensing image height estimation.
2. The remote sensing image-based farmland building height prediction method according to claim 1, wherein: in the step S1), the remote sensing optical image includes a high-resolution satellite No. 1 with a resolution of 2m, a high-resolution satellite No. two with a resolution of 0.8m, a high-resolution satellite No. three with a resolution of 0.5m, a SuperView satellite with a resolution of 0.5m, a resource satellite No. one with a resolution of 16m or 30m, a resource satellite No. two with a resolution of 2m, 10m or 16m, a resource satellite No. three with a resolution of 3m, 5m, 10m or 30m, and a space hawk satellite No. one with a resolution of 1m or a space hawk satellite No. two with a resolution of 0.6 m; the SAR image comprises a third high-resolution satellite with the resolution of 1m, 3m or 5m-30 m.
3. The remote sensing image-based farmland building height prediction method according to claim 2, wherein: in the step S1), the normalized digital surface model is generated according to the stereo images captured by the high-resolution satellite II and the WorldView satellite, and the ground sampling distance GSD is 7m; both the telemetry optical image and the SAR image were resampled to 512 x 512 pictures at the same resolution of 2.0 m.
4. The remote sensing image-based farmland building height prediction method according to claim 1, wherein: in step S2), the data fusion operation is to fuse the remote sensing optical image and the SAR image, and the formula is as follows:
I＝Cat(Z(R)，Z(G)，Z(B)，Z(Sar))；
wherein, I is the fused image, R, G, B is three-channel data of the remote sensing optical image, sar is SAR satellite data, and Cat () is a splicing function; z () is a normalization function, and the specific formula is:
wherein the method comprises the steps ofx is original image data, mean () is a Mean function, std () is a variance function.
5. The remote sensing image-based farmland building height prediction method according to claim 1, wherein: in step S2), the data enhancement operation includes a calculation process by a random flip function, a random rotation function, and a random clipping function.
6. The remote sensing image-based farmland building height prediction method according to claim 1, wherein: in step S3), the farmland building remote sensing image height estimation network model includes an encoder for downsampling and a decoder for upsampling:
the encoder comprises five feature extraction downsampling blocks, wherein a pooling layer is used for increasing receptive fields at the tail of each block, and the relative positions after pooling are stored, and the encoder comprises 13 convolutions layers and 5 pooling layers;
the decoder comprises five upsampling blocks, wherein the top of each block uses an upsampling layer to amplify the remote sensing image, and the remote sensing image is restored according to the relative position saved in the encoder stage, and the decoder has 5 upsampling layers; the last upsampling block adds a Softmax activation layer to get the same size pixel level height value as the original.
7. A remote sensing image based farmland building height prediction system for implementing the method of any one of claims 1-6, characterized in that: the system comprises a data acquisition module, a data preprocessing module, a farmland building remote sensing image height estimation network model, a verification and evaluation module and a prediction output module;
the data acquisition module is used for: the method comprises the steps of correspondingly forming a training set and a testing set by remote sensing optical images and SAR images of a farmland building and a normalized digital surface model nDSM one by one;
the data preprocessing module is used for: the data fusion module fuses the remote sensing optical image and the SAR image, and the formula is as follows:
I＝Cat(Z(R)，Z(G)，Z(B)，Z(Sar))；
wherein, I is the fused image, R, G, B is three-channel data of the remote sensing optical image, sar is SAR satellite data, and Cat () is a splicing function; z () is a normalization function, and the specific formula is:
wherein the method comprises the steps ofx is original image data, mean () is a Mean function, std () is a variance function;
the data enhancement module: the method comprises the steps of performing calculation processing on image data by using a random overturn function, a random rotation function and a random clipping function;
the farmland building remote sensing image height estimation network model comprises the following steps: comprising an encoder for downsampling and a decoder for upsampling;
the verification and evaluation module: the method comprises the steps of verifying and evaluating a farmland building height estimation task under a remote sensing scene for a farmland building remote sensing image height estimation network model until an evaluation index meets the requirement;
the prediction output module: and the method is used for inputting the remote sensing image of the farmland building to be predicted into the network model for estimating the height of the remote sensing image of the farmland building and outputting the estimated height of the remote sensing image of the farmland building.
8. The farmland building height prediction system according to claim 7, wherein: the encoder comprises five feature extraction downsampling blocks, wherein a pooling layer is used for increasing receptive fields at the tail of each block, and the relative positions after pooling are stored, and the encoder comprises 13 convolutions layers and 5 pooling layers;
the decoder comprises five upsampling blocks, wherein the top of each block uses an upsampling layer to amplify the remote sensing image, and the remote sensing image is restored according to the relative position saved in the encoder stage, and the decoder has 5 upsampling layers; the last upsampling block adds a Softmax activation layer to get the same size pixel level height value as the original.
9. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the method of any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310747563.6A CN116503464B (en) | 2023-06-25 | 2023-06-25 | Farmland building height prediction method based on remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310747563.6A CN116503464B (en) | 2023-06-25 | 2023-06-25 | Farmland building height prediction method based on remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116503464A CN116503464A (en) | 2023-07-28 |
| CN116503464B true CN116503464B (en) | 2023-10-03 |
Family
ID=87325036
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310747563.6A Active CN116503464B (en) | 2023-06-25 | 2023-06-25 | Farmland building height prediction method based on remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116503464B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113902792A (en) * | 2021-11-05 | 2022-01-07 | 长光卫星技术有限公司 | Building height detection method and system based on improved RetinaNet network and electronic equipment |
| CN113902793A (en) * | 2021-11-05 | 2022-01-07 | 长光卫星技术有限公司 | End-to-end building height prediction method and system based on single vision remote sensing image and electronic equipment |
| 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 |
| CN115631412A (en) * | 2022-10-18 | 2023-01-20 | 安徽大学 | Remote sensing image building extraction method based on coordinate attention and data correlation upsampling |
-
2023
- 2023-06-25 CN CN202310747563.6A patent/CN116503464B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113902792A (en) * | 2021-11-05 | 2022-01-07 | 长光卫星技术有限公司 | Building height detection method and system based on improved RetinaNet network and electronic equipment |
| CN113902793A (en) * | 2021-11-05 | 2022-01-07 | 长光卫星技术有限公司 | End-to-end building height prediction method and system based on single vision remote sensing image and electronic equipment |
| 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 |
| CN115631412A (en) * | 2022-10-18 | 2023-01-20 | 安徽大学 | Remote sensing image building extraction method based on coordinate attention and data correlation upsampling |
Non-Patent Citations (3)
| Title |
|---|
| Chao-Jung Liu.IM2ELEVATION: Building Height Estimation from Single-View Aerial Imagery.《MDPI》.2022,第1-22页. * |
| 基于遥感的建筑物高度快速提取研究综述;钱瑶 等;《生态学报》;第35卷(第12期);第3886-3895页 * |
| 陈亚雄.一种多尺度平衡深度哈希图像检索方法.计算机应用研究.第36卷(第2期),第622-629页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116503464A (en) | 2023-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113780296B (en) | Remote sensing image semantic segmentation method and system based on multi-scale information fusion | |
| CN113657388B (en) | Image semantic segmentation method for super-resolution reconstruction of fused image | |
| CN111598778B (en) | Super-resolution reconstruction method for insulator image | |
| CN113706482A (en) | High-resolution remote sensing image change detection method | |
| CN104217404A (en) | Video image sharpness processing method in fog and haze day and device thereof | |
| CN113205520B (en) | Method and system for semantic segmentation of image | |
| CN114332385A (en) | Monocular camera target detection and spatial positioning method based on three-dimensional virtual geographic scene | |
| CN115147731A (en) | SAR image target detection method based on full-space coding attention module | |
| CN115359195B (en) | Method and device for generating orthophoto, storage medium and electronic equipment | |
| CN114758337A (en) | Semantic instance reconstruction method, device, equipment and medium | |
| CN115293992B (en) | Polarization image defogging method and device based on unsupervised weight depth model | |
| Xu et al. | Building height calculation for an urban area based on street view images and deep learning | |
| CN116524189A (en) | High-resolution remote sensing image semantic segmentation method based on coding and decoding indexing edge characterization | |
| CN116912675B (en) | Underwater target detection method and system based on feature migration | |
| CN116503464B (en) | Farmland building height prediction method based on remote sensing image | |
| CN110751699B (en) | Color reconstruction method of optical remote sensing image based on convolutional neural network | |
| CN117422619A (en) | Training method of image reconstruction model, image reconstruction method, device and equipment | |
| CN115496788A (en) | Deep completion method using airspace propagation post-processing module | |
| Xu | [Retracted] Application of Remote Sensing Image Data Scene Generation Method in Smart City | |
| CN115439738A (en) | Underwater target detection method based on self-supervision cooperative reconstruction | |
| CN108416815A (en) | Assay method, equipment and the computer readable storage medium of air light value | |
| CN113744152A (en) | Tide water image denoising processing method, terminal and computer readable storage medium | |
| CN113902744A (en) | Image detection method, system, equipment and storage medium based on lightweight network | |
| CN114372987A (en) | Sub-pixel flood inundation mapping method based on irregular area spatial spectrum information | |
| CN113192204A (en) | Three-dimensional reconstruction method of building in single inclined remote sensing image |
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 |