CN108830870B - Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning - Google Patents
Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning Download PDFInfo
- Publication number
- CN108830870B CN108830870B CN201810491945.6A CN201810491945A CN108830870B CN 108830870 B CN108830870 B CN 108830870B CN 201810491945 A CN201810491945 A CN 201810491945A CN 108830870 B CN108830870 B CN 108830870B
- Authority
- CN
- China
- Prior art keywords
- farmland
- edge
- scale
- segmentation
- contour
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 36
- 230000011218 segmentation Effects 0.000 claims abstract description 62
- 238000003708 edge detection Methods 0.000 claims abstract description 37
- 238000000034 method Methods 0.000 claims abstract description 26
- 230000004927 fusion Effects 0.000 claims abstract description 15
- 238000005457 optimization Methods 0.000 claims abstract description 11
- 230000009466 transformation Effects 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 8
- 238000007637 random forest analysis Methods 0.000 claims description 7
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000012706 support-vector machine Methods 0.000 claims description 5
- 230000035945 sensitivity Effects 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000002156 mixing Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims 1
- 230000003595 spectral effect Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 238000001514 detection method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 239000000575 pesticide Substances 0.000 description 2
- 238000009331 sowing Methods 0.000 description 2
- 238000005507 spraying Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 241000607479 Yersinia pestis Species 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a satellite image high-precision farmland boundary extraction method based on multi-scale structure learning, which comprises the following steps of: step 1, edge detection based on structure learning: after a high-resolution satellite image is input, edge detection is firstly carried out on the satellite image based on local structural pattern learning, and an edge strength value of each pixel point in the satellite image on a specific scale is obtained; step 2, multi-scale edge information fusion; step 3, multi-scale segmentation based on the hypermetrological contour map: converting multi-scale edge information into bottom layer segmentation block information through watershed transformation; analyzing the contour intensity between adjacent segmentation blocks by utilizing the hypermetrological contour map to combine the segmentation blocks layer by layer to obtain segmentation blocks with different scales; step 4, farmland boundary optimization based on semantic modeling: the spatial distribution positions of the farmland blocks are modeled by utilizing the conditional random field, so that the usability of the farmland boundary is improved. The method can realize accurate extraction of the farmland boundary.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a satellite image high-precision farmland boundary extraction method based on multi-scale structure learning.
Background
With the rapid development of satellite remote sensing technology, sensing technology and high-precision positioning technology, high-resolution satellite remote sensing has the advantages of large image coverage, detection in high-risk areas, low cost, high resolution, more stock data and the like, and meanwhile, due to the continuous formation of satellite/small satellite constellations, the revisit time of the satellite remote sensing to the same area is shorter and shorter, and some satellites even have the revisit capability of 1 to 2 days. The satellite image can realize large-area dynamic monitoring under medium and low resolution, can realize small-range accurate detection under high resolution, and is widely applied to various fields of military and civil at home and abroad. With the massive acquisition of the satellite remote sensing data, the intelligent and precise agricultural application becomes a hot spot and a trend, such as intelligent pest and disease monitoring, crop growth monitoring, pesticide spraying, unmanned sowing and the like, and how to extract spatial distribution elements of farmlands and crops from the massive high-resolution satellite remote sensing data becomes a key technology for restricting the development of precise agriculture.
At present, the boundary of a farmland block (an independent farmland area with obvious edges in satellite images) required in the pesticide spraying and unmanned sowing processes of an unmanned aerial vehicle is basically extracted in a manual mode, the efficiency is low, the cost is high, and centralized management and updating cannot be realized due to scattered data distribution. Therefore, the automatic and accurate extraction of the boundary of the farmland block is a key technology for precise agricultural application. In recent years, computer vision and machine learning have been rapidly developed, wherein methods of image ground object classification (road classification) and semantic segmentation (semantic segmentation, which means that all pixels in an image are labeled to obtain a plurality of homogeneity regions, so that all pixels in each region belong to the same type of ground object or target) are diversified. However, the existing image classification and semantic segmentation methods can only distinguish different types of ground feature types (scenes or target types on the ground, such as buildings, roads, farmlands, woodlands, and the like) or targets, and cannot well realize the segmentation and boundary extraction of a single farmland block. Therefore, in practical application, a rapid and effective farmland boundary accurate extraction method and scheme for a high-resolution satellite image are urgently needed, and a farmland block boundary with complete semantics and accuracy can be provided. In recent two years, as an extension of semantic segmentation, instance segmentation (instance segmentation, which is a further extension of image semantic segmentation) for single object extraction is beginning to attract industrial attention, and instance segmentation can distinguish different object individuals belonging to the same type (such as different adjacent farmland blocks, two adjacent buildings, etc.) and corresponding structure learning (structure learning) technologies while distinguishing different land and object types (such as farmlands and buildings).
The technical scheme close to the invention is as follows:
1) a method and a device for extracting a ground feature of a high-resolution image (publication No. CN106951877A) define three types of spectrum-space features of the high-resolution image to be extracted according to a spectrum feature image and three types of multi-scale feature images of the high-resolution image to be extracted: the method comprises the following steps of firstly, preliminarily classifying the three types of spectrum-space characteristics by using a support vector machine, and then further distinguishing the ground objects by using a probability fusion method according to a classification result.
2) A high-resolution image residential area extraction method based on spatial correlation and heterogeneity characteristics (publication number CN106529430A) comprises performing false color synthesis on multiband original images to obtain false color images; selecting a ground object sample from the false color image, and calculating spectral characteristic curves of various ground objects, wherein the spectral values corresponding to the spectral characteristic wave bands are spectral characteristics; adopting the spatial autocorrelation statistic as a local measurement index of spatial correlation to construct spatial correlation characteristics of the false color image; calculating the spatial heterogeneity characteristic of the false color image by adopting a spatial variation function; and combining the spectral feature, the spatial correlation feature and the spatial heterogeneity feature to construct a feature set, and extracting the target ground object from the false color image.
Note: the two schemes are based on the mode of combining bottom layer feature extraction and classifier classification to realize the extraction of large-area ground object types, and the accurate extraction of the farmland blocks cannot be realized.
3) The invention discloses an unmanned aerial vehicle aerial image farmland block object accurate extraction method (publication number CN 107563413A). The invention provides an unmanned aerial vehicle aerial image farmland block object accurate extraction method, which comprises the steps of firstly obtaining image edge information by utilizing contour detection based on spectral information, then obtaining bottom segmentation blocks by adopting watershed transformation, generating multi-scale segmentation maps based on contour intensity, and finally realizing non-farmland region elimination by supervised image classification.
Note: the scheme only utilizes spectral information to carry out farmland edge detection, and does not design a structured edge detection method by combining the space geometric characteristics of farmland objects, so that the farmland edge is extracted and the edge of a non-farmland area is enhanced; meanwhile, the scheme does not utilize the space semantic information of the farmland object to optimize the farmland boundary after the farmland boundary extraction is carried out on the high-resolution aerial photography or the satellite image, the obtained farmland boundary interference is more, and the boundary usability is to be improved.
The existing methods for extracting land features (such as cultivated land, forest land, urban area and the like) basically adopt the idea of feature extraction and classifier classification, only large-area farmland areas can be extracted, different farmland blocks cannot be distinguished, and the accurate boundary of a single farmland cannot be determined. Under the influence of different geographical positions, different crop types and different growth periods, the spectrum and texture distribution rules of adjacent farmland blocks in the image are not obvious, and the boundary between the two is also unstable, so that the extraction of the boundary of a single farmland needs comprehensive spectrum, texture, edge and object structure information.
Disclosure of Invention
The invention provides a satellite image high-precision farmland boundary extraction method based on multi-scale structure learning, solves the technical problem of accurate extraction of a single farmland boundary, and provides basic support information for accurate agricultural application.
The technical scheme adopted by the invention is as follows:
a satellite image high-precision farmland boundary extraction method based on multi-scale structure learning comprises the following steps:
step 1, edge detection based on structure learning: after a high-resolution satellite image is input, edge detection is firstly carried out on the satellite image based on local structural pattern learning, and an edge strength value of each pixel point in the satellite image on a specific scale is obtained;
step 2, multi-scale edge information fusion: by means of multi-scale edge information fusion, the sensitivity of edge detection to different resolutions of satellite images is reduced, so that edge detection results can be combined with high-level semantic information to eliminate edge noise while edge positioning accuracy is kept;
step 3, multi-scale segmentation based on the hypermetrological contour map: step 31, converting multi-scale edge information into bottom layer segmentation block information, namely superpixels, through watershed transformation; step 32, analyzing the contour intensity between adjacent segmentation blocks by utilizing the hypermetrological contour map to combine the segmentation blocks layer by layer to obtain segmentation blocks with different scales;
step 4, farmland boundary optimization based on semantic modeling: the spatial distribution positions of the farmland blocks are modeled by utilizing the conditional random field, so that the usability of the farmland boundary is improved.
Further, the structure learning-based edge detection in step 1 specifically includes the following steps:
step 11, feature extraction: performing feature extraction based on an image slice of 16 × 16 pixel size;
step 12, training a structure classifier: using a random forest as a structure classifier, obtaining an edge/non-edge training truth value required by the structure classifier by marking a farmland boundary in a satellite image, and training the structure classifier based on the training truth value;
step 13, predicting the edge strength: and predicting the edge intensity value of the satellite image by using the trained structure classifier to realize edge detection.
Further, in the step 12, the random forest is replaced by a support vector machine, an artificial neural network or a deep neural network.
Further, the extracted features include pixel intensity, color, gradient magnitude and direction, and the difference of the features between two pixels is calculated.
Further, the multi-scale edge information fusion in the step 2 specifically includes the following steps:
setting the contour intensity on a scale s at each pixel point position (x, y) in the image obtained by edge detection as Ls (x, y), and performing weighted average on Ls (x, y) on a plurality of scales s to obtain the average contour intensity Lm (x, y) at the pixel point (x, y):
wherein, wsAre weighting factors for the contour intensity Ls (x, y) at different scales s.
Furthermore, for satellite images with resolutions better than 1 m, the scale s is {1/6,1/4,1/2,1}, that is, edge blending is performed on the original image and the original image after downsampling the original image by 1/6,1/4, 1/2.
Further, the step 31 specifically includes the following steps:
obtaining an edge intensity map through edge detection and multi-scale edge fusion;
selecting at least one local minimum point as a seed point;
converting the multi-scale edge intensity information into continuous bottom layer segmentation blocks through watershed transformation;
the average value of the average contour intensity Lm (x, y) at all pixel points on the contour between two adjacent divided blocks is taken as the intensity value of this contour.
Further, the step 32 specifically includes the following steps:
for the current segmentation map, selecting a section of the contour with the minimum intensity value in the contours between all two segmentation blocks;
combining two segmentation blocks at two sides of the section of the contour into a new segmentation block, and deleting the corresponding contour;
updating the intensity value of each contour in the segmentation graph;
and repeating the steps to obtain the segmentation blocks with different scales.
Furthermore, semantic information of the farmland is represented by modeling in the step 4, wherein the semantic information comprises category consistency and spatial distribution morphology.
Further, assuming that the class distribution of the image region with the characteristic f is z, the farmland semantic distribution model based on the conditional random field is represented in an energy form as follows:
wherein the content of the first and second substances,
is a univariate potential function representing a class z for a pixel iiA penalty of (2);
is a binary potential function representing a class z of pairs of pixelsiAnd zjA penalty of (2);
and (3) performing iterative computation on the E (z | f), finally enabling the internal categories of the farmland region to tend to be consistent, and simultaneously enabling the shape of the obtained farmland region to be consistent with the actual situation, so that the optimization of the farmland boundary is realized.
The method has the advantages that through multi-scale structure learning, the high-precision farmland boundary extraction method facing the high-resolution satellite image is realized, and basic geographic information support can be provided for precision agriculture. By testing on a large-scale high-resolution satellite image data set (comprising 0.3 meter, 0.6 meter and the like), the accuracy of farmland boundary extraction by the method can reach about 70%, and the accuracy of farmland boundary extraction by the conventional method is less than 50%, so that the method can realize accurate extraction of farmland boundaries.
Drawings
FIG. 1 is a general flow chart of satellite image high-precision farmland boundary extraction based on multi-scale structure learning;
FIG. 2 is a schematic diagram of a local edge pattern classification based on a structure classifier;
FIG. 3 is a schematic diagram of a field boundary before (left) and after (right) optimization based on semantic modeling.
Detailed Description
The invention is further illustrated below with reference to the figures and examples. FIG. 1 is a general flow chart of satellite image high-precision farmland boundary extraction based on multi-scale structure learning, which specifically comprises the following steps:
step 1, edge detection based on structure learning: after a high-resolution satellite image is input, edge detection is firstly carried out on the satellite image based on local structural pattern learning, and an edge intensity value of each pixel point in the image on a specific scale is obtained.
Step 2, multi-scale edge fusion: by means of multi-scale edge information fusion, the sensitivity of edge detection to different resolutions of satellite images can be reduced, edge detection results can be combined with high-level semantic information to eliminate edge noise while edge positioning accuracy is kept, and robustness of edge detection is improved.
Step 3, multi-scale segmentation based on the hypermetrological contour map: the edge information is converted into the bottom layer of segmentation block information, namely superpixels, through Watershed Transform (WT), and then, the contour strength between adjacent segmentation blocks is analyzed by using an ultra metric contour map (UCM) to combine the segmentation blocks layer by layer, so as to obtain segmentation maps with different scales, wherein the segmentation blocks with larger scales represent field block objects with larger sizes.
Super-pixel: homogeneity segmentation generated by over-segmentation is called superpixel (superpixel) quickly, a farmland can be segmented into one or more superpixels, and therefore the superpixels need to be combined (grouping) or clustered (clustering) to obtain a complete farmland block.
Under-segmentation: the adjacent farmlands are divided into one area, namely insufficient division and too few division blocks.
And (3) over-segmentation: a farmland is divided into a plurality of areas, namely, the number of divided blocks is too large.
Step 4, farmland boundary optimization based on semantic modeling: by modeling the spatial distribution position and form of the farmland block by using a Conditional Random Field (CRF), the problems of burrs, irregularity and the like in the extracted high-precision farmland boundary can be eliminated, thereby improving the usability of the farmland boundary. The invention finally obtains the boundary extraction result of each farmland block, different farmland segmentation blocks are represented by different colors in the result diagram shown in figure 1, and the outlines of the segmentation blocks represent the accurate boundary of the farmland.
Steps 1-4 are further illustrated below with reference to specific examples:
step 1, edge detection based on structure learning:
the invention provides an edge detection method based on structure mode learning, which can represent a local edge structure mode aiming at a farmland block object and realize high-precision and high-credibility farmland edge detection by modeling a local edge structure in an image slice with the size of 16 multiplied by 16 pixels and combining structure classifier learning to carry out edge detection. The edge detection based on the structure pattern learning can obtain the contour intensity value L (x, y) at each pixel point (x, y) in the image, and the larger the contour intensity value is, the higher the probability that the contour line exists at the point is. The edge detection process based on structure learning comprises the following steps:
step 11, feature extraction: feature extraction is carried out on the basis of image slices with the size of 16 x 16 pixels, the extracted features comprise pixel intensity, color, gradient amplitude, direction and the like, and the difference of the features between two pixels is calculated.
Step 12, training a structure classifier: using Random Forest (RF) as a structure classifier, obtaining an edge/non-edge training truth value required by the classifier by labeling a farmland boundary in one image, and training the structure classifier based on the training truth value.
The structure classifier used in the present invention may also select other supervised classification methods besides the random forest, such as Support Vector Machine (SVM), Artificial Neural Network (ANN), or Deep Neural Network (DNN).
Step 13, predicting the edge strength: and predicting the edge intensity value of the test image by using the trained structure classifier to realize edge detection.
The process of classifying the structure mode of the local edge by using the structure classifier is shown in fig. 2, and the obtained image slice edge mode is used for realizing more accurate farmland edge detection.
Step 2, multi-scale edge fusion:
the multi-scale edge information fusion has the function of enabling edge detection to be combined with high-precision edge positioning information of the satellite image on the original scale and structural semantic information on the large scale. Let Ls (x, y) be the contour intensity on the scale s at the pixel point position (x, y) in the image obtained by edge detection. By performing weighted averaging on Ls (x, y) over multiple scales s, the average contour intensity Lm (x, y) at the pixel point (x, y) can be obtained:
wherein wsIs a weighting factor for the profile intensity Ls (x, y) at different scales s, the scale s and the corresponding weighting factor wsDifferent assignments may be chosen depending on the application. For satellite images with resolution better than 1 m, the scale adopted by the invention is s ═ {1/6,1/4,1/2,1}, namely, edge fusion is carried out on the original image and the original image after downsampling the original image according to 1/6,1/4 and 1/2. By carrying out multi-scale information fusion on the pixel edge strength obtained by edge detection, the false edge interference existing on a high-resolution (such as 0.3 meter) image can be eliminated by utilizing the semantic information on a large scale, the robustness of edge detection is improved, meanwhile, the high edge positioning precision can be kept, and a foundation is provided for subsequent high-precision farmland boundary extraction.
Step 3, multi-scale segmentation based on the hypermetrological contour map:
firstly, in an edge intensity image obtained by multi-scale edge detection and fusion, a plurality of local minimum value points are selected as seed points, edge intensity information is converted into a plurality of continuous bottom segmentation blocks through watershed transformation, and the average of all pixel point edge intensity values Lm (x, y) on a contour between two adjacent segmentation blocks is taken as the intensity value of the contour. A greater intensity value of the contour between the segments indicates a greater probability of the existence of a field boundary there.
And carrying out hierarchical semantic combination on the bottom layer segmentation blocks generated by watershed transformation by utilizing an ultra metric contour map (UCM), so as to generate segmentation blocks with different scales, and enabling the segmentation blocks to correspond to actual farmland blocks one by selecting segmentation blocks with larger scales. The process of generating higher level partitions from the lower level partitions is as follows: 1) for the current segmentation map, selecting a section of the contour with the minimum intensity from the contours between all two segmentation blocks; 2) combining two segmentation blocks at two sides of the section of the contour into a new segmentation block, and deleting the corresponding contour; 3) updating the intensity value of each contour in the segmentation graph; 4) the process is repeated to obtain the segmentation blocks with different scales.
Step 4, farmland boundary optimization based on semantic modeling:
modeling the spatial semantic distribution of a farmland block by using a Conditional Random Field (CRF) can represent semantic information on two aspects of the farmland: 1) the category consistency is that the probability that pixels with similar characteristics and similar distances are farmland (non-farmland) categories at the same time is higher; 2) the spatial distribution forms, such as that most farmlands are rectangular and the farmlands in the same region tend to be consistent. Therefore, the problems of burrs, irregularity and the like of the farmland boundary extracted under the high-resolution condition can be solved, and the usability of the farmland boundary is improved. Assuming that the class distribution of an image region with the characteristic of f is z, the farmland semantic distribution model based on the conditional random field is represented by the following energy form:
wherein the content of the first and second substances,
is a univariate potential function representing a class z for a pixel iiPenalty of (in energy);
is a binary potential function representing a class z of pairs of pixelsiAnd ziPenalty of if ziAnd zjThe closer the class distribution of the model (a) is to the field block distribution mode obtained by training, the smaller the punishment is. By carrying out iterative computation on E (z | f), the internal categories of the farmland region finally tend to be consistent, and meanwhile, the obtained farmland region shape is more practical, and finally, the optimization of the farmland boundary is realized. The farmland boundaries before and after optimization based on semantic modeling are shown in fig. 3, and the farmland boundaries after optimization can be found to be more in line with the actual situation.
The method is mainly used for high-resolution satellite images with the resolution ratio superior to 1 meter, and is also applicable to aerial images or unmanned aerial vehicle aerial images with the possibly higher resolution ratio after simple parameter adaptation.
Although the present invention has been described with reference to the preferred embodiments, it is not intended to limit the present invention, and those skilled in the art can make variations and modifications of the present invention without departing from the spirit and scope of the present invention by using the methods and technical contents disclosed above.
Claims (8)
1. A satellite image high-precision farmland boundary extraction method based on multi-scale structure learning is characterized by comprising the following steps:
step 1, edge detection based on structure learning: after a high-resolution satellite image is input, edge detection is firstly carried out on the satellite image based on local structural pattern learning, and an edge strength value of each pixel point in the satellite image on a specific scale is obtained;
step 2, multi-scale edge information fusion: an edge intensity graph is obtained through multi-scale edge information fusion, the sensitivity of edge detection to different resolutions of a satellite image is reduced, and edge noise can be eliminated by combining high-level semantic information while the edge positioning accuracy of an edge detection result is kept;
step 3, multi-scale segmentation based on the hypermetrological contour map: step 31, converting multi-scale edge information into bottom layer segmentation block information, namely superpixels, through watershed transformation; step 32, analyzing the contour intensity between adjacent segmentation blocks by utilizing the hypermetrological contour map to combine the segmentation blocks layer by layer to obtain segmentation blocks with different scales;
step 4, farmland boundary optimization based on semantic modeling: the spatial distribution positions of the farmland blocks are modeled by utilizing the conditional random field, so that the usability of the farmland boundary is improved;
the structure learning-based edge detection in the step 1 specifically includes the following steps:
step 11, feature extraction: feature extraction is performed based on image slices of size 16 × 16 pixels, the extracted features including: calculating the difference of the characteristics of the two pixels according to the pixel intensity, the color, the gradient amplitude and the direction;
step 12, training a structure classifier: using a random forest as a structure classifier, obtaining an edge/non-edge training truth value required by the structure classifier by marking a farmland boundary in a satellite image, and training the structure classifier based on the training truth value;
step 13, predicting the edge strength: and predicting the edge intensity value of the satellite image by using the trained structure classifier to realize edge detection.
2. The method for extracting the satellite image high-precision farmland boundary based on the multi-scale structure learning as claimed in claim 1, wherein the random forest is replaced by a support vector machine or an artificial neural network in the step 12.
3. The method for extracting the satellite image high-precision farmland boundary based on the multi-scale structure learning as claimed in claim 1, wherein the multi-scale edge information fusion in the step 2 specifically comprises the following steps:
setting the contour intensity on a scale s at each pixel point position (x, y) in the image obtained by edge detection as Ls (x, y), and performing weighted average on Ls (x, y) on a plurality of scales s to obtain the average contour intensity Lm (x, y) at the pixel point (x, y):
wherein, wsAre weighting factors for the contour intensity Ls (x, y) at different scales s.
4. The method as claimed in claim 3, wherein for satellite images with resolution better than 1 m, the scale is {1/6,1/4,1/2,1}, that is, edge blending is performed by using images obtained by down-sampling original images according to 1/6,1/4,1/2 and the original images.
5. The method for extracting the satellite image high-precision farmland boundary based on the multi-scale structure learning as claimed in claim 3, wherein the step 31 specifically comprises the following steps:
acquiring an edge intensity map;
selecting at least one local minimum point as a seed point;
converting the multi-scale edge intensity information into continuous bottom layer segmentation blocks through watershed transformation;
the average value of the average contour intensity Lm (x, y) at all pixel points on the contour between two adjacent divided blocks is taken as the intensity value of this contour.
6. The method for extracting the satellite image high-precision farmland boundary based on the multi-scale structure learning as claimed in claim 5, wherein the step 32 specifically comprises the following steps:
for the current segmentation map, selecting a section of the contour with the minimum intensity value in the contours between all two segmentation blocks;
combining two segmentation blocks at two sides of the section of the contour into a new segmentation block, and deleting the corresponding contour;
updating the intensity value of each contour in the segmentation graph;
and repeating the steps to obtain the segmentation blocks with different scales.
7. The method for extracting farmland boundary with high precision based on the multi-scale structure learning as claimed in claim 6, wherein semantic information of the farmland, including category consistency and spatial distribution form, is represented by modeling in the step 4.
8. The method for extracting the satellite image high-precision farmland boundary based on the multi-scale structure learning as claimed in claim 7, wherein the image region with the characteristic of f is assumed to have the class distribution of z, and the farmland semantic distribution model based on the conditional random field is represented by the following energy form:
wherein the content of the first and second substances,
is a univariate potential function representing a class z for a pixel iiA penalty of (2);
is a binary potential function representing a class z of pairs of pixelsiAnd zjA penalty of (2);
and (3) performing iterative computation on the E (z | f), finally enabling the internal categories of the farmland region to tend to be consistent, and simultaneously enabling the shape of the obtained farmland region to be consistent with the actual situation, so that the optimization of the farmland boundary is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810491945.6A CN108830870B (en) | 2018-05-21 | 2018-05-21 | Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810491945.6A CN108830870B (en) | 2018-05-21 | 2018-05-21 | Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108830870A CN108830870A (en) | 2018-11-16 |
| CN108830870B true CN108830870B (en) | 2021-12-28 |
Family
ID=64148047
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810491945.6A Active CN108830870B (en) | 2018-05-21 | 2018-05-21 | Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108830870B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111612792B (en) * | 2019-02-22 | 2024-03-08 | 曹生 | VRDS 4D medical image-based Ai endoscope analysis method and product |
| CN111753834B (en) * | 2019-03-29 | 2024-03-26 | 中国水利水电科学研究院 | Planting land block structure semantic segmentation method and device based on deep neural network |
| KR102040702B1 (en) * | 2019-05-07 | 2019-11-05 | (주)컨텍 | Apparatus and method for providing application service using satellite image |
| CN110216082B (en) * | 2019-05-23 | 2020-07-14 | 上海交通大学 | Fluorescence labeling seed dynamic identification method and system |
| CN110188778B (en) * | 2019-05-31 | 2023-04-07 | 中国人民解放军61540部队 | Residential area element outline regularization method based on image extraction result |
| CN111968141B (en) * | 2020-06-30 | 2023-06-16 | 广西科技大学 | Non-classical receptive field contour detection method based on multi-scale superposition |
| WO2022061496A1 (en) * | 2020-09-22 | 2022-03-31 | 深圳市大疆创新科技有限公司 | Object boundary extraction method and apparatus, mobile platform and storage medium |
| CN112308872B (en) * | 2020-11-09 | 2023-06-23 | 西安工程大学 | Image edge detection method based on multi-scale Gabor first derivative |
| CN113063375B (en) * | 2021-03-16 | 2022-04-08 | 成都理工大学 | Unmanned aerial vehicle remote sensing extraction method for linear farming ridges |
| CN115482246B (en) * | 2021-05-31 | 2023-06-16 | 数坤(上海)医疗科技有限公司 | Image information extraction method and device, electronic equipment and readable storage medium |
| CN113870278A (en) * | 2021-08-24 | 2021-12-31 | 广州华农大智慧农业科技有限公司 | Improved Mask R-CNN model-based satellite remote sensing image farmland block segmentation method |
| CN114092813B (en) * | 2021-11-25 | 2022-08-05 | 中国科学院空天信息创新研究院 | Industrial park image extraction method and system, electronic equipment and storage medium |
| CN116805396B (en) * | 2023-08-24 | 2023-12-29 | 杭州稻道农业科技有限公司 | Satellite remote sensing-based farmland weed precise identification method and device |
| CN117593201B (en) * | 2024-01-18 | 2024-06-14 | 杭州稻道农业科技有限公司 | High-resolution satellite image-based field-combining field integration method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530878A (en) * | 2013-10-12 | 2014-01-22 | 北京工业大学 | Edge extraction method based on fusion strategy |
| WO2015031641A1 (en) * | 2013-08-29 | 2015-03-05 | Mayo Foundation For Medical Education And Research | System and method for boundary classification and automatic polyp detection |
| CN105740869A (en) * | 2016-01-28 | 2016-07-06 | 北京工商大学 | Square operator edge extraction method and system based on multiple scales and multiple resolutions |
| CN106558022A (en) * | 2016-11-30 | 2017-04-05 | 重庆大学 | A kind of single image super resolution ratio reconstruction method constrained based on Edge difference |
| CN107016415A (en) * | 2017-04-12 | 2017-08-04 | 合肥工业大学 | A kind of coloured image Color Semantic sorting technique based on full convolutional network |
| CN107563413A (en) * | 2017-08-09 | 2018-01-09 | 千寻位置网络有限公司 | The accurate extracting method of unmanned plane image farmland block object |
| CN107909597A (en) * | 2017-11-14 | 2018-04-13 | 西安建筑科技大学 | A kind of Multiscale Markov Random Field Models image partition method kept with edge |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7505606B2 (en) * | 2005-05-19 | 2009-03-17 | Microsoft Corporation | Detecting doctored images using camera response normality and consistency |
| US9842282B2 (en) * | 2015-05-22 | 2017-12-12 | Here Global B.V. | Method and apparatus for classifying objects and clutter removal of some three-dimensional images of the objects in a presentation |
-
2018
- 2018-05-21 CN CN201810491945.6A patent/CN108830870B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015031641A1 (en) * | 2013-08-29 | 2015-03-05 | Mayo Foundation For Medical Education And Research | System and method for boundary classification and automatic polyp detection |
| CN103530878A (en) * | 2013-10-12 | 2014-01-22 | 北京工业大学 | Edge extraction method based on fusion strategy |
| CN105740869A (en) * | 2016-01-28 | 2016-07-06 | 北京工商大学 | Square operator edge extraction method and system based on multiple scales and multiple resolutions |
| CN106558022A (en) * | 2016-11-30 | 2017-04-05 | 重庆大学 | A kind of single image super resolution ratio reconstruction method constrained based on Edge difference |
| CN107016415A (en) * | 2017-04-12 | 2017-08-04 | 合肥工业大学 | A kind of coloured image Color Semantic sorting technique based on full convolutional network |
| CN107563413A (en) * | 2017-08-09 | 2018-01-09 | 千寻位置网络有限公司 | The accurate extracting method of unmanned plane image farmland block object |
| CN107909597A (en) * | 2017-11-14 | 2018-04-13 | 西安建筑科技大学 | A kind of Multiscale Markov Random Field Models image partition method kept with edge |
Non-Patent Citations (5)
| Title |
|---|
| "Contour Detection and Hierarchical Image Segmentation";Pablo Arbela´ez等;《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》;20110531;第33卷(第5期);全文 * |
| "Remote Sensing Image Scene Classification Using Multi-Scale Completed Local Binary Patterns and Fisher Vectors";Huang L等;《Remote Sensing》;20160608;第8卷(第6期);全文 * |
| "基于多尺度分割的土地利用分类研究";张东梅;《中国优秀硕士学位论文全文数据库·基础科学辑》;20180115;第2018年卷(第1期);全文 * |
| "基于边缘信息增强的面向对象高分影像信息提取研究";秦礼明;《中国优秀硕士学位论文全文数据库·基础科学辑》;20160815;第2016年卷(第8期);全文 * |
| "高分辨率遥感影像多尺度纹理、形状特征提取与面向对象分类研究";黄昕;《中国博士学位论文全文数据库·信息科技辑》;20130915;第2013年卷(第9期);第六章 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108830870A (en) | 2018-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108830870B (en) | Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning | |
| CN108846832B (en) | Multi-temporal remote sensing image and GIS data based change detection method and system | |
| US10984532B2 (en) | Joint deep learning for land cover and land use classification | |
| Zhang et al. | Scale Sequence Joint Deep Learning (SS-JDL) for land use and land cover classification | |
| CN111986099B (en) | Tillage monitoring method and system based on convolutional neural network with residual error correction fused | |
| EP3614308B1 (en) | Joint deep learning for land cover and land use classification | |
| Blaschke et al. | Image segmentation methods for object-based analysis and classification | |
| CN107563413B (en) | Unmanned aerial vehicle aerial image farmland block object accurate extraction method | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN108052966B (en) | Remote sensing image scene automatic extraction and classification method based on convolutional neural network | |
| Zhang et al. | Automated delineation of agricultural field boundaries from Sentinel-2 images using recurrent residual U-Net | |
| Doxani et al. | Monitoring urban changes based on scale-space filtering and object-oriented classification | |
| CN101840581B (en) | Method for extracting profile of building from satellite remote sensing image | |
| CN110263717A (en) | It is a kind of incorporate streetscape image land used status determine method | |
| CN113449594A (en) | Multilayer network combined remote sensing image ground semantic segmentation and area calculation method | |
| CN111028255A (en) | Farmland area pre-screening method and device based on prior information and deep learning | |
| CN107330875A (en) | Based on the forward and reverse heterogeneous water body surrounding enviroment change detecting method of remote sensing images | |
| CN109829426B (en) | Railway construction temporary building monitoring method and system based on high-resolution remote sensing image | |
| CN111611960B (en) | Large-area ground surface coverage classification method based on multilayer perceptive neural network | |
| Sutha | Object based classification of high resolution remote sensing image using HRSVM-CNN classifier | |
| CN109344837B (en) | SAR image semantic segmentation method based on deep convolutional network and weak supervised learning | |
| Ganchenko et al. | Special Area Detection and Recognition on Agricultural Fields Images | |
| Kalpana | Analysis of rain fall and the temperature of Coimbatore District using land use and land cover change detection by image segmentation | |
| Mohammadzadeh et al. | Automatic main road extraction from high resolution satellite imageries by means of particle swarm optimization applied to a fuzzy-based mean calculation approach | |
| CN115984689A (en) | Multi-scale earth surface complexity feature extraction and land utilization segmentation method |
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 |