CN108446588B - Double-temporal remote sensing image change detection method and system - Google Patents
Double-temporal remote sensing image change detection method and system Download PDFInfo
- Publication number
- CN108446588B CN108446588B CN201810113902.4A CN201810113902A CN108446588B CN 108446588 B CN108446588 B CN 108446588B CN 201810113902 A CN201810113902 A CN 201810113902A CN 108446588 B CN108446588 B CN 108446588B
- Authority
- CN
- China
- Prior art keywords
- image
- distance
- feature map
- file
- generate
- 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.)
- Expired - Fee Related
Links
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
- G06V20/13—Satellite images
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C11/00—Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
- G01C11/02—Picture taking arrangements specially adapted for photogrammetry or photographic surveying, e.g. controlling overlapping of pictures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N2021/1793—Remote sensing
Landscapes
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Immunology (AREA)
- General Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Pathology (AREA)
- Chemical & Material Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radar, Positioning & Navigation (AREA)
- Health & Medical Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a double-time-phase remote sensing image change detection method and a double-time-phase remote sensing image change detection system, wherein the method comprises the following steps: according to the size of the appointed block, performing block cutting on the front and rear time phase images to generate a front and rear time phase image block data set, a list file of block image information and a vector file for recording the position and the characteristic information of the block image; performing high-dimensional feature extraction on the front and rear time phase image block data sets by using a deep learning network model to generate a high-dimensional feature file; calculating the characteristic distance of high-dimensional characteristics of front and rear time phase image blocks, generating a characteristic distance file, and adjusting characteristic distance parameters to obtain changed image blocks; and comparing the changed image block with the image block of the reference data, obtaining a precision evaluation result by using an error matrix, and judging whether the precision evaluation result meets the requirement. The invention provides a double-temporal remote sensing image change detection method and system, which can improve the change detection precision.
Description
Technical Field
The invention relates to the technical field of remote sensing, in particular to a method and a system for detecting changes of double-temporal remote sensing images.
Background
The remote sensing image change detection is an indispensable technology for geographic national condition monitoring, land utilization, coverage change monitoring and the like, and mainly comprises two means of manual visual interpretation and computer automatic/semi-automatic change detection.
The manual visual interpretation is to manually distinguish the response spectrum information on the images in different periods, and find out different parts from the response spectrum information to judge the change. The method relies on the knowledge of the interpreter on the interpretation area and the accumulated interpretation prior knowledge, and has the advantages of higher accuracy and precision, the greatest disadvantages of low efficiency and higher labor cost, and different interpretators can have different image analysis results due to different interpretation experiences and knowledge.
The computer automatic/semi-automatic change detection method is the key point of the current research, and is mainly divided into two major categories of pixel-oriented methods and object-oriented methods seen from an analysis unit, wherein the pixel-oriented methods comprise direct comparison (such as difference, ratio and regression), transformation-based methods (such as PCA principal component analysis method, Thyshat transformation, change vector analysis and texture analysis), change detection after classification (such as comparison after classification and multi-temporal direct comparison), machine learning (such as artificial neural network, support vector machine and decision tree), GIS (geographic information system) auxiliary methods and the like; the object-oriented method comprises direct object comparison, object classification comparison, multi-temporal superposition detection and the like. Each of these methods has its advantages and disadvantages, and new methods are developed for different data sources and application purposes and requirements, but it is still very difficult to select a suitable method for a specific application and research area, and change detection is still a hot point of research.
With the development of new-generation information technologies such as artificial intelligence and big data, it is necessary and advanced to study the change detection method supported by the new-generation information technologies, deep learning is a door machine learning technology which is developed in the big data environment in recent years, and the essence is to learn the complex features in the data by constructing a machine learning model with a multilayer structure and massive training data, and use the complex features in the data for later analysis, prediction and other applications. Due to the fact that the method has the learning capacity and high feature abstraction capacity of mass data, revolutionary success is achieved in the application fields of voice recognition, image recognition, information retrieval and the like compared with the traditional method, the method is mainly applied to image classification, object detection, image restoration, image fusion and the like in the remote sensing field, research on change detection is few, and research on a remote sensing image change detection method is lacked.
Disclosure of Invention
In view of the above, the present invention provides a method and a system for detecting changes in a dual-temporal remote sensing image, which can improve the accuracy of change detection.
Based on the above object, the present invention provides a method for detecting changes in a dual-temporal remote sensing image, which includes:
according to the size of the appointed block, performing block cutting on the front and rear time phase images to generate a front and rear time phase image block data set, a list file of block image information and a vector file for recording the position and the characteristic information of the block image;
performing high-dimensional feature extraction on the front and rear time phase image block data sets by using a deep learning network model to generate a high-dimensional feature file;
calculating the characteristic distance of the high-dimensional characteristics of the front and rear time phase image blocks according to the list file of the block image information, the vector file for recording the positions of the block images and the characteristic information and the high-dimensional characteristic file to generate a characteristic distance file;
adjusting the characteristic distance parameters according to the characteristic distance file to obtain changed image blocks;
comparing the changed image block with an image block of reference data, obtaining a precision evaluation result by using an error matrix, and judging whether the precision evaluation result meets the requirement; if the requirement is met, outputting the changed image block; and if the characteristic distance parameter does not meet the requirement, adjusting the characteristic distance parameter to continue judging.
Optionally, in the process of segmenting and cropping the front and rear time phase images, if the background pixel proportion of an image block at a certain position exceeds 50%, the front and rear time phase image blocks at the certain position are removed.
Optionally, the step of performing high-dimensional feature extraction on the front-time phase image block data set and the back-time phase image block data set by using a deep learning network model to generate a high-dimensional feature file includes: inputting accurate manual sampling data in advance, training the deep learning network model, optimizing parameters of the deep learning network model to obtain a fine-tuned deep learning network model, and performing high-dimensional feature extraction on the front-time phase image block data set and the rear-time phase image block data set by using the fine-tuned deep learning network model to generate a high-dimensional feature file.
Optionally, the performing, by using the fine-tuned deep learning network model, high-dimensional feature extraction on the front-time phase image block data set and the rear-time phase image block data set includes:
the first convolution processing, inputting 227 x 3 front and rear time phase image block data sets, firstly performing convolution operation through 96 11 x 11 convolution kernels to generate a 55 x 96 feature map, then performing activation processing on the 55 x 96 feature map through a correction linear unit to generate an activation feature map, performing pooling operation with the scale of 3 x 3 and the step size of 2 to generate a 27 x 96 feature map, and finally performing normalization processing to form a 27 x 96 feature map;
a second convolution process, wherein a feature map of 27 × 96 is input, and a convolution operation is performed through 256 convolution kernels of 5 × 5 to generate a feature map of 27 × 256; then, performing activation processing on the feature map of 27 × 256 by a correction linear unit to generate an activated feature map, performing pooling operation with the scale of 3 × 3 and the step size of 2 to generate a feature map of 13 × 256, and finally performing normalization processing to form a feature map of 13 × 256;
the third convolution processing is carried out, 13 × 256 feature maps are input, firstly, the convolution operation is carried out through 384 convolution kernels with 3 × 3 to generate 13 × 384 feature maps, then, the activation processing is carried out on the 13 × 384 feature maps through a correction linear unit to generate activation feature maps, and the 13 × 384 feature maps are formed;
a fourth convolution process, wherein a feature map of 13 × 384 is input, and a convolution operation is performed through 384 convolution kernels of 3 × 3 to generate a feature map of 13 × 384; then, performing activation processing on the feature map of 13 × 384 through a correction linear unit to generate an activation feature map, and forming a feature map of 13 × 384;
in the fifth convolution processing, a feature map of 13 × 384 is input, and convolution operation is performed through 256 convolution kernels of 3 × 3 to generate a feature map of 13 × 256; then, performing activation processing on the feature map of 13 × 256 through a correction linear unit to generate an activation feature map; forming a feature map of 6 × 256 through pooling operation with a scale of 3 × 3 and a step size of 2;
a first full-connection process, wherein feature map data of 6 × 256 is fully connected with 4096 neurons, and a feature map of 4096 is output;
and a second full-connection process of fully connecting 4096 feature map data to 4096 neurons and outputting a 4096 feature map.
Optionally, the feature distance of the high-dimensional feature of the anterior-posterior temporal phase image block includes: the minimum distance, the related distance and the distance ratio of the high-dimensional features of the front and rear time phase image blocks.
Optionally, the distance ratio (i, j) of the high-dimensional features of the front-time phase image block and the rear-time phase image block is:
Optionally, adjusting the characteristic distance parameter according to the characteristic distance file to obtain the changed image block includes: adjusting the minimum distance parameter, the related distance parameter and the distance ratio parameter according to the characteristic distance file;
adjusting the minimum distance parameter includes: sorting the minimum distances of all the image blocks, calculating a histogram and a cumulative function of the minimum distances, determining a minimum distance threshold value under the parameter, and outputting the image blocks with the minimum distances exceeding the threshold value, namely the changed image blocks;
adjusting the relevant distance parameter includes: sorting the correlation distances of all the image blocks, calculating a histogram and a cumulative function of the correlation distances, determining a correlation distance threshold value under the parameter, and outputting the image blocks of which the correlation distances exceed the threshold value, namely the changed image blocks;
adjusting the distance ratio parameter includes: sorting the distance ratios of all the image blocks, calculating a histogram and a cumulative function of the distance ratios, determining a distance ratio threshold value under the parameter, and outputting the image blocks with the distance ratios exceeding the threshold value, namely the changed image blocks.
Optionally, the comparing the changed image block with the image block of the reference data, and obtaining the precision evaluation result by using the error matrix includes:
wherein N is11Representing the number of the areas of which the reference data are marked as changes and the detection result is changed; n is a radical of12Representing the number of the areas of which the reference data are marked as unchanged and the detection result is changed; n is a radical of21Representing the number of the areas of which the reference data are marked as changed but the detection result is unchanged; n is a radical of22The representative reference data is marked as unchanged and the detection result is the number of the unchanged areas.
The invention provides a system for detecting the change of a double-time phase remote sensing image by adopting the method, which comprises the following steps:
the image blocking module is used for blocking and cutting the front and rear time phase images according to the size of the specified block to generate a front and rear time phase image blocking data set, a list file of blocking image information and a vector file for recording the positions and the characteristic information of the blocking images;
the high-dimensional feature extraction module is used for performing high-dimensional feature extraction on the front and rear time phase image block data sets by using a deep learning network model to generate a high-dimensional feature file;
the characteristic distance calculation module is used for calculating the characteristic distance of the high-dimensional characteristics of the front and rear time phase image blocks according to the list file of the block image information, the vector file for recording the block image positions and the characteristic information and the high-dimensional characteristic file to generate a characteristic distance file;
the change finding module is used for adjusting the characteristic distance parameters according to the characteristic distance file to obtain changed image blocks;
the precision evaluation module is used for comparing the changed image block with an image block of reference data, obtaining a precision evaluation result by using an error matrix and judging whether the precision evaluation result meets the requirement; if the requirement is met, outputting the changed image block; and if the characteristic distance parameter does not meet the requirement, adjusting the characteristic distance parameter to continue judging.
From the above, according to the double-temporal remote sensing image change detection method and system provided by the invention, the front and rear temporal images are cut in blocks, the high-dimensional feature extraction is performed on the front and rear temporal image block data sets by using the deep learning network model, the feature distance is calculated, the feature distance parameter is adjusted to obtain the changed image block, the changed image block is compared with the image block of the reference data, and the precision evaluation result is obtained by using the error matrix, so that the manual intervention is reduced, the change detection precision is improved, and an automatic means can be provided for the remote sensing image change detection.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flow chart of an embodiment of a method for detecting changes in a dual-temporal remote sensing image according to the present invention;
fig. 2 is a schematic diagram of a high-dimensional feature extraction flow of an embodiment of a double-temporal remote sensing image change detection method provided by the invention;
fig. 3 is a block data set diagram of a front time-phase remote sensing image according to an embodiment of a method and a system for detecting changes in a dual time-phase remote sensing image provided by the present invention;
fig. 4 is a block data set diagram of a rear time-phase remote sensing image according to an embodiment of a method and a system for detecting changes in a dual time-phase remote sensing image provided by the present invention;
fig. 5 is a schematic diagram of a block information list file according to an embodiment of a method and a system for detecting changes in a dual-temporal remote sensing image provided by the present invention;
fig. 6 is a schematic diagram of a block vector file according to an embodiment of a method and a system for detecting changes of a dual-temporal remote sensing image provided by the present invention;
fig. 7 is a schematic diagram of a characteristic distance file of an embodiment of a method and a system for detecting changes in a dual-temporal remote sensing image according to the present invention;
fig. 8 is a schematic diagram of a change finding result of an embodiment of a method and a system for detecting a change of a dual-temporal remote sensing image according to the present invention;
fig. 9 is a schematic diagram illustrating a change detection result and reference data superposition according to an embodiment of a method and a system for detecting a change of a dual-temporal remote sensing image provided by the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to specific embodiments and the accompanying drawings.
Based on the above purpose, the embodiments of the present invention provide a method and a system for detecting a change of a dual-temporal remote sensing image, which can improve the accuracy of change detection.
As shown in fig. 1, a schematic flow chart of an embodiment of a method for detecting changes in a dual-temporal remote sensing image according to the present invention is shown; the invention provides a double-time-phase remote sensing image change detection method, which comprises the following steps:
step 101: according to the size of the appointed block, performing block cutting on the front and rear time phase images to generate a front and rear time phase image block data set, a list file of block image information and a vector file for recording the position and the characteristic information of the block image;
step 102: performing high-dimensional feature extraction on the front and rear time phase image block data sets by using a deep learning network model to generate a high-dimensional feature file;
step 103: calculating the characteristic distance of the high-dimensional characteristics of the front and rear time phase image blocks according to the list file of the block image information, the vector file for recording the positions of the block images and the characteristic information and the high-dimensional characteristic file to generate a characteristic distance file;
step 104: adjusting the characteristic distance parameters according to the characteristic distance file to obtain changed image blocks;
step 105: comparing the changed image block with the image block of the reference data, and obtaining a precision evaluation result by using an error matrix, wherein the step 106 is as follows: judging whether the precision evaluation result meets the requirement or not; step 108: if the requirement is met, outputting the changed image block; step 107: and if the characteristic distance parameter does not meet the requirement, adjusting the characteristic distance parameter to continue judging.
Further, step 107: if too many image blocks with changed output and a large number of image blocks with changed misjudgment appear, increasing the characteristic distance parameter; if the number of the image blocks with output changes is too small, and a large number of image blocks with missed judgment changes appear, reducing the characteristic distance parameter and continuing the judgment.
In the process of cutting the front and rear time phase images into blocks, if the background pixel ratio of an image block at a certain position exceeds 50%, the front and rear time phase image blocks at the position are removed.
Step 102: the step of utilizing a deep learning network model to extract high-dimensional features of the front-time phase image block data set and the back-time phase image block data set and generating a high-dimensional feature file comprises the following steps: inputting accurate manual sampling data in advance, training the deep learning network model, optimizing parameters of the deep learning network model to obtain a fine-tuned deep learning network model, and performing high-dimensional feature extraction on the front-time phase image block data set and the rear-time phase image block data set by using the fine-tuned deep learning network model to generate a high-dimensional feature file.
As shown in fig. 2, it is a schematic diagram of a high-dimensional feature extraction flow of an embodiment of a double-temporal remote sensing image change detection method provided by the present invention; the high-dimensional feature extraction of the front and rear time phase image block data sets by using the finely adjusted deep learning network model comprises the following steps:
step 202: the method comprises the steps of performing first convolution processing, inputting 227 × 3 front and rear time phase image block data sets, performing convolution operation through 96 11 × 11 convolution kernels (step size is 4, and external expansion is 0) to generate a feature map of 55 × 96, performing activation processing on the feature map of 55 × 96 through a modified Linear Unit (ReLU), generating an activated feature map, performing pooling operation with the scale of 3 × 3 and the step size of 2 to generate a feature map of 27 × 96, and performing normalization processing to form a feature map of 27 × 96;
step 203: a second convolution process, in which feature maps 27 × 96 are input, and convolution operation is performed by 256 convolution kernels 5 × 5 (step size 1, outer expansion 2) to generate feature maps 27 × 256; then, performing activation processing on the feature map of 27 × 256 by a correction linear unit to generate an activated feature map, performing pooling operation with the scale of 3 × 3 and the step size of 2 to generate a feature map of 13 × 256, and finally performing normalization processing to form a feature map of 13 × 256;
step 204: the third convolution processing, inputting the feature map of 13 × 256, firstly performing convolution operation through 384 convolution kernels of 3 × 3 (step size is 1, and external expansion is 1) to generate the feature map of 13 × 384, and then performing activation processing on the feature map of 13 × 384 through a correction linear unit to generate an activation feature map, so as to form the feature map of 13 × 384;
step 205: a fourth convolution process, in which a feature map of 13 × 384 is input, and a convolution operation is performed by 384 convolution kernels of 3 × 3 (step size 1, outer expansion 1) to generate a feature map of 13 × 384; then, performing activation processing on the feature map of 13 × 384 through a correction linear unit to generate an activation feature map, and forming a feature map of 13 × 384;
step 206: a fifth convolution process, in which feature maps of 13 × 384 are input, and convolution operation is performed by 256 convolution kernels of 3 × 3 (step size 1, outer expansion 1) to generate feature maps of 13 × 256; then, performing activation processing on the feature map of 13 × 256 through a correction linear unit to generate an activation feature map; forming a feature map of 6 × 256 through pooling operation with a scale of 3 × 3 and a step size of 2;
step 207: a first full-connection process, wherein feature map data of 6 × 256 is fully connected with 4096 neurons, and a feature map of 4096 is output;
step 208: and a second full-connection process of fully connecting 4096 feature map data to 4096 neurons and outputting a 4096 feature map.
Step 103: the characteristic distance of the high-dimensional characteristic of the front and rear time phase image blocks comprises: the minimum distance, the related distance and the distance ratio of the high-dimensional features of the front and rear time phase image blocks.
The distance ratio (i, j) of the high-dimensional features of the front and rear time phase image blocks is as follows:
where cor _ dist (i, j) is the correlation distance of the high-dimensional features of the front and rear time phase image blocks, and min _ dist (i, j) is the minimum distance of the high-dimensional features of the front and rear time phase image blocks.
First, the distance of the image blocks is defined, the feature vector F of each image block is calculated based on step 102,
F＝(f1，f2，…，f4096)
wherein f isn(n-1, 2, …, 4096) represents the nth feature value of the image block.
Defining a distance dist (F) of any two image blocks1，F2)：
Preceding time image block feature vector:
wherein the content of the first and second substances,
representing the nth eigenvalue of the previous image block.
Later-time image block feature vector:
wherein the content of the first and second substances,
representing the nth eigenvalue of the later image block.
Secondly, calculating the image block distance between the corresponding position in the rear time phase image block and the front time phase image, namely the correlation distance cor _ dist (i, j) of the front and rear time phase image blocks:
and calculating the distances between the rear time phase image block and all the image blocks in the front time phase again, and finding out the minimum distance, namely the minimum distance min _ dist (i, j) of the rear time phase image block:
wherein: w is the number of rows of image blocks, H is the number of columns of image blocks, i, j respectively represent the serial numbers of the rows and the columns of the image blocks in the previous time, x, y respectively represent the serial numbers of the rows and the columns of the image blocks in the later time, i is more than or equal to 1 and less than or equal to W, and i is more than or equal to 1 and less than or equal to H.
And finally, calculating the ratio of the correlation distance to the minimum distance, namely the distance ratio (i, j) of the later image block:
step 104: adjusting the characteristic distance parameter according to the characteristic distance file to obtain the changed image block comprises the following steps: adjusting the minimum distance parameter, the related distance parameter and the distance ratio parameter according to the characteristic distance file;
and adjusting the characteristic distance parameter to find a changed image block, wherein the smaller the related distance is, the image block is not changed, and the image block with unchanged target and large time phase difference is judged by a distance ratio, and if the distance ratio is 1, the similarity is high, and the image block is probably not changed.
Adjusting the minimum distance parameter includes: sorting the minimum distances of all the image blocks, calculating a histogram and a cumulative function of the minimum distances, determining a minimum distance threshold value under the parameter, and outputting the image blocks with the minimum distances exceeding the threshold value, namely the changed image blocks; the parameter represents the percentage of output image blocks based on the minimum distance feature, the adjustment range is 0-100, when the parameter is 0, all image blocks are output, and when the parameter is 100, all image blocks are not output.
Adjusting the relevant distance parameter includes: sorting the correlation distances of all the image blocks, calculating a histogram and a cumulative function of the correlation distances, determining a correlation distance threshold value under the parameter, and outputting the image blocks of which the correlation distances exceed the threshold value, namely the changed image blocks; the parameter represents the percentage of the output image blocks based on the relevant distance features, the adjustment range is 0-100, when the parameter is 0, all image blocks are output, and when the parameter is 100, all image blocks are not output.
Adjusting the distance ratio parameter includes: sorting the distance ratios of all the image blocks, calculating a histogram and a cumulative function of the distance ratios, determining a distance ratio threshold value under the parameter, and outputting the image blocks with the distance ratios exceeding the threshold value, namely the changed image blocks. The parameter represents the percentage of output image blocks based on the distance ratio, the adjustment range is 0-100, when the parameter is 0, all image blocks are output, and when the parameter is 100, all image blocks are not output.
Step 105: the comparing the changed image block with the image block of the reference data, and obtaining the precision evaluation result by using the error matrix comprises:
TABLE 1 error matrix calculation
Wherein N is11Representing the number of the areas of which the reference data are marked as changes and the detection result is changed; n is a radical of12Representing the number of the areas of which the reference data are marked as unchanged and the detection result is changed; n is a radical of21Representing the number of the areas of which the reference data are marked as changed but the detection result is unchanged; n is a radical of22The representative reference data is marked as unchanged and the detection result is the number of the unchanged areas.
The invention provides a system for detecting the change of a double-time phase remote sensing image by adopting the method, which comprises the following steps:
the image blocking module is used for blocking and cutting the front and rear time phase images according to the size of the specified block to generate a front and rear time phase image blocking data set, a list file of blocking image information and a vector file for recording the positions and the characteristic information of the blocking images;
the high-dimensional feature extraction module is used for performing high-dimensional feature extraction on the front and rear time phase image block data sets by using a deep learning network model to generate a high-dimensional feature file;
the characteristic distance calculation module is used for calculating the characteristic distance of the high-dimensional characteristics of the front and rear time phase image blocks according to the list file of the block image information, the vector file for recording the block image positions and the characteristic information and the high-dimensional characteristic file to generate a characteristic distance file;
the change finding module is used for adjusting the characteristic distance parameters according to the characteristic distance file to obtain changed image blocks;
the precision evaluation module is used for comparing the changed image block with an image block of reference data, obtaining a precision evaluation result by using an error matrix and judging whether the precision evaluation result meets the requirement; if the requirement is met, outputting the changed image block; and if the characteristic distance parameter does not meet the requirement, adjusting the characteristic distance parameter to continue judging.
As a preferred embodiment, the high-resolution remote sensing images of the new region of the world city in sichuan area 2015 and 2016 are taken as an embodiment, and the system for change detection of double-time-phase remote sensing images provided by the invention is adopted to perform a deep learning change detection experiment. The image of the new area of the Sichuan Tianfu comprises R, G, B three wave bands, the resolution is 1 meter, and the specific steps are as follows:
step 1: the remote sensing images of the new region 2015 and 1026 of the sikawa prefecture are loaded, 3 displayed wave bands (default R, G, B) are selected, 256 × 256 blocking operations are performed on the two-time phase remote sensing images, as shown in fig. 3 and 4, a block data set (361 blocks are cut out from each image) of the two-time phase remote sensing images, as shown in fig. 5, a list file of blocking image information, and as shown in fig. 6, a vector file for recording the positions and feature information of the blocking images are respectively generated.
Step 2: accurate manual sampling data is input in advance, deep learning network models (such as AlexNet, VGG, GoogleNet, ResNet and the like) are trained, and model parameters are finely adjusted, so that the deep learning network models are suitable for the embodiment. And then loading the block data sets of the two-time-phase remote sensing images and the list files of the block image information in the new Sichuan Tianfu area 2015 and 2016 generated in the step 1, and respectively performing feature extraction on the block data sets of the two-time-phase remote sensing images by using a fine-tuned deep learning network model to generate 4096-dimensional feature files of corresponding records.
And step 3: and (3) loading the vector file for recording the position and the characteristic information of the blocked image generated in the step (1) and the record 4096-dimensional characteristic file generated in the step (2), and calculating three characteristics of the minimum distance, the related distance and the distance ratio, so as to generate a characteristic distance file as shown in fig. 7.
And 4, step 4: and (4) loading the characteristic distance file obtained in the step (3), and adjusting the proportion of the minimum distance, the related distance and the distance ratio within the threshold value [0,100] to generate the changed image block shown in the figure 8.
And 5: by comparing the changed image block with the image block of the reference data, an accuracy evaluation result is obtained by using an error matrix, as shown in fig. 9, a schematic diagram of the change detection result and the reference data is superimposed.
TABLE 2 evaluation results of accuracy
The overall accuracy is 85.22%, the omission factor is 13.04%, and the false detection rate is 1.74%, thus verifying the effectiveness of the method.
It can be seen from the foregoing embodiments that, according to the method and system for detecting changes in a dual-temporal remote sensing image provided by the embodiments of the present invention, a front temporal image and a rear temporal image are cut into blocks, a deep learning network model is used to perform high-dimensional feature extraction on a front temporal image block data set and a rear temporal image block data set, feature distance calculation is performed, feature distance parameters are adjusted to obtain changed image blocks, the changed image blocks are compared with image blocks of reference data, and an accuracy evaluation result is obtained by using an error matrix, so that manual intervention is reduced, the accuracy of change detection is improved, and an automatic means can be provided for remote sensing image change detection.
Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, is limited to these examples; within the idea of the invention, also features in the above embodiments or in different embodiments may be combined, steps may be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.
The embodiments of the invention are intended to embrace all such alternatives, modifications and variances that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, substitutions, improvements and the like that may be made without departing from the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (8)
1. A double-time phase remote sensing image change detection method is characterized by comprising the following steps:
according to the size of the appointed block, performing block cutting on the front and rear time phase images to generate a front and rear time phase image block data set, a list file of block image information and a vector file for recording the position and the characteristic information of the block image;
performing high-dimensional feature extraction on the front and rear time phase image block data sets by using a deep learning network model to generate a high-dimensional feature file;
calculating the characteristic distance of the high-dimensional characteristics of the front and rear time phase image blocks according to the list file of the block image information, the vector file for recording the positions of the block images and the characteristic information and the high-dimensional characteristic file to generate a characteristic distance file;
adjusting the characteristic distance parameters according to the characteristic distance file to obtain changed image blocks;
comparing the changed image block with an image block of reference data, obtaining a precision evaluation result by using an error matrix, and judging whether the precision evaluation result meets the requirement; if the requirement is met, outputting the changed image block; and if the characteristic distance parameter does not meet the requirement, adjusting the characteristic distance parameter to continue judging.
2. The method for detecting the change of the double-temporal remote sensing image according to claim 1, wherein in the process of segmenting and cutting the front and rear temporal images, if the background pixel proportion of an image block at a certain position is more than 50%, the front and rear temporal image blocks at the certain position are removed.
3. The method for detecting changes of double-temporal remote sensing images according to claim 1, wherein the step of performing high-dimensional feature extraction on the front and rear temporal image block data sets by using a deep learning network model to generate a high-dimensional feature file comprises the following steps: inputting accurate manual sampling data in advance, training the deep learning network model, optimizing parameters of the deep learning network model to obtain a fine-tuned deep learning network model, and performing high-dimensional feature extraction on the front-time phase image block data set and the rear-time phase image block data set by using the fine-tuned deep learning network model to generate a high-dimensional feature file.
4. The method for detecting changes of double-temporal remote sensing images according to claim 3, wherein the step of performing high-dimensional feature extraction on the front and rear temporal image block data sets by using the fine-tuned deep learning network model comprises the following steps:
the first convolution processing, inputting 227 x 3 front and rear time phase image block data sets, firstly performing convolution operation through 96 11 x 11 convolution kernels to generate a 55 x 96 feature map, then performing activation processing on the 55 x 96 feature map through a correction linear unit to generate an activation feature map, performing pooling operation with the scale of 3 x 3 and the step size of 2 to generate a 27 x 96 feature map, and finally performing normalization processing to form a 27 x 96 feature map;
a second convolution process, wherein a feature map of 27 × 96 is input, and a convolution operation is performed through 256 convolution kernels of 5 × 5 to generate a feature map of 27 × 256; then, performing activation processing on the feature map of 27 × 256 by a correction linear unit to generate an activated feature map, performing pooling operation with the scale of 3 × 3 and the step size of 2 to generate a feature map of 13 × 256, and finally performing normalization processing to form a feature map of 13 × 256;
the third convolution processing is carried out, 13 × 256 feature maps are input, firstly, the convolution operation is carried out through 384 convolution kernels with 3 × 3 to generate 13 × 384 feature maps, then, the activation processing is carried out on the 13 × 384 feature maps through a correction linear unit to generate activation feature maps, and the 13 × 384 feature maps are formed;
a fourth convolution process, wherein a feature map of 13 × 384 is input, and a convolution operation is performed through 384 convolution kernels of 3 × 3 to generate a feature map of 13 × 384; then, performing activation processing on the feature map of 13 × 384 through a correction linear unit to generate an activation feature map, and forming a feature map of 13 × 384;
in the fifth convolution processing, a feature map of 13 × 384 is input, and convolution operation is performed through 256 convolution kernels of 3 × 3 to generate a feature map of 13 × 256; then, performing activation processing on the feature map of 13 × 256 through a correction linear unit to generate an activation feature map; forming a feature map of 6 × 256 through pooling operation with a scale of 3 × 3 and a step size of 2;
a first full-connection process, wherein feature map data of 6 × 256 is fully connected with 4096 neurons, and a feature map of 4096 is output;
and a second full-connection process of fully connecting 4096 feature map data to 4096 neurons and outputting a 4096 feature map.
5. The method for detecting changes of double-temporal remote sensing images according to claim 1, wherein the characteristic distance of the high-dimensional characteristics of the front and rear temporal image blocks comprises: the minimum distance, the related distance and the distance ratio of the high-dimensional features of the front and rear time phase image blocks,
the distance ratio (i, j) of the high-dimensional features of the front and rear time phase image blocks is as follows:
6. The method for detecting changes in double-temporal remote sensing images according to claim 5, wherein adjusting the characteristic distance parameters according to the characteristic distance file to obtain changed image blocks comprises: adjusting the minimum distance parameter, the related distance parameter and the distance ratio parameter according to the characteristic distance file;
adjusting the minimum distance parameter includes: sorting the minimum distances of all the image blocks, calculating a histogram and a cumulative function of the minimum distances, determining a minimum distance threshold value under the parameter, and outputting the image blocks with the minimum distances exceeding the threshold value, namely the changed image blocks;
adjusting the relevant distance parameter includes: sorting the correlation distances of all the image blocks, calculating a histogram and a cumulative function of the correlation distances, determining a correlation distance threshold value under the parameter, and outputting the image blocks of which the correlation distances exceed the threshold value, namely the changed image blocks;
adjusting the distance ratio parameter includes: sorting the distance ratios of all the image blocks, calculating a histogram and a cumulative function of the distance ratios, determining a distance ratio threshold value under the parameter, and outputting the image blocks with the distance ratios exceeding the threshold value, namely the changed image blocks.
7. The method for detecting changes in double-temporal remote sensing images according to claim 1, wherein the obtaining of the precision evaluation result by comparing the changed image block with the image block of the reference data using the error matrix comprises:
wherein N is11Representing the number of the areas of which the reference data are marked as changes and the detection result is changed; n is a radical of12Representing the number of the areas of which the reference data are marked as unchanged and the detection result is changed; n is a radical of21Representing reference data marked as changed but detectedThe result is the number of unchanged areas; n is a radical of22The representative reference data is marked as unchanged and the detection result is the number of the unchanged areas.
8. A system for change detection of a two-time-phase remote sensing image by using the method of any one of claims 1 to 7, comprising:
the image blocking module is used for blocking and cutting the front and rear time phase images according to the size of the specified block to generate a front and rear time phase image blocking data set, a list file of blocking image information and a vector file for recording the positions and the characteristic information of the blocking images;
the high-dimensional feature extraction module is used for performing high-dimensional feature extraction on the front and rear time phase image block data sets by using a deep learning network model to generate a high-dimensional feature file;
the characteristic distance calculation module is used for calculating the characteristic distance of the high-dimensional characteristics of the front and rear time phase image blocks according to the list file of the block image information, the vector file for recording the block image positions and the characteristic information and the high-dimensional characteristic file to generate a characteristic distance file;
the change finding module is used for adjusting the characteristic distance parameters according to the characteristic distance file to obtain changed image blocks;
the precision evaluation module is used for comparing the changed image block with an image block of reference data, obtaining a precision evaluation result by using an error matrix and judging whether the precision evaluation result meets the requirement; if the requirement is met, outputting the changed image block; and if the characteristic distance parameter does not meet the requirement, adjusting the characteristic distance parameter to continue judging.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810113902.4A CN108446588B (en) | 2018-02-05 | 2018-02-05 | Double-temporal remote sensing image change detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810113902.4A CN108446588B (en) | 2018-02-05 | 2018-02-05 | Double-temporal remote sensing image change detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108446588A CN108446588A (en) | 2018-08-24 |
| CN108446588B true CN108446588B (en) | 2020-09-15 |
Family
ID=63191808
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810113902.4A Expired - Fee Related CN108446588B (en) | 2018-02-05 | 2018-02-05 | Double-temporal remote sensing image change detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108446588B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110334581B (en) * | 2019-05-09 | 2022-02-18 | 宁波市测绘和遥感技术研究院 | Multi-source remote sensing image change detection method |
| CN111553272B (en) * | 2020-04-28 | 2022-05-06 | 上海市测绘院 | High-resolution satellite optical remote sensing image building change detection method based on deep learning |
| CN112488025B (en) * | 2020-12-10 | 2022-06-14 | 武汉大学 | Double-temporal remote sensing image semantic change detection method based on multi-modal feature fusion |
| CN115311550B (en) * | 2022-06-22 | 2024-05-28 | 中国科学院空天信息创新研究院 | Remote sensing image semantic change detection method and device, electronic equipment and storage medium |
| CN117576517B (en) * | 2024-01-15 | 2024-04-12 | 西南交通大学 | Optical remote sensing image self-supervision contrast learning change detection method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015042772A1 (en) * | 2013-09-24 | 2015-04-02 | 中国科学院自动化研究所 | Remote sensing image salient object change detection method |
| CN105069811A (en) * | 2015-09-08 | 2015-11-18 | 中国人民解放军重庆通信学院 | Multi-temporal remote sensing image change detection method |
| CN105225227A (en) * | 2015-09-07 | 2016-01-06 | 中国测绘科学研究院 | The method and system that remote sensing image change detects |
| CN106844739A (en) * | 2017-02-14 | 2017-06-13 | 中国科学院遥感与数字地球研究所 | A kind of Remote Sensing Imagery Change information retrieval method based on neutral net coorinated training |
| CN107358176A (en) * | 2017-06-26 | 2017-11-17 | 武汉大学 | Sorting technique based on high score remote sensing image area information and convolutional neural networks |
| CN107610164A (en) * | 2017-09-11 | 2018-01-19 | 北京空间飞行器总体设计部 | A kind of No. four Image registration methods of high score based on multiple features mixing |
-
2018
- 2018-02-05 CN CN201810113902.4A patent/CN108446588B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015042772A1 (en) * | 2013-09-24 | 2015-04-02 | 中国科学院自动化研究所 | Remote sensing image salient object change detection method |
| CN105225227A (en) * | 2015-09-07 | 2016-01-06 | 中国测绘科学研究院 | The method and system that remote sensing image change detects |
| CN105069811A (en) * | 2015-09-08 | 2015-11-18 | 中国人民解放军重庆通信学院 | Multi-temporal remote sensing image change detection method |
| CN106844739A (en) * | 2017-02-14 | 2017-06-13 | 中国科学院遥感与数字地球研究所 | A kind of Remote Sensing Imagery Change information retrieval method based on neutral net coorinated training |
| CN107358176A (en) * | 2017-06-26 | 2017-11-17 | 武汉大学 | Sorting technique based on high score remote sensing image area information and convolutional neural networks |
| CN107610164A (en) * | 2017-09-11 | 2018-01-19 | 北京空间飞行器总体设计部 | A kind of No. four Image registration methods of high score based on multiple features mixing |
Non-Patent Citations (5)
| Title |
|---|
| 《An ObjectT-Based Semantic Classification Method for High Resolution Remote Sensing Imagery Using Ontology》;Haiyan Gu et al;《Remote Sensing》;20170330;第09卷(第04期);第329-336页 * |
| 《Change detection and dynamic analysis based on remote sensing images》;G.Luzi et al;《2013 3rd International Workshop on Image and Data Fusion(IWIDF2013)》;20131203;185-188 * |
| 《Fuzzy neural network architecture for change detection in remotely sensed imagery》;H.Nemmour et al;《Journal International Journal of Remote Sensing》;20070222;第27卷(第04期);第705-717页 * |
| 《基于人工神经网络的遥感影像变化信息提取方法研究》;邵飞;《中国优秀博硕士学位论文全文数据库 (硕士) 基础科学辑》;20070215(第02期);A011-71 * |
| 《遥感影像变化检测方法及应用研究》;张家琦;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160115(第01期);I140-364 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108446588A (en) | 2018-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108446588B (en) | Double-temporal remote sensing image change detection method and system | |
| CN111080629B (en) | Method for detecting image splicing tampering | |
| CN112733749B (en) | Real-time pedestrian detection method integrating attention mechanism | |
| CN110310264A (en) | A kind of large scale object detection method, device based on DCNN | |
| CN112801169B (en) | Camouflage target detection method, system, device and storage medium based on improved YOLO algorithm | |
| CN111461083A (en) | Rapid vehicle detection method based on deep learning | |
| CN110889863B (en) | Target tracking method based on target perception correlation filtering | |
| CN111382690B (en) | Vehicle re-identification method based on multi-loss fusion model | |
| CN111709313B (en) | Pedestrian re-identification method based on local and channel combination characteristics | |
| CN107784288A (en) | A kind of iteration positioning formula method for detecting human face based on deep neural network | |
| CN105243154A (en) | Remote sensing image retrieval method and system based on significant point characteristics and spare self-encodings | |
| CN112906816B (en) | Target detection method and device based on optical differential and two-channel neural network | |
| CN111524117A (en) | Tunnel surface defect detection method based on characteristic pyramid network | |
| CN115937697A (en) | Remote sensing image change detection method | |
| CN110163294A (en) | Remote Sensing Imagery Change method for detecting area based on dimensionality reduction operation and convolutional network | |
| CN114419406A (en) | Image change detection method, training method, device and computer equipment | |
| CN117037004A (en) | Unmanned aerial vehicle image detection method based on multi-scale feature fusion and context enhancement | |
| CN110866931A (en) | Image segmentation model training method and classification-based enhanced image segmentation method | |
| CN113673618A (en) | Tobacco insect target detection method fused with attention model | |
| CN112464725A (en) | First arrival picking method and device based on deep learning network | |
| CN110472092B (en) | Geographical positioning method and system of street view picture | |
| CN116311004A (en) | Video moving target detection method based on sparse optical flow extraction | |
| CN116343078A (en) | Target tracking method, system and equipment based on video SAR | |
| CN103903258A (en) | Method for detecting changes of remote sensing image based on order statistic spectral clustering | |
| CN113723469A (en) | Interpretable hyperspectral image classification method and device based on space-spectrum combined network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200915 Termination date: 20210205 |