CN111738201B - Method and system for extracting remote sensing image of woodland based on region-of-interest network - Google Patents
Method and system for extracting remote sensing image of woodland based on region-of-interest network Download PDFInfo
- Publication number
- CN111738201B CN111738201B CN202010618625.XA CN202010618625A CN111738201B CN 111738201 B CN111738201 B CN 111738201B CN 202010618625 A CN202010618625 A CN 202010618625A CN 111738201 B CN111738201 B CN 111738201B
- Authority
- CN
- China
- Prior art keywords
- image
- remote sensing
- forest land
- forest
- 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 37
- 238000000605 extraction Methods 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 48
- 238000007781 pre-processing Methods 0.000 claims abstract description 18
- 238000005457 optimization Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 8
- 238000005520 cutting process Methods 0.000 claims description 6
- 238000013135 deep learning Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 8
- 239000000284 extract Substances 0.000 description 7
- 238000005070 sampling Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000011835 investigation Methods 0.000 description 4
- 230000011218 segmentation Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000007797 corrosion Effects 0.000 description 2
- 238000005260 corrosion Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000010339 dilation Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000003064 k means clustering Methods 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 244000025254 Cannabis sativa Species 0.000 description 1
- PNEYBMLMFCGWSK-UHFFFAOYSA-N aluminium oxide Inorganic materials [O-2].[O-2].[O-2].[Al+3].[Al+3] PNEYBMLMFCGWSK-UHFFFAOYSA-N 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a method and a system for extracting remote sensing images of a wooded area based on a region-of-interest network. Acquiring forest land survey data and multispectral wide-coverage image data; preprocessing the forest land survey data and the multispectral wide-coverage image data to obtain preprocessed forest land survey data and preprocessed multispectral wide-coverage image data; obtaining a sample set based on a geographic space corresponding relation according to the preprocessed forest land survey data and the preprocessed multispectral wide coverage image data; constructing a region-of-interest network; training the network of the region of interest according to the sample set to obtain the trained network of the region of interest; acquiring a remote sensing image to be extracted; inputting the remote sensing image to be extracted into the trained region-of-interest network to obtain an extracted image of the forest land; and optimizing the image to obtain an optimized extracted image of the forest land. The method can ensure that the woodland has higher extraction precision in images of different time phases.
Description
Technical Field
The invention relates to the field of remote sensing image extraction of a woodland, in particular to a method and a system for extracting a remote sensing image of a woodland based on a region-of-interest network.
Background
The forest land/non-forest land information has important significance for forest resource investigation, investigation planning and design and other forestry remote sensing applications. Aiming at the requirements of forest resource investigation business application in China on multispectral, high spatial resolution, high-frequency imaging and large-area coverage of the remote sensing image, a forest land/non-forest land rapid identification project of the remote sensing image is developed, forest resource investigation efficiency is improved, forest resources are conveniently planned and designed, and the remote sensing image can serve for forest development and ecological environment construction in China.
The analysis of the existing forest land extraction method shows that the high accuracy is difficult to ensure at the same time of rapidness. At present, the rapid extraction of the forest land is mainly realized by improving and accelerating key algorithms such as image noise reduction, feature extraction, clustering and the like in the process of extracting the forest land. For example, Sharifi et al (2015) adopt a fast independent principal component analysis method to perform noise reduction on the SAR image, so that the woodland extraction precision is guaranteed, and the efficiency is improved; in the aspect of feature extraction, most researchers currently extract forest land information from texture features contained in high-resolution images (Dian et al, 2015), and adding texture features can avoid the problem of mixed division of forest lands and cultivated lands to a certain extent, however, most of texture extraction uses a sliding window as a unit to extract forest land textures pixel by pixel, and texture analysis pixel by pixel makes the calculation amount too large and the time consumption too long, so that the efficiency of forest land extraction is difficult to guarantee. Huhualong et al (2016) propose a forest land boundary extraction method based on texture primitive merging, and through carrying out experiments on multi-source high-resolution images, the advantages of the extraction method in comparison with the traditional method in extraction precision and calculation efficiency are verified. The alundum ice (2013) clusters the images by adopting an iterative unsupervised classification method, and extracts the forest lands according to the interpreted sample points at the later stage, and the result shows that the method is suitable for large-scale forest land identification and is a quick and effective method. Boukir et al (2012) propose a fast Mean Shift algorithm (PAMS) for the problem of large Mean Shift clustering calculation amount, compared with the traditional Mean Shift, the operation efficiency of the algorithm is improved by about 5 times, the algorithm is equivalent to the K-means clustering efficiency, unsupervised woodland mapping can be realized, and compared with the K-means clustering, the algorithm has more application advantages due to the fact that the clustering number needs to be manually set.
Generally, the emphasis of forest land extraction is on identification accuracy, the attention to identification efficiency is low, and with the development of high-performance calculation, the rapid identification of large-area forest lands applying a GPU and a multi-core CPU in the future is a necessary trend. In addition, most of the existing forest land identification methods are only suitable for small-range areas, the application and popularization capabilities are poor, the forest lands are easily affected by terrain and seasonal factors, and the fast and accurate identification of the forest lands in different areas can be realized only by fully mining invariant features of the forest lands under different conditions. Deep learning currently achieves better results in large-scale target detection and classification applications (Ren et al, 2016), shows good potential for recognizing surface features in complex environments, and the feature extraction and classification modes thereof can be used for reference in the project.
In recent years, deep learning methods have been rapidly developed in the field of remote sensing image target extraction due to high accuracy and high speed. Since the idea of Deep Learning (Deep Learning) was proposed by Hinton in 2006 (Hinton,2006), Deep Learning models represented by Deep Belief Networks (DBN) and Deep Convolutional Neural Networks (DCNN) have gained remarkable results in applications such as character recognition, speech signal processing, and image understanding, while research on applications of Deep Learning in remote sensing image classification is underway (Zhang et al, 2015), and research related to remote sensing image classification is mainly focused on two aspects of remote sensing image feature Learning and classification model construction.
The existing classification research based on deep learning mostly focuses on hyperspectral image classification, and possible reasons are that open-source hyperspectral classification data sets such as Pavia are easy to obtain, and a small amount of research based on optical images (Hung, 2014; Pentiti, 2015) adopts open-source data sets such as UCMerced or establishes special data sets.
Hu et al (2015) designs a DCNN with a 5-layer structure and applies the DCNN to hyperspectral remote sensing image classification research based on spectral features; xing et al (2016) classify hyperspectral images using SDAE; chen (2014), Yue and the like (2015) extract main component information in the neighborhood from the main component transformation image as spatial information of the pixel, combine the spatial information with spectral information of the pixel, respectively extract abstract features by using SAE and DCNN, and finally classify the hyperspectral image by using logistic regression. Zhao et al (2015) extracted global and robust features from hyperspectral images using a Multi-scale Convolutional Auto-Encoder (MCAE) and classified images using logistic regression, for the problem that DCNN spatial receptive field is fixed and Multi-scale spatial information cannot be extracted. Lv et al (2015) performed urban land cover and land use classification of polarized SAR images based on DBN. Hung et al (2014) utilize sparse self-encoders for feature learning and invasive grass seed identification for very high resolution UAV imagery. The Frank-rabeye studio at the university of kanachi merlon has recently created a visual search tool Terrapattern for satellite images that uses deep convolutional neural networks to assist in image recognition, referred to as a "visual search engine for a satellite image". In terms of data sets, DigitalGlobe has opened its online satellite data download platform, SpaceNet, which is the first publicly released satellite imagery data platform of high resolution, dedicated to training machine learning algorithms worldwide. It is conceivable that with the help of satellite images of classification marks issued by a platform, an image recognition and analysis algorithm based on the satellite images will make new progress, but at present, a set of standard data sets are not established for forest lands/non-forest lands, and due to the characteristics of different shapes and uneven distribution of the forest lands, a method particularly suitable for extracting the forest lands is not provided at present.
Disclosure of Invention
The invention aims to provide a method and a system for extracting remote sensing images of a woodland based on a region-of-interest network, which can ensure that the woodland has higher extraction precision in images at different time phases.
In order to achieve the purpose, the invention provides the following scheme:
a method for extracting remote sensing images of wooded land based on a region of interest network comprises the following steps:
acquiring forest land survey data and multispectral wide-coverage image data;
performing data preprocessing on the forest land survey data and the multispectral wide-coverage image data to obtain preprocessed forest land survey data and preprocessed multispectral wide-coverage image data;
obtaining a sample set based on a geographic space corresponding relation according to the preprocessed forest land survey data and the preprocessed multispectral wide coverage image data;
constructing a region-of-interest network;
training the interested area network according to the sample set to obtain the trained interested area network;
acquiring a remote sensing image to be extracted;
inputting the remote sensing image to be extracted into the trained region-of-interest network to obtain an extracted image of the woodland;
and optimizing the extracted image of the forest land to obtain the optimized extracted image of the forest land.
Optionally, the preprocessing the forest land survey data to obtain preprocessed forest land survey data specifically includes:
and performing forest area survey data cutting, survey data and image registration, survey data type merging, survey data rasterization and image radiometric calibration processing on the forest area survey data and the multispectral wide coverage image to obtain preprocessed forest area survey data.
Optionally, the training the network of interest according to the sample set to obtain a trained network of interest specifically includes:
dividing the sample set into a plurality of multispectral remote sensing images and forest land survey marked pictures;
and taking each multispectral remote sensing image as input, taking a forest area survey marked picture corresponding to each multispectral remote sensing image as standard output, comparing the output of the interested area network with the standard output during training, adjusting the network parameters of the interested area, and training to obtain the trained interested area network.
Optionally, the optimizing the extracted image of the existing forest land to obtain an optimized extracted image of the existing forest land specifically includes:
and splicing the extracted images of the forested areas, and performing boundary optimization on the overall result by using a mathematical morphology method to obtain the optimized extracted images of the forested areas.
A remote sensing image extraction system of woodland based on area of interest network includes:
the data acquisition module is used for acquiring forest area survey data and multispectral wide-coverage image data;
the data preprocessing module is used for preprocessing the forest land survey data and the multispectral wide coverage image data to obtain preprocessed forest land survey data and preprocessed multispectral wide coverage image data;
the sample set determining module is used for obtaining a sample set according to the preprocessed forest land survey data and the preprocessed multispectral wide-coverage image data based on a geographic space corresponding relation;
the interesting area network model building module is used for building an interesting area network;
the interesting area network model training module is used for training the interesting area network according to the sample set to obtain the trained interesting area network;
the remote sensing image acquisition module is used for acquiring a remote sensing image to be extracted;
the forest land extraction image determining module is used for inputting the remote sensing image to be extracted into the trained region-of-interest network to obtain a forest land extraction image;
and the forest land extraction image optimization module is used for optimizing the forest land extraction image to obtain an optimized forest land extraction image.
Optionally, the data preprocessing module specifically includes:
and the data preprocessing unit is used for performing forest land survey data cutting, survey data and image registration, survey data type merging, survey data rasterization and image radiometric calibration processing on the forest land survey data and the multispectral wide coverage image to obtain preprocessed forest land survey data.
Optionally, the training module of the network model of the region of interest specifically includes:
the system comprises a sample set dividing unit, a forest survey marking unit and a multi-spectral remote sensing image acquiring unit, wherein the sample set dividing unit is used for dividing a sample set into a plurality of multispectral remote sensing images and forest survey marked pictures;
and the interesting region network model training unit is used for taking each multispectral remote sensing image as input, taking a forest area survey marked picture corresponding to each multispectral remote sensing image as standard output, comparing the output of the interesting region network with the standard output during training, adjusting interesting region network parameters, and training to obtain the trained interesting region network.
Optionally, the woodland extraction image optimization module specifically includes:
and the woodland extraction image optimization unit is used for splicing the woodland extraction images and performing boundary optimization on the overall result by using a mathematical morphology method to obtain an optimized woodland extraction image.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the method uses the interesting area network to automatically extract the woodland in the remote sensing image, the interesting area network (POI-Net) is a neural network for extracting the characteristics of the interesting position based on deep learning, and the interesting area network is extracted aiming at the woodland in the remote sensing image.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a method for extracting remote sensing images of a wooded area based on a network of interest (ROI);
FIG. 2 is a schematic diagram of a network structure of a region of interest;
FIG. 3 is a schematic diagram of a cascaded residual fusion module;
FIG. 4 is a schematic diagram of a network training of a region of interest;
fig. 5 is a structural diagram of the remote sensing image extraction system based on the region of interest network.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a flow chart of the method for extracting remote sensing images of woodland based on the network of interest. As shown in fig. 1, a method for extracting a remote sensing image of a woodland based on a network of interest (roi) comprises:
step 101: forest region survey data and multispectral wide-coverage image data are obtained.
Step 102: performing data preprocessing on the forest land survey data and the multispectral wide-coverage image data to obtain preprocessed forest land survey data and preprocessed multispectral wide-coverage image data, and specifically comprising:
and performing forest area survey data cutting, survey data and image registration, survey data type merging, survey data rasterization and image radiometric calibration processing on the forest area survey data and the multispectral wide coverage image to obtain preprocessed forest area survey data.
Step 103: and obtaining a sample set based on the geographic space corresponding relation according to the preprocessed forest land survey data and the preprocessed multispectral wide coverage image data.
Based on the rasterized survey data and the image data processed in the step 1, a plurality of 256 × 256 pixel sample pairs are generated according to the geographic space corresponding relationship, and each pair comprises a picture representing the multispectral remote sensing image and a picture representing the forest survey land type mark.
The sample set is made based on forest land survey data and multispectral wide-coverage remote sensing images. The forest land survey data need to be actually surveyed to obtain the vegetation condition of the region and integrate the vegetation condition into a professional marking sample set. After the marked sample set is obtained, the multispectral wide-coverage remote sensing image and the marked data are registered through geographic information. Next, the labeled images are classified into a plurality of categories according to the definition of the land category in "national forest land change survey technical project 2011 edition". And finally, dividing each image material according to a fixed size, screening out an image pair with better quality, and obtaining a final sample set. It is recommended that the size of the remote sensing image and the size of the marked image are both 256 × 256 and the pixel positions correspond to each other.
Step 104: and constructing a region of interest network.
Area of interest network (POI-Net) introduction:
the network of the region of interest is an end-to-end neural network based on deep learning, belongs to one of image semantic segmentation models, is an image segmenter based on a convolutional neural network and used for extracting the characteristics of the position of interest, and can continuously explore effective position information of a target during up-sampling and down-sampling in the model training process. Fig. 2 is a schematic diagram of a network structure of a region of interest.
And the down-sampling stage of the interested area network consists of successive layers of descending blocks, and initialization weight values are added. The up-sampling stage adds cascade residual fusion modules which can combine the extracted high-order features with the low-order features. In the up-sampling stage, the up-sampling operation is carried out on the smaller feature graph, the size of the feature graph is expanded to be the same as that of other larger primary images, and after a convolution layer is carried out, the feature graph and the convolution layer are spliced and fused. They are then passed through a cascade of residual building blocks, where crossing the pass path results in different feature fields, thereby generating features at different levels of abstraction, fig. 3 is a schematic diagram of a cascade of residual fusion modules.
Step 105: training the network of the region of interest according to the sample set to obtain the trained network of the region of interest, specifically comprising:
dividing the sample set into a plurality of multispectral remote sensing images and forest land survey marked pictures;
and taking each multispectral remote sensing image as input, taking a forest area survey marked picture corresponding to each multispectral remote sensing image as standard output, comparing the output of the interested area network with the standard output during training, adjusting the network parameters of the interested area, and training to obtain the trained interested area network.
After the area-of-interest network used is determined and a woodland/non-woodland sample library is prepared, training of the area-of-interest network (POI-Net) is started. Fig. 4 is a schematic diagram of training a network of interest. As shown in fig. 4, the region-of-interest network is loaded first, then the training times are specified, a prepared sample library needs to be input in each round of training, the region-of-interest network extracts forest land features of the remote sensing images according to the relationship between the remote sensing images and corresponding forest land survey marks, and then parameters of the region-of-interest network are adjusted, and finally the trained region-of-interest network is obtained.
Through multiple times of training and selection of a proper model, an interested area network with the best image segmentation effect, namely the interested area network, can be obtained.
After sample pairs (such as 300) and training times (such as 10000) required by the model are set, the interesting area network model is automatically trained, parameters in the model are continuously updated and iterated during the training, and after the training times are reached, the training is finished, and at this time, the parameters of the interesting area network are adjusted. And training for multiple times by adjusting the training times, and selecting the parameter with the highest extraction accuracy in the woodland for storage. The number of training times is related to the number and quality of the sample pairs in the training set, and usually, 3000 iterations are performed according to every 100 sample pairs, and if the texture of the samples is complex, the number of training times is appropriately increased.
Step 106: and acquiring a remote sensing image to be extracted.
Preparing a remote sensing image to be extracted, dividing the image into a plurality of image blocks with proper sizes (such as 256 multiplied by 256 pixel sizes) and sequencing.
Step 107: and inputting the remote sensing image to be extracted into the trained region-of-interest network to obtain the extracted image of the forest land.
And sequentially taking out the image blocks in sequence, and inputting the image blocks into the interested area network with complete parameters obtained by model training. Because the network of the region of interest is an end-to-end network, after the remote sensing image is input, the network performs semantic segmentation on the image according to the parameters obtained in the step 105, and outputs the segmentation result of the image. The segmentation result is an extracted image of the wooded area with the same size as the remote sensing image, wherein pixels corresponding to the wooded area are marked as a specific value (such as 1) to distinguish the non-wooded area.
In the image segmentation model based on deep learning, the trained images are consistent in size, that is, only the images in the sample library are consistent in size, so that the images are required to be fixed in size when being detected, that is, only the remote sensing images with the specified size can be segmented and detected every time.
Therefore, for the multispectral wide-coverage image, the whole to-be-detected image needs to be divided into a plurality of image blocks to be placed in an interested area network, the interested area network sequentially predicts all the image blocks, and finally, the predicted extracted image blocks of the forest land are spliced, so that the whole forest land extraction result of the remote sensing image can be obtained.
Step 108: optimizing the extracted image of the forest land to obtain an optimized extracted image of the forest land, which specifically comprises the following steps:
and splicing the extracted images of the forested areas, and performing boundary optimization on the overall result by using a mathematical morphology method to obtain the optimized extracted images of the forested areas.
The overall result image optimization method of mathematical morphology comprises the following steps:
the basic operations of mathematical morphology include "erosion" and "dilation", open and closed operations. After obtaining a plurality of image blocks and using the region of interest network to extract the forest land, the extraction result can be further processed. Firstly, performing mathematical morphology 'corrosion' operation on an image woodland extraction result, performing convolution by using an image and an inner core, sliding the inner core over an image block, and extracting the minimum pixel value of an inner core coverage area and replacing the pixel at an anchor point position. Then, a "dilation" operation is performed to sweep the kernel through the image, extract the maximum pixel value of the kernel coverage area, and replace the pixel at the anchor point location.
Through the two operations, the edge of the extracted result of the woodland can be smoothed, and the purpose of reducing the edge connection of the edge window is achieved. In addition, noise in the image can be eliminated through 'corrosion' and 'expansion' operations, so that the visual effect of the extracted result is better.
Compared with the prior art, the invention has the following advantages:
1. compared with other methods, the remote sensing image woodland extraction method has higher accuracy. The trained region-of-interest network is used, and can be fitted more quickly compared with other deep learning neural networks, and the network enhances the transfer of necessary characteristics between layers. The problem of the precision of forest land extraction descends when the terrain is broken finely and the land type change is large in the ordinary deep learning neural network is solved.
2. The result optimization speed of forest land extraction is faster than that of other computer image processing modes. While a single neural network is used, the woodland extraction result is optimized by using a mathematical morphology method. On the premise of ensuring the woodland extraction efficiency, the method of combining the mathematical morphology in consideration of the information continuity of the image relieves the problem that the wide-field image generates the edge connection of the image block edge, and reduces the noise generated in the result image.
Fig. 5 is a structural diagram of the remote sensing image extraction system based on the region of interest network. As shown in fig. 5, a remote sensing image extraction system based on a region of interest network includes:
the data acquisition module 201 is configured to acquire forest survey data and multispectral wide-coverage image data.
The data preprocessing module 202 is configured to perform data preprocessing on the forest land survey data and the multispectral wide-coverage image data to obtain preprocessed forest land survey data and preprocessed multispectral wide-coverage image data.
And the sample set determining module 203 is configured to obtain a sample set based on a geographic spatial correspondence relationship according to the preprocessed forest land survey data and the preprocessed multispectral wide-coverage image data.
And the region-of-interest network model building module 204 is used for building a region-of-interest network.
And the interesting area network model training module 205 is configured to train the interesting area network according to the sample set, so as to obtain an interesting area network after training.
The to-be-extracted remote sensing image acquisition module 206 is configured to acquire a remote sensing image to be extracted.
And the forest land extraction image determining module 207 is used for inputting the remote sensing image to be extracted into the trained region of interest network to obtain a forest land extraction image.
And the woodland extraction image optimization module 208 is used for optimizing the woodland extraction image to obtain an optimized woodland extraction image.
The data preprocessing module 202 specifically includes:
and the data preprocessing unit is used for performing forest land survey data cutting, survey data and image registration, survey data type merging, survey data rasterization and image radiometric calibration processing on the forest land survey data and the multispectral wide coverage image to obtain preprocessed forest land survey data.
The region-of-interest network model training module 205 specifically includes:
and the sample set dividing unit is used for dividing the sample set into a plurality of multispectral remote sensing images and forest land survey marked pictures.
And the interesting region network model training unit is used for taking each multispectral remote sensing image as input, taking a forest area survey marked picture corresponding to each multispectral remote sensing image as standard output, comparing the output of the interesting region network with the standard output during training, adjusting interesting region network parameters, and training to obtain the trained interesting region network.
The woodland extraction image optimization module 208 specifically includes:
and the woodland extraction image optimization unit is used for splicing the woodland extraction images and performing boundary optimization on the overall result by using a mathematical morphology method to obtain an optimized woodland extraction image.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (6)
1. A method for extracting remote sensing images of wooded land based on a network of interest area is characterized by comprising the following steps:
acquiring forest land survey data and multispectral wide-coverage image data;
performing data preprocessing on the forest land survey data and the multispectral wide-coverage image data to obtain preprocessed forest land survey data and preprocessed multispectral wide-coverage image data;
obtaining a sample set based on a geographic space corresponding relation according to the preprocessed forest land survey data and the preprocessed multispectral wide coverage image data;
constructing a region-of-interest network;
training the interested area network according to the sample set to obtain the trained interested area network;
acquiring a remote sensing image to be extracted;
inputting the remote sensing image to be extracted into the trained region-of-interest network to obtain an extracted image of the woodland;
optimizing the extracted image of the forest land to obtain an optimized extracted image of the forest land;
the training of the network of the region of interest according to the sample set to obtain the trained network of the region of interest specifically includes:
dividing the sample set into a plurality of multispectral remote sensing images and forest land survey marked pictures;
and taking each multispectral remote sensing image as input, taking a forest area survey marked picture corresponding to each multispectral remote sensing image as standard output, comparing the output of the interested area network with the standard output during training, adjusting the network parameters of the interested area, and training to obtain the trained interested area network.
2. The method for extracting remote sensing images of woodland based on area-of-interest network according to claim 1, wherein the preprocessing the data of forest land survey to obtain preprocessed forest land survey data specifically comprises:
and performing forest area survey data cutting, survey data and image registration, survey data type merging, survey data rasterization and image radiometric calibration processing on the forest area survey data and the multispectral wide coverage image to obtain preprocessed forest area survey data.
3. The method for extracting remote sensing images of forest lands based on the area-of-interest network according to claim 1, wherein the method for optimizing the extracted images of forest lands to obtain the optimized extracted images of forest lands specifically comprises:
and splicing the extracted images of the forested areas, and performing boundary optimization on the overall result by using a mathematical morphology method to obtain the optimized extracted images of the forested areas.
4. A remote sensing image extraction system of woodland based on area of interest network, characterized by that, includes:
the data acquisition module is used for acquiring forest area survey data and multispectral wide-coverage image data;
the data preprocessing module is used for preprocessing the forest land survey data and the multispectral wide coverage image data to obtain preprocessed forest land survey data and preprocessed multispectral wide coverage image data;
the sample set determining module is used for obtaining a sample set according to the preprocessed forest land survey data and the preprocessed multispectral wide-coverage image data based on a geographic space corresponding relation;
the interesting area network model building module is used for building an interesting area network;
the interesting area network model training module is used for training the interesting area network according to the sample set to obtain the trained interesting area network;
the remote sensing image acquisition module is used for acquiring a remote sensing image to be extracted;
the forest land extraction image determining module is used for inputting the remote sensing image to be extracted into the trained region-of-interest network to obtain a forest land extraction image;
the forest land extraction image optimization module is used for optimizing the forest land extraction image to obtain an optimized forest land extraction image;
the training module for the network model of the region of interest specifically comprises:
the system comprises a sample set dividing unit, a forest survey marking unit and a multi-spectral remote sensing image acquiring unit, wherein the sample set dividing unit is used for dividing a sample set into a plurality of multispectral remote sensing images and forest survey marked pictures;
and the interesting area network model training unit is used for taking each multispectral remote sensing image as input, taking the forest area survey marked picture corresponding to each multispectral remote sensing image as standard output, comparing the output of the interesting area network with the standard output during training, adjusting the interesting area network parameters, and training to obtain the trained interesting area network.
5. The remote sensing image extraction system based on the region of interest network with the woodland according to claim 4, wherein the data preprocessing module specifically comprises:
and the data preprocessing unit is used for performing forest land survey data cutting, survey data and image registration, survey data type merging, survey data rasterization and image radiometric calibration processing on the forest land survey data and the multispectral wide coverage image to obtain preprocessed forest land survey data.
6. The remote sensing image extraction system based on the forest land of the region of interest network according to claim 4, wherein the image optimization module for extracting the forest land specifically comprises:
and the woodland extraction image optimization unit is used for splicing the woodland extraction images and performing boundary optimization on the overall result by using a mathematical morphology method to obtain an optimized woodland extraction image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010618625.XA CN111738201B (en) | 2020-07-01 | 2020-07-01 | Method and system for extracting remote sensing image of woodland based on region-of-interest network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010618625.XA CN111738201B (en) | 2020-07-01 | 2020-07-01 | Method and system for extracting remote sensing image of woodland based on region-of-interest network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111738201A CN111738201A (en) | 2020-10-02 |
| CN111738201B true CN111738201B (en) | 2022-08-09 |
Family
ID=72653954
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010618625.XA Active CN111738201B (en) | 2020-07-01 | 2020-07-01 | Method and system for extracting remote sensing image of woodland based on region-of-interest network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111738201B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115082793B (en) * | 2022-06-28 | 2023-02-17 | 生态环境部卫星环境应用中心 | Method and device for rapidly investigating space background condition of forest and grass in water source area |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543630A (en) * | 2018-11-28 | 2019-03-29 | 苏州中科天启遥感科技有限公司 | Remote sensing image forest land extracting method and system, storage medium, electronic equipment based on deep learning |
| CN110136194A (en) * | 2019-05-21 | 2019-08-16 | 吉林大学 | Snow Cover measuring method based on satellite-borne multispectral remotely-sensed data |
-
2020
- 2020-07-01 CN CN202010618625.XA patent/CN111738201B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543630A (en) * | 2018-11-28 | 2019-03-29 | 苏州中科天启遥感科技有限公司 | Remote sensing image forest land extracting method and system, storage medium, electronic equipment based on deep learning |
| CN110136194A (en) * | 2019-05-21 | 2019-08-16 | 吉林大学 | Snow Cover measuring method based on satellite-borne multispectral remotely-sensed data |
Non-Patent Citations (1)
| Title |
|---|
| Yuanyuan Gui等.Woodland_Detection_Using_Most-Sure_Strategy_to_Fuse_Segmentation_Results_of_Deep_Learning.《IEEE Xplore》.2019,第6724-6727页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111738201A (en) | 2020-10-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113011427B (en) | Remote sensing image semantic segmentation method based on self-supervision contrast learning | |
| Wang et al. | Scene classification of high-resolution remotely sensed image based on ResNet | |
| CN111598174B (en) | Model training method based on semi-supervised antagonistic learning and image change analysis method | |
| Shi et al. | Road detection from remote sensing images by generative adversarial networks | |
| Kampffmeyer et al. | Semantic segmentation of small objects and modeling of uncertainty in urban remote sensing images using deep convolutional neural networks | |
| Osco et al. | The segment anything model (sam) for remote sensing applications: From zero to one shot | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| Ghassemi et al. | Learning and adapting robust features for satellite image segmentation on heterogeneous data sets | |
| CN110633708A (en) | Deep network significance detection method based on global model and local optimization | |
| Lu et al. | P_SegNet and NP_SegNet: New neural network architectures for cloud recognition of remote sensing images | |
| Zuo et al. | HF-FCN: Hierarchically fused fully convolutional network for robust building extraction | |
| Shen et al. | Biomimetic vision for zoom object detection based on improved vertical grid number YOLO algorithm | |
| Xing et al. | Traffic sign recognition using guided image filtering | |
| Marochov et al. | Image classification of marine-terminating outlet glaciers in Greenland using deep learning methods | |
| Li et al. | Using a semantic edge-aware multi-task neural network to delineate agricultural parcels from remote sensing images | |
| CN111104850A (en) | Remote sensing image building automatic extraction method and system based on residual error network | |
| Huang et al. | Attention-guided label refinement network for semantic segmentation of very high resolution aerial orthoimages | |
| Jiang et al. | Arbitrary-shaped building boundary-aware detection with pixel aggregation network | |
| Cheng et al. | Multi-scale Feature Fusion and Transformer Network for urban green space segmentation from high-resolution remote sensing images | |
| Sathyanarayanan et al. | A multiclass deep learning approach for LULC classification of multispectral satellite images | |
| Xia et al. | Submesoscale oceanic eddy detection in SAR images using context and edge association network | |
| Hu et al. | Supervised multi-scale attention-guided ship detection in optical remote sensing images | |
| Chu et al. | Glacier extraction based on high-spatial-resolution remote-sensing images using a deep-learning approach with attention mechanism | |
| Yadav et al. | Geological information extraction from satellite imagery using machine learning | |
| CN111738201B (en) | Method and system for extracting remote sensing image of woodland based on region-of-interest network |
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 |