CN103034863A - Remote-sensing image road acquisition method combined with kernel Fisher and multi-scale extraction - Google Patents
Remote-sensing image road acquisition method combined with kernel Fisher and multi-scale extraction Download PDFInfo
- Publication number
- CN103034863A CN103034863A CN2012105654258A CN201210565425A CN103034863A CN 103034863 A CN103034863 A CN 103034863A CN 2012105654258 A CN2012105654258 A CN 2012105654258A CN 201210565425 A CN201210565425 A CN 201210565425A CN 103034863 A CN103034863 A CN 103034863A
- Authority
- CN
- China
- Prior art keywords
- road
- image
- built
- area
- extraction
- 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.)
- Granted
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 114
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000012549 training Methods 0.000 claims abstract description 26
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 25
- 230000000694 effects Effects 0.000 claims abstract description 22
- 230000003595 spectral effect Effects 0.000 claims abstract description 21
- 239000011159 matrix material Substances 0.000 claims description 25
- 238000012545 processing Methods 0.000 claims description 19
- 238000001228 spectrum Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 230000011218 segmentation Effects 0.000 claims description 4
- 241000764238 Isis Species 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 2
- 239000003550 marker Substances 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 238000010276 construction Methods 0.000 abstract description 4
- 238000011156 evaluation Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000011160 research Methods 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 238000007796 conventional method Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 240000007594 Oryza sativa Species 0.000 description 1
- 235000007164 Oryza sativa Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 235000009566 rice Nutrition 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 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
- G06V20/182—Network patterns, e.g. roads or rivers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a remote-sensing image road acquisition method combined with kernel Fisher and multi-scale extraction. The remote-sensing image road acquisition method combined with the kernel Fisher and the multi-scale extraction comprises the steps of extracting simple surface features (such as vegetation, shadows of water bodies and constructions, and bare lands), training a kernel Fisher classifier through spectral signature, achieving classification of the simple surface features, removing influences of the surface features on a road, to obtain a built-up area containing road and construction information, and extracting the road in the built-up area. On the one hand, the spectral signature and texture feature are simultaneously used for training the kernel Fisher classifier to divide the built-up area into the road and constructions. On the other hand, road information of common lane count is extracted in sequence according to the fact that different roads with different lane counts are different in width. Finally, road information obtained by kernel Fisher classification and road information obtained by the multi-scale extraction are combined and final road information is acquired. The remote-sensing image road acquisition method combined with the kernel Fisher and the multi-scale extraction improves effect of road extraction and improves accuracy of the road extraction.
Description
Technical Field
The invention relates to extraction of a linear ground object target, in particular to a method for acquiring road information from a remote sensing image by combining kernel Fisher classification and multi-scale extraction.
Background
Roads as a modern urban traffic framework have important geographic, political and economic meanings, and are also main recording and identifying objects in maps and geographic information systems. Road information is used as basic geographic information, is an important component of geographic information data, and is basic data applied to many geographic information systems, such as vehicle navigation, traffic management, emergency response and the like. At present, remote sensing satellites, imaging radars and unmanned planes develop rapidly, earth observation means are more complete, and obtained image data are increasingly abundant. How to accurately and timely extract and utilize information from the massive data becomes a problem which needs to be solved. On the other hand, the speed of city construction is increased, so that road information must be updated regularly to ensure the availability and accuracy of the road information. At present, the road extraction method in actual production mainly adopts manual interpretation and identification, and the efficiency and the precision are difficult to ensure. The extraction of the important feature of the road by using a computer has become a research hotspot in the field of pattern recognition and remote sensing at present. In recent years, many scholars have studied road extraction methods from different angles and different application fields, and the road extraction methods are classified into semi-automatic road extraction and automatic road extraction according to the degree of automation of road extraction.
The semi-automatic road extraction method is carried out by using a man-machine interaction mode, and the main idea is that an initial point of a road is manually provided, and then a computer carries out identification and processing according to a certain rule. The automatic road extraction method is used for automatically and accurately positioning the position of a road by recognizing and understanding the characteristics of the road. The automatic road extraction method is the development direction and the final target of remote sensing image target identification and extraction. The extraction effect of the current various methods is still not ideal, and due to the complexity and diversity of remote sensing images and the technical limitation on the extraction of the ground objects such as roads in various fields such as computer target recognition, artificial intelligence and mode recognition, the method becomes the bottleneck of research on the extraction of linear ground objects such as roads by domestic and foreign experts.
The currently common road extraction methods mainly include a template matching method, a knowledge-based method, an object-oriented extraction method, a ridge and valley line extraction method, a mathematical morphology method, a region segmentation method and other extraction methods. The methods are mainly suitable for extracting rural roads from medium-low resolution remote sensing images, but for high-resolution remote sensing images provided by QuickBird, IKONOS, WorldView-2, Geoeye-1 and other satellites, foreign matter co-spectrum conditions often exist between roads and buildings in the images, and the roads and the buildings are often classified by mistake when the extraction is carried out by using a conventional method. In addition, the information contained in the high-resolution remote sensing image is richer, and the interference information contained in the road is increased gradually, such as vehicle, house and tree shadows, which increase the difficulty of extracting the road information. Many existing road extraction methods are not suitable for high-resolution remote sensing images.
In recent years, the application of the kernel Fisher classification method as an effective mode classification technology in remote sensing image classification processing is gradually a research hotspot. The kernel Fisher classification has two advantages over support vector machines: (1) the kernel Fisher classification has no concept of support vectors, the complexity of the kernel Fisher classification is proportional to the number of training samples, and the complexity of a support vector machine is closely related to the number of the support vectors; (2) the performance of kernel Fisher is superior to support vector machines in some ways, the main reason for this is that the training of the former relies on the whole training samples, while the latter relies mainly on support vectors. At present, only researchers put forward 'hyperspectral remote sensing image classification based on kernel Fisher discriminant analysis' (remote sensing science report, 2008, 12 (4): 579-. The research of road extraction is carried out on the remote sensing image with high spatial resolution by applying a nuclear Fisher classification method, and reports are not found at home and abroad.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides the remote sensing image road acquisition method combining kernel Fisher and multi-scale extraction.
The technical solution of the invention for realizing the above purpose is as follows:
a remote sensing image road acquisition method combining kernel Fisher and multi-scale extraction divides land and objects in a remote sensing image into five categories of vegetation, water and building shadows, bare land, buildings and roads; the processing steps are as follows:
1) extracting simple ground objects, wherein the simple ground objects refer to vegetation, water bodies, building shadows and bare land;
1.1) aiming at three types of ground objects, namely vegetation, water, building shadow and bare land, respectively selecting sample data of the three simple ground objects from a remote sensing image, and carrying out normalization processing on spectral features of the sample data to obtain training samples of the three simple ground objects;
1.2) training a nuclear Fisher classifier of the vegetation, the water body and building shadow and the bare land in sequence by using the obtained training samples of the vegetation, the water body and building shadow and the bare land information;
1.3) calculating the spectral characteristics of the remote sensing image data to be classified and normalizing to obtain a test set; classifying the test set by utilizing a nuclear Fisher classifier of three simple ground objects obtained by training, sequentially extracting vegetation, water and building shadows and bare land information, and eliminating the influence of the three simple ground objects on road extraction to obtain a built-up area image only containing road and building information;
2) extracting roads in the built-up area image;
2.1) respectively calculating the spectrum and texture characteristics of sample data of two types of ground objects, namely a road and a building, respectively obtaining spectrum characteristic training samples and texture characteristic training samples of the road and the building, and training a kernel Fisher classifier by using the training samples;
2.2) calculating spectral features and texture features of roads and buildings in the built-up area image, normalizing to obtain a test set, extracting the ground objects in the built-up area image by using the kernel Fisher classifier obtained by training in the step 2.1), and eliminating the influence of the buildings on road extraction to obtain initial road information 1;
2.3) carrying out multi-scale extraction on the built-up area by utilizing the characteristic that roads with different lane numbers have different widths, and respectively extracting to obtain road information of various lane numbers; merging the road information of each lane number to obtain initial road information 2;
and 2.4) merging the road information results obtained in the steps 2.2) and 2.3), and removing holes in the road by using mathematical morphology to obtain a final road extraction effect graph.
The spectral characteristics of the steps 1) and 2) comprise R, G, B data of three wave bands; the texture features of step 2) comprise energy features and homogeneity features.
And (3) respectively processing the initial road information extracted in the steps 2.2) and 2.3) by using mathematical morphology and an area threshold.
The mathematical morphology is that according to the structure of non-road pixels in the image, proper structural elements are selected for mathematical morphology processing, and part of non-road pixels are separated from road pixels; the area threshold is used for removing objects with undersized areas by selecting a proper area threshold and filtering part of noise.
And 2.3) the method for extracting the road in multiple scales comprises the following steps:
2.3.1) the size of the area formed by the built-up area (including roads and buildings) and the non-built-up area (including vegetation, water body, shadow of buildings and bare land) is as followsBinary image of
Is divided into
Each size isThe built-up area of the image block matrix is represented by '1', and the non-built-up area is represented by '0'; definition ofTo correspond to a binary image
The flag matrix of (2); counting '1's in each image blockNumber ofIf it is greater than the threshold value
Then the image block can be considered as a built-up area image block, i.e.
Otherwise, the image block is a non-as-built area image block, i.e.
(ii) a Expressed by the formula:
wherein,
，
，
，
，；
2.3.2) on the created area marking matrix obtained in the last step
Traversing the entire token matrix by selecting an appropriate sliding windowDefinition of
For marking the image blocks corresponding to the built-up area
A road block marking matrix of (1); counting the number of '1' in each sliding windowIf it is less than the threshold valueThen the image block can be considered to characterize the road, i.e.
Otherwise, the image block is considered to characterize the building, i.e.
(ii) a Expressed by the formula:
wherein,
，
，
judging the threshold value of the road block in the sliding window;
2.3.3) for the segmented image blocks of the non-built area, due to the influence of the segmentation scale and the threshold, part of road blocks are misjudged into the image blocks of the non-built area; for the image blocks, the road blocks are secondarily judged in the following modes to ensure the accuracy of road extraction:
if it isIs 1, then；
Wherein,
，
，“"is a logical AND operation; if not built-up area image block
If one or more road blocks exist in the window range, the non-built area image block can be judged to be a road image block;
2.3.4) Using road Block Mark matrixExtracting the built-up area to obtain the road with the scale; definition of
Is a sizeAs road effect image, if
And then:
wherein,
，
，
；
2.3.5) carrying out multi-scale extraction on the whole built-up area by adopting the method, and sequentially extracting road information containing two lanes, four lanes, six lanes and eight lanes.
And merging the road information of each lane number, and then carrying out mathematical morphology and area threshold processing to obtain the road information obtained by multi-scale extraction.
Compared with the prior art, the invention has the following effects:
(1) the invention adopts the idea of one-to-one multi-time classification, firstly extracts obvious non-road areas, removes the influence of the areas on the extraction of road information, and then further adopts a corresponding algorithm to extract the road information.
(2) The invention utilizes the spectrum and texture characteristics of various ground objects in the high-resolution remote sensing image and the kernel Fisher classifier to extract roads, fully utilizes the advantages of the kernel Fisher classifier and the spectrum and texture characteristics of the ground objects, and improves the road extraction effect.
(3) The method combines the characteristics that the shapes of roads and buildings are different in images and the roads containing different numbers of lanes have different widths, carries out multi-scale extraction on the built-up area to obtain the roads with different widths, fully considers the influence of the road widths on the road extraction effect, and improves the road extraction precision.
Drawings
Fig. 1 is a flowchart of a high-resolution remote sensing image road extraction method provided by the invention.
FIG. 2 shows the shape of a common road and the type of a road intersectionPresentation under the window; wherein (a) a vertical road; (b) a horizontal road; (c) a curved road; (d) t-shaped intersecting roads; (e) y-shaped intersecting roads; (f) the intersecting roads are perpendicular.
FIG. 3 is an original WorldView-2 image.
Fig. 4 is a diagram of vegetation information extraction results.
Fig. 5 is a water body and building shadow information extraction result diagram.
Fig. 6 is a diagram of bare land information extraction results.
FIG. 7 is a result diagram of road extraction using kernel Fisher classification; wherein, (a) a regional binary image is built; (b) a road initial result graph extracted by a kernel Fisher classification method; (c) and (4) checking the road result extracted by Fisher classification.
FIG. 8 shows a schematic view of a process using
The sliding window extracts road information obtained by multi-scale extraction of the road; wherein, (a) a two-lane road extraction result map; (b) extracting a result graph of the four-lane road; (c) extracting a result graph of a road with six lanes; (d) extracting a result graph of the road with eight lanes; (e) extracting a merged graph from roads with two to eight lanes; (f) and (4) obtaining a road result graph after multi-scale extraction and mathematical morphology processing.
FIG. 9 is a graph of road acquisition effects of kernel Fisher classification combined with multi-scale extraction.
Fig. 10 is a diagram illustrating the effect of the final road extraction.
Fig. 11 is a manually extracted reference road.
FIG. 12 is a diagram of road extraction effect of a conventional method; wherein, (a) the method proposed by Mingjun Song, (b) the method proposed by Liliwei; (c) envi 4.6 object oriented approach.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
The invention selects the size of a certain area as
The WorldView-2 image of (A) was the subject of the study, which provided a 1.88 meter spaceThe resolution includes R, G, B multispectral images, as shown in FIG. 3. The invention divides the ground objects in the remote sensing image into five categories of vegetation, water body and building shadow, bare land, buildings and roads;
the method comprises the following specific steps, and the process is shown in a figure 1:
(1) extracting simple ground objects, wherein the simple ground objects refer to vegetation, water bodies, building shadows and bare land;
Calculating spectral characteristics of remote sensing image data to be classified and normalizing to obtain a test set; classifying the test set by using the trained nuclear Fisher classifiers of three simple ground objects, extracting vegetation, water body and building shadow and bare land information in sequence, eliminating the influence of the three simple ground objects on road extraction, and obtaining a built-up area image only containing road and building information。
The invention aims to realize extraction of road information, and for the land features with larger difference from roads, such as vegetation, water body and building shadow, bare land and the like, the spectral features can be adopted as the feature vector of a kernel Fisher classifier, so that the influence of the land features on road extraction can be eliminated. Fig. 4, 5 and 6 show vegetation, water and building shadow, bare land information extracted by using a nuclear Fisher classifier. As can be seen from the figure, better extraction effect can be obtained by adopting the spectral characteristics. Because the water body and the building shadow formed by illumination have similar spectral characteristics, the water body and the building shadow are unified into a ground object. The object of the invention is the extraction of road information, so that the processing has no influence on the extraction of road information. On the other hand, roads have spectral features the same as or similar to those of buildings, and cannot be classified correctly only by using the spectral features. The invention selects the texture characteristics and the geometric characteristics to be matched with the spectral characteristics to classify roads and buildings.
(2) Extraction of complex ground objects
The texture feature is extracted by using a gray level co-occurrence matrix (GLCM), wherein the gray level co-occurrence matrix is a second-order statistical measure of image gray level change and is a basic function for describing texture structure property features. It is defined as the distance in the image as
The joint probability distribution of two simultaneous gray scale pixels. If the image gray level is N, the gray level co-occurrence matrix with a certain spatial relationship is
WhereindIs composed of
，The pixel pair pitch is such that,
which is the direction of the two pixel connection lines. Get and decidedValues are respectively set to，
，
，
The direction influence is eliminated by adopting the superposition of the 4 directions, and 4 gray level co-occurrence matrixes are formed.
The gray level co-occurrence matrix can obtain a plurality of statistical indexes such as mean value, energy, contrast, entropy, homogeneity and the like. They reflect the gray level distribution, information amount and texture thickness of the image from different angles. The quality of the feature selection directly affects the classification effect. The invention selects two texture features of energy and homogeneity to assist the spectral feature to carry out classification research on roads and buildings.
(Energy)
Is used to measure texture consistency or uniformity, which is defined as:
degree of homogeneity
Is used to reflect the degree of image uniform adjustment, which is defined as:
the texture feature extraction algorithm adopted by the invention is as follows:
1) calculating the intensity component of the multispectral image, namely:
2) and carrying out gray level quantization processing. Since the calculation amount for obtaining the gray level co-occurrence matrix is large, in order to reduce the calculation amount, the gray level is generally coarsely quantized, i.e., 256 gray levels are reduced to N gray levels, and N is 8 in the present invention. Although the quantized image has a certain distortion, the influence of the quantized image on the extraction of the texture features is small.
3) By using
Processing the image by the template of the window, calculating the gray level co-occurrence matrix in four directions of each window, respectively calculating the characteristic values of the four directions of each gray level co-occurrence matrix, and averaging the characteristic values of the four directions of each characteristic quantity to obtain the final texture characteristic value.
4) Traversing the template of the sliding window through the whole image, and extracting a gray level co-occurrence matrix of the whole image, wherein the distance of each moving of the sliding window is 1 pixel.
The spectrum and the texture features are used as the feature vectors of the kernel Fisher classifier to train the kernel Fisher classifier, the design method of the classifier is similar to that of the classifier, and the details are not repeated here.
Fig. 7 (a) is a built-up region binary image. Fig. 7 (b) is a result of extracting initial road information by using spectrum and texture features, it can be seen from the figure that most of the block building information has been basically removed, and for non-removed roads, the following method can be used for processing, specifically:
1) mathematical morphology. According to the structure of the non-road pixels in the image, selecting proper structural elements to perform mathematical morphology processing, and separating partial non-road pixels from road pixels.
2) And (4) area threshold value. The area of the road cannot be very small, objects with small areas can be removed by selecting a proper area threshold value T, meanwhile, partial noise is filtered, and the follow-up operation time is saved.
Fig. 7 (c) shows the road obtained by mathematical morphology and area threshold processing. It can be seen from the figure that roads connected with buildings are basically proposed, but a part of isolated roads are not proposed yet, and for the roads, road information is acquired by using a multi-scale extraction method.
The multi-scale extraction is mainly applied to the extraction of isolated roads separated from buildings in a built-up area, and is a supplement to the extraction result of the kernel Fisher. Roads differ from buildings in that: (1) roads with different lane numbers have different widths; (2) roads appear as long stripes on the image, while buildings appear as regular geometric bodies on the image. By combining the characteristics with the kernel Fisher classification, a better classification effect can be achieved.
The width of each motor vehicle lane has certain standard requirements and contains differentThe number of lanes of the road has different widths. The invention utilizes the relationship between the number of lanes and the width of the road, adopts a multi-scale extraction method to obtain the road information, and extracts the scale
Is defined as:
extraction scale = resolution of road actual width/image
The correspondence between the number of lanes, the road width, and the extraction scale is shown in table 1.
TABLE 1 correspondence of number of lanes, road width and extraction scale
| Number of lanes | Road width (Rice) | Extraction scale |
| Double lane | 7 | 5 |
| Four lanes | 15 | 9 |
| Six lanes | 22.5 | 13 |
| Eight lanes | 30 | 19 |
The method for extracting the road by utilizing the multi-scale comprises the following steps:
1) will be made up of a built-up area (denoted by "1") and a non-built-up area (denoted by "0") of a size of
Binary image of
Is divided into
Each size is
The image block matrix of (1). Definition ofTo correspond to a binary image
The flag matrix of (2). Counting the number of '1' in each image block
If it is greater than the threshold value
Then the image block can be considered as a built-up area image block, i.e.
Otherwise, the image block is a non-as-built area image block, i.e.
. Expressed by the formula:
wherein,
，
，
，
，。
2) marking the constructed area obtained in the last step with a matrix
By selecting a suitable sliding window size
In the invention, choose and useSliding windows of size traverse the entire marker matrix
Definition ofFor marking the image blocks corresponding to the built-up areaThe road block marking matrix in (1). Make statistics of each
Number of "1" in window
If it is less than the threshold valueThen the image block can be considered to characterize the road, i.e.
Otherwise, the image block is considered to characterize the building, i.e.
. Expressed by the formula:
wherein,
，
，
and judging the threshold value of the road block in the sliding window.
Common road and road intersection types are: horizontal roads, vertical roads, curved roads, T-shaped intersecting roads, Y-shaped intersecting roads, and vertical intersecting roads, as shown in fig. 2. In order to effectively reduce the calculation amount, the invention discloses a road block judgment threshold value
Are all 11.
3) For the segmented image blocks of the non-built-up area, due to the influence of the segmentation scale and the threshold, part of road blocks are misjudged into the image blocks of the non-built-up area. For the image blocks, the road blocks are secondarily judged in the following modes to ensure the accuracy of road extraction:
if it isIs 1, then
。
Wherein,
，
，“
"is a logical" and "operation. If not built-up area image block
If one or more road blocks exist in the window range, the non-built area image block can be judged to be a road image block, so that the integrity of extracting the road image block is ensured.
4) Using road block marking matrices
And extracting the built-up area to obtain the road with the scale. Definition of
Is a size
As road effect image, if
And then:
wherein,
，
，
。
the whole built-up area is subjected to multi-scale extraction by adopting the mode, and road information containing two lanes, four lanes, six lanes and eight lanes is extracted in sequence.
5) And merging the road information of each lane number to obtain initial road information obtained after multi-scale extraction, and performing mathematical morphology processing to obtain the road information obtained by multi-scale extraction.
FIG. 8 is a road result graph obtained by multi-scale extraction. Fig. 9 is a road extraction effect graph obtained by combining kernel Fisher classification and multi-scale extraction, fig. 10 is a final road information extraction effect graph obtained by removing holes in roads by using mathematical morphology, and as can be seen from fig. 10, urban roads in high-resolution remote sensing images are basically extracted.
In order to verify the accuracy and effectiveness of the invention, the experiment adopts a Road extraction method based on a support vector machine, which is proposed in the literature of "Road extraction SVM and image segmentation" (photoelectric Engineering and Remote Sensing, 2004, 70 (12): 1365) by Mingjun Song et al, a Road extraction method based on mathematical morphology, which is proposed in the literature of "high-resolution Remote Sensing image Road extraction based on mathematical morphology" (Remote Sensing information, 2005, 9-11) by Listanding Wei et al, and a Road extraction method based on an object-oriented technology, which is provided by ENVI 4.6 software, for comparison. The experiment compares the extraction effect of the invention and three common methods with the artificially extracted reference road, and obtains the subjective evaluation of the road extraction quality. In addition, three evaluation indexes of accuracy, namely accuracy, omission error and redundancy error, proposed by Wiedemann C. et al in the literature "External evaluation of road networks" (International industries of photography and Remote Sensing, 2003, 34 (3): 93-98) are adopted as objective evaluation criteria of experimental effects. The higher the accuracy, the higher the accuracy of road extraction, the lower the missing error and the redundant error, and the better the road extraction effect. The definition of accuracy, missing errors and redundant errors is as follows:
the experiment selection is configured to be a desktop computer with an AMD dual-core Sempron central processing unit (with a main frequency of 2.30 GHz) and a memory of 2GB, and simulation experiments are carried out under a Windows XP operating system by utilizing Matlab7.0 software. The road extraction effect of the invention is shown in fig. 10, except that a small part of roads are arranged at the upper right corner of fig. 10, and the other most roads are extracted because the spectral and textural features of the roads are very close to surrounding buildings and cannot be extracted accurately. The reference roads extracted manually are shown in fig. 11. The road extraction results of the conventional method are shown in fig. 12. As can be seen from fig. 12, there is a high degree of misjudgment by using the three common methods. In fig. 12 (a), although the accuracy of the road is high, the redundancy error is high, and there are cases where a building is mistakenly broken as a road. The missing road error in fig. 12 (b) is large, and many missing roads exist. The road in fig. 12 (c) is less accurate and less effective in extracting a road having a narrow width. In general, the road extracted by the method has higher accuracy, smaller missing error and moderate redundant error, and is obviously superior to the common road extraction method. The objective evaluation results of the various algorithm road extractions are shown in table 2.
TABLE 2 Objective evaluation results of road extraction Algorithm
| Method of producing a composite material | Accuracy (%) | Missing error (%) | Redundant error (%) |
| The method proposed by the invention | 94.03 | 6.05 | 20.21 |
| Method proposed by Mingjun Song | 87.34 | 13.08 | 47.24 |
| Method for extracting Liliwei | 82.19 | 18.19 | 11.24 |
| Envi 4.6 object-oriented method | 76.68 | 23.43 | 30.87 |
As can be seen from table 2, the evaluation results of the present invention are greater than the method proposed by liliiwei et al only in redundancy error, and the other evaluation results are superior to the conventional methods.
The invention provides a high-resolution remote sensing image road acquisition method combining kernel Fisher classification and multi-scale extraction. Multi-scale extraction has mainly two advantages: firstly, extracting common road information of two lanes, four lanes, six lanes and eight lanes by using different scales; secondly, the process of extracting roads in multiple scales has less calculation amount, and has higher extraction precision for independent roads far away from buildings. The nuclear Fisher classification method fully utilizes the spectral and textural features of the ground objects in the high-resolution remote sensing image, and the multi-scale extraction mainly utilizes the width information of roads, and the roads extracted by the two methods complement each other to jointly complete the extraction of the urban road information of the high-resolution remote sensing image.
Finally, the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered in the claims of the present invention.
Claims (4)
1. A remote sensing image road acquisition method combining kernel Fisher and multi-scale extraction divides land and objects in a remote sensing image into five categories of vegetation, water and building shadows, bare land, buildings and roads; the method is characterized by comprising the following steps:
1) extracting simple ground objects, wherein the simple ground objects refer to vegetation, water bodies, building shadows and bare land;
1.1) aiming at three types of ground objects, namely vegetation, water, building shadow and bare land, respectively selecting sample data of the three simple ground objects from a remote sensing image, and carrying out normalization processing on spectral features of the sample data to obtain training samples of the three simple ground objects;
1.2) training a nuclear Fisher classifier of the vegetation, the water body and building shadow and the bare land in sequence by using the obtained training samples of the vegetation, the water body and building shadow and the bare land information;
1.3) calculating the spectral characteristics of the remote sensing image data to be classified and normalizing to obtain a test set; classifying the test set by utilizing a nuclear Fisher classifier of three simple ground objects obtained by training, sequentially extracting vegetation, water and building shadows and bare land information, and eliminating the influence of the three simple ground objects on road extraction to obtain a built-up area image only containing road and building information;
2) extracting roads in the built-up area image;
2.1) respectively calculating the spectrum and texture characteristics of sample data of two types of ground objects, namely a road and a building, respectively obtaining spectrum characteristic training samples and texture characteristic training samples of the road and the building, and training a kernel Fisher classifier by using the training samples;
2.2) calculating spectral features and texture features of roads and buildings in the built-up area image, normalizing to obtain a test set, extracting the ground objects in the built-up area image by using the kernel Fisher classifier obtained by training in the step 2.1), and eliminating the influence of the buildings on road extraction to obtain initial road information 1;
2.3) carrying out multi-scale extraction on the built-up area by utilizing the characteristic that roads with different lane numbers have different widths, and respectively extracting to obtain road information of various lane numbers; merging the road information of each lane number to obtain initial road information 2;
and 2.4) merging the road information results obtained in the steps 2.2) and 2.3), and removing holes in the road by using mathematical morphology to obtain a final road extraction effect graph.
2. The method for acquiring the road of the remote sensing image by combining the kernel Fisher and the multi-scale extraction as claimed in claim 1, wherein the spectral features in the steps 1) and 2) comprise data of R, G, B three wave bands; the texture features of step 2) comprise energy features and homogeneity features.
3. The method for acquiring the remote sensing image road by combining the kernel Fisher and the multi-scale extraction according to claim 1 or 2, wherein the road information extracted in the step 2.2) and the step 2.3) is processed by mathematical morphology and an area threshold value respectively;
the mathematical morphology is that according to the structure of non-road pixels in an image, proper structural elements are selected for mathematical morphology processing, and part of non-road pixels are separated from road pixels; the area threshold is used for removing objects with undersized areas by selecting a proper area threshold and filtering part of noise.
4. The method for acquiring the remote sensing image road by combining the kernel Fisher and the multi-scale extraction as claimed in claim 1, wherein the method for extracting the road in the 2.3) step by the multi-scale extraction is as follows:
2.3.1) size to be made up of built-on and non-built-on areasBinary image of
Is divided into
Each size isThe built-up area of the image block matrix is represented by '1', and the non-built-up area is represented by '0'; definition of
To correspond to a binary image
The flag matrix of (2); counting the number of '1' in each image block
If it is greater than the threshold value
Then the image block can be considered as a built-up area image block, i.e.
Otherwise, the image block is a non-as-built area image block, i.e.
(ii) a Expressed by the formula:
wherein,
，
，，
，；
2.3.2) on the created area marking matrix obtained in the last stepBy selecting a size of
The sliding window traverses the entire marker matrix
Definition of
For marking the image blocks corresponding to the built-up area
A road block marking matrix of (1); counting the number of '1' in each sliding window
If it is less than the threshold value
Then the image block can be considered to characterize the road, i.e.
Otherwise, the image block is considered to characterize the building, i.e.
(ii) a Expressed by the formula:
wherein,
，
，
judging the threshold value of the road block in the sliding window;
2.3.3) for the segmented image blocks of the non-built area, due to the influence of the segmentation scale and the threshold, part of road blocks are misjudged into the image blocks of the non-built area; for the image blocks, the road blocks are secondarily judged in the following modes to ensure the accuracy of road extraction:
if it is
Is 1, then
；
Wherein,
，
，“
"is a logical AND operation; if not built-up area image block
If one or more road blocks exist in the window range, the non-built area image block can be judged to be a road image block;
2.3.4) Using road Block Mark matrix
Extracting the original image to obtain the road with the scale; definition of
Is a size
As road effect image, if
And then:
wherein:
，
，；
performing multi-scale extraction on the whole built-up area by adopting the mode, and sequentially extracting road information containing two lanes, four lanes, six lanes and eight lanes;
2.3.5) merging the road information of each lane number, and then obtaining the road information obtained by multi-scale extraction through mathematical morphology and area threshold processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210565425.8A CN103034863B (en) | 2012-12-24 | 2012-12-24 | The remote sensing image road acquisition methods of a kind of syncaryon Fisher and multiple dimensioned extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210565425.8A CN103034863B (en) | 2012-12-24 | 2012-12-24 | The remote sensing image road acquisition methods of a kind of syncaryon Fisher and multiple dimensioned extraction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103034863A true CN103034863A (en) | 2013-04-10 |
| CN103034863B CN103034863B (en) | 2015-08-12 |
Family
ID=48021742
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210565425.8A Active CN103034863B (en) | 2012-12-24 | 2012-12-24 | The remote sensing image road acquisition methods of a kind of syncaryon Fisher and multiple dimensioned extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103034863B (en) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103279951A (en) * | 2013-05-13 | 2013-09-04 | 武汉理工大学 | Object-oriented remote sensing image building and shade extraction method of remote sensing image building |
| CN104036295A (en) * | 2014-06-18 | 2014-09-10 | 西安电子科技大学 | Road center line auto-detection method employing multispectral remote sensing images |
| CN104239885A (en) * | 2014-09-05 | 2014-12-24 | 北京航天控制仪器研究所 | Earthquake disaster damage degree evaluation method based on unmanned aerial vehicle aerial photos |
| CN104406715A (en) * | 2014-12-15 | 2015-03-11 | 重庆市勘测院 | Precision evaluation method and system for remote sensing estimation of surface sensible heat/latent heat flux |
| CN104966091A (en) * | 2015-07-30 | 2015-10-07 | 王植 | Strip mine road extraction method based on unmanned plane remote sensing images |
| CN105279519A (en) * | 2015-09-24 | 2016-01-27 | 四川航天系统工程研究所 | Remote sensing image water body extraction method and system based on cooperative training semi-supervised learning |
| CN106249601A (en) * | 2016-09-29 | 2016-12-21 | 广东华路交通科技有限公司 | A kind of road section length division methods based on Ordered Clustering Analysis |
| CN103971367B (en) * | 2014-04-28 | 2017-01-11 | 河海大学 | Hydrologic data image segmenting method |
| CN106372599A (en) * | 2016-08-30 | 2017-02-01 | 水利部水土保持监测中心 | Method and system for extracting silt arrester in water and soil retaining period |
| CN106529484A (en) * | 2016-11-16 | 2017-03-22 | 哈尔滨工业大学 | Combined spectrum and laser radar data classification method based on class-fixed multinucleated learning |
| CN107507202A (en) * | 2017-09-28 | 2017-12-22 | 武汉大学 | A kind of vegetation rotary island towards high-resolution remote sensing image automates extracting method |
| CN108108721A (en) * | 2018-01-09 | 2018-06-01 | 北京市遥感信息研究所 | A kind of method that road extraction is carried out using EO-1 hyperion |
| CN108229273A (en) * | 2017-02-27 | 2018-06-29 | 北京市商汤科技开发有限公司 | Multilayer neural network model training, the method and apparatus of roadway characteristic identification |
| CN108287845A (en) * | 2017-01-09 | 2018-07-17 | 北京四维图新科技股份有限公司 | A kind of Automatic extraction method for road information and device and hybrid navigation system |
| CN109409438A (en) * | 2018-11-07 | 2019-03-01 | 重庆市勘测院 | The Remote Image Classification inferred based on IFCM cluster with variation |
| CN109493320A (en) * | 2018-10-11 | 2019-03-19 | 苏州中科天启遥感科技有限公司 | Method for extracting remote sensing image road and system, storage medium, electronic equipment based on deep learning |
| CN109948693A (en) * | 2019-03-18 | 2019-06-28 | 西安电子科技大学 | Expand and generate confrontation network hyperspectral image classification method based on super-pixel sample |
| CN110619651A (en) * | 2019-09-09 | 2019-12-27 | 博云视觉（北京）科技有限公司 | Driving road segmentation method based on monitoring video |
| CN110795994A (en) * | 2019-09-16 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Intersection image selection method and device |
| CN110809767A (en) * | 2017-07-06 | 2020-02-18 | 华为技术有限公司 | Advanced driver assistance system and method |
| CN112613371A (en) * | 2020-12-16 | 2021-04-06 | 上海大学 | Hyperspectral image road extraction method based on dense connection convolution neural network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070058863A1 (en) * | 2005-09-09 | 2007-03-15 | Honeywell International Inc. | Label Detection |
| CN102682301A (en) * | 2010-12-08 | 2012-09-19 | 通用汽车环球科技运作有限责任公司 | Adaptation for clear path detection with additional classifiers |
-
2012
- 2012-12-24 CN CN201210565425.8A patent/CN103034863B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070058863A1 (en) * | 2005-09-09 | 2007-03-15 | Honeywell International Inc. | Label Detection |
| CN102682301A (en) * | 2010-12-08 | 2012-09-19 | 通用汽车环球科技运作有限责任公司 | Adaptation for clear path detection with additional classifiers |
Non-Patent Citations (1)
| Title |
|---|
| 楚恒等: "纹理与几何特征在道路提取中的应用", 《计算机与现代化》, 31 July 2012 (2012-07-31), pages 96 - 99 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103279951B (en) * | 2013-05-13 | 2016-03-09 | 武汉理工大学 | A kind of method of OO remote sensing image building and shadow extraction thereof |
| CN103279951A (en) * | 2013-05-13 | 2013-09-04 | 武汉理工大学 | Object-oriented remote sensing image building and shade extraction method of remote sensing image building |
| CN103971367B (en) * | 2014-04-28 | 2017-01-11 | 河海大学 | Hydrologic data image segmenting method |
| CN104036295A (en) * | 2014-06-18 | 2014-09-10 | 西安电子科技大学 | Road center line auto-detection method employing multispectral remote sensing images |
| CN104036295B (en) * | 2014-06-18 | 2017-03-01 | 西安电子科技大学 | Multi-spectrum remote sensing image road axis automatic testing method |
| CN104239885B (en) * | 2014-09-05 | 2017-11-28 | 北京航天控制仪器研究所 | A kind of earthquake disaster damage degree appraisal procedure based on unmanned plane |
| CN104239885A (en) * | 2014-09-05 | 2014-12-24 | 北京航天控制仪器研究所 | Earthquake disaster damage degree evaluation method based on unmanned aerial vehicle aerial photos |
| CN104406715A (en) * | 2014-12-15 | 2015-03-11 | 重庆市勘测院 | Precision evaluation method and system for remote sensing estimation of surface sensible heat/latent heat flux |
| CN104966091A (en) * | 2015-07-30 | 2015-10-07 | 王植 | Strip mine road extraction method based on unmanned plane remote sensing images |
| CN105279519A (en) * | 2015-09-24 | 2016-01-27 | 四川航天系统工程研究所 | Remote sensing image water body extraction method and system based on cooperative training semi-supervised learning |
| CN105279519B (en) * | 2015-09-24 | 2018-09-21 | 四川航天系统工程研究所 | Remote sensing image Clean water withdraw method and system based on coorinated training semi-supervised learning |
| CN106372599A (en) * | 2016-08-30 | 2017-02-01 | 水利部水土保持监测中心 | Method and system for extracting silt arrester in water and soil retaining period |
| CN106249601A (en) * | 2016-09-29 | 2016-12-21 | 广东华路交通科技有限公司 | A kind of road section length division methods based on Ordered Clustering Analysis |
| CN106529484A (en) * | 2016-11-16 | 2017-03-22 | 哈尔滨工业大学 | Combined spectrum and laser radar data classification method based on class-fixed multinucleated learning |
| CN108287845A (en) * | 2017-01-09 | 2018-07-17 | 北京四维图新科技股份有限公司 | A kind of Automatic extraction method for road information and device and hybrid navigation system |
| CN108229273A (en) * | 2017-02-27 | 2018-06-29 | 北京市商汤科技开发有限公司 | Multilayer neural network model training, the method and apparatus of roadway characteristic identification |
| CN110809767A (en) * | 2017-07-06 | 2020-02-18 | 华为技术有限公司 | Advanced driver assistance system and method |
| CN110809767B (en) * | 2017-07-06 | 2022-09-09 | 华为技术有限公司 | Advanced driver assistance system and method |
| CN107507202A (en) * | 2017-09-28 | 2017-12-22 | 武汉大学 | A kind of vegetation rotary island towards high-resolution remote sensing image automates extracting method |
| CN108108721A (en) * | 2018-01-09 | 2018-06-01 | 北京市遥感信息研究所 | A kind of method that road extraction is carried out using EO-1 hyperion |
| CN109493320A (en) * | 2018-10-11 | 2019-03-19 | 苏州中科天启遥感科技有限公司 | Method for extracting remote sensing image road and system, storage medium, electronic equipment based on deep learning |
| CN109493320B (en) * | 2018-10-11 | 2022-06-17 | 苏州中科天启遥感科技有限公司 | Remote sensing image road extraction method and system based on deep learning, storage medium and electronic equipment |
| CN109409438B (en) * | 2018-11-07 | 2021-09-07 | 重庆市勘测院 | Remote sensing image classification method based on IFCM clustering and variational inference |
| CN109409438A (en) * | 2018-11-07 | 2019-03-01 | 重庆市勘测院 | The Remote Image Classification inferred based on IFCM cluster with variation |
| CN109948693A (en) * | 2019-03-18 | 2019-06-28 | 西安电子科技大学 | Expand and generate confrontation network hyperspectral image classification method based on super-pixel sample |
| CN110619651A (en) * | 2019-09-09 | 2019-12-27 | 博云视觉（北京）科技有限公司 | Driving road segmentation method based on monitoring video |
| CN110619651B (en) * | 2019-09-09 | 2023-01-17 | 博云视觉（北京）科技有限公司 | Driving road segmentation method based on monitoring video |
| CN110795994A (en) * | 2019-09-16 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Intersection image selection method and device |
| CN112613371A (en) * | 2020-12-16 | 2021-04-06 | 上海大学 | Hyperspectral image road extraction method based on dense connection convolution neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103034863B (en) | 2015-08-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103034863B (en) | The remote sensing image road acquisition methods of a kind of syncaryon Fisher and multiple dimensioned extraction | |
| CN110321963B (en) | Hyperspectral image classification method based on fusion of multi-scale and multi-dimensional space spectrum features | |
| CN108846832B (en) | Multi-temporal remote sensing image and GIS data based change detection method and system | |
| CN111898688B (en) | Airborne LiDAR data tree classification method based on three-dimensional deep learning | |
| CN103400151B (en) | The optical remote sensing image of integration and GIS autoregistration and Clean water withdraw method | |
| CN107358260B (en) | Multispectral image classification method based on surface wave CNN | |
| CN102496034B (en) | High-spatial resolution remote-sensing image bag-of-word classification method based on linear words | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| CN109635733B (en) | Parking lot and vehicle target detection method based on visual saliency and queue correction | |
| Zhou et al. | An integrated skeleton extraction and pruning method for spatial recognition of maize seedlings in MGV and UAV remote images | |
| CN111639587B (en) | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network | |
| CN101714254A (en) | Registering control point extracting method combining multi-scale SIFT and area invariant moment features | |
| CN104700398A (en) | Point cloud scene object extracting method | |
| CN103632363A (en) | Object-level high-resolution remote sensing image change detection method based on multi-scale fusion | |
| CN110309780A (en) | High resolution image houseclearing based on BFD-IGA-SVM model quickly supervises identification | |
| CN103606164B (en) | SAR image segmentation method based on high-dimensional triple Markov field | |
| CN103226826B (en) | Based on the method for detecting change of remote sensing image of local entropy visual attention model | |
| CN109635789B (en) | High-resolution SAR image classification method based on intensity ratio and spatial structure feature extraction | |
| Yuan et al. | Learning to count buildings in diverse aerial scenes | |
| CN110070545B (en) | Method for automatically extracting urban built-up area by urban texture feature density | |
| CN103854290A (en) | Extended target tracking method combining skeleton characteristic points and distribution field descriptors | |
| CN116205879A (en) | Unmanned aerial vehicle image and deep learning-based wheat lodging area estimation method | |
| CN113361407B (en) | PCANet-based spatial spectrum feature combined hyperspectral sea ice image classification method | |
| CN105825215B (en) | It is a kind of that the instrument localization method of kernel function is embedded in based on local neighbor and uses carrier | |
| CN105005962B (en) | Islands and reefs Remote Sensing Image Matching method based on hierarchical screening strategy |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240305 Address after: 401120 No. 6, Qingzhu East Road, Dazhulin, Yubei District, Chongqing Patentee after: Chongqing Institute of Surveying and Mapping Science and Technology (Chongqing Map Compilation Center) Country or region after: China Address before: 400020 Jiangbei District, Chongqing electric measuring Village No. 231 Patentee before: CHONGQING SURVEY INSTITUTE Country or region before: China |
|
| TR01 | Transfer of patent right |