CN115830297A - Processing method of remote sensing image change detection sample library - Google Patents
Processing method of remote sensing image change detection sample library Download PDFInfo
- Publication number
- CN115830297A CN115830297A CN202211541397.6A CN202211541397A CN115830297A CN 115830297 A CN115830297 A CN 115830297A CN 202211541397 A CN202211541397 A CN 202211541397A CN 115830297 A CN115830297 A CN 115830297A
- Authority
- CN
- China
- Prior art keywords
- image
- remote sensing
- sensing image
- change detection
- sample library
- 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.)
- Pending
Links
Images
Abstract
The invention discloses a processing method of a remote sensing image change detection sample library, which comprises the following steps: acquiring a multi-source remote sensing image in an experimental area; according to the multi-source remote sensing image, image preprocessing is carried out on the image to be processed, and a preprocessed target image is obtained; marking a change area in the target image according to the change condition between the acquired historical images in different time phases and the target image; sampling the target image, adjusting the size and sliding interval of image blocks in the remote sensing image, and constructing to obtain a self-constructed sample library; and training the sample data in the self-built sample library by adopting a deep learning network model to obtain a change detection model. The embodiment of the invention has high efficiency, effectively utilizes the small sample library to fully extract the deep characteristics of the remote sensing image change class, realizes the optimization and updating of the network parameters, and can be widely applied to the technical field of remote sensing image information processing.
Description
Technical Field
The invention relates to the technical field of remote sensing image information processing, in particular to a processing method of a remote sensing image change detection type sample library.
Background
While the remote sensing technology is rapidly developed, the remote sensing images are applied more and more, and a plurality of good research results are obtained, but some disadvantages exist. On one hand, the remote sensing image is difficult to obtain, the shooting cost is high, and the remote sensing image can be put into use only through a complicated processing flow; on the other hand, the obtained samples are labeled less, and a worker needs to spend a great deal of energy on labeling the samples. These problems lead to a small number of labeled samples, poor quality, insufficient sample diversity, and thus a negative impact on subsequent research tasks.
The deep learning has high requirements on the number of samples, the precision of a detection model generally depends on the number of samples, and a larger sample scale is a precondition for achieving better detection precision. However, in practice, the number of samples of the changed building is often insufficient or the quality of the samples is not good enough, and a lot of labor and time are required to mark the samples of the changed building. In addition, in practical situations, the proportion of the variable buildings in cities is low, and the construction of a large sample library is difficult. Therefore, a change sample library needs to be constructed, data enhancement is introduced, sample quality is improved, and performance and robustness of the model are improved.
Disclosure of Invention
In view of this, the embodiment of the present invention provides an efficient processing method for a remote sensing image change detection type sample library, so as to effectively utilize a small sample library to fully extract deep features of remote sensing image change types, and implement optimization and update of network parameters.
One aspect of the embodiments of the present invention provides a method for processing a remote sensing image change detection type sample library, including:
acquiring a multi-source remote sensing image in an experimental area;
according to the multi-source remote sensing image, image preprocessing is carried out on the image to be processed, and a preprocessed target image is obtained;
marking a change area in the target image according to the change condition between the acquired historical images in different time phases and the target image;
sampling the target image, adjusting the size and sliding interval of image blocks in the remote sensing image, and constructing to obtain a self-constructed sample library;
training sample data in the self-built sample library by adopting a deep learning network model to obtain a change detection model; the change detection model is used for carrying out change detection on the remote sensing image; and marking a changed area on the remote sensing image in the sample data.
Optionally, the image preprocessing is performed on the image to be processed according to the multi-source remote sensing image to obtain a preprocessed target image, and the method includes:
performing histogram matching processing on two remote sensing images in the same region at different time phases;
carrying out image registration processing on the two remote sensing images of the front time sequence and the rear time sequence after the histogram matching processing;
and denoising the remote sensing image subjected to the image registration processing.
Optionally, the performing image registration processing on the two remote sensing images of the time sequence before and after the histogram matching processing includes:
using the previous historical period image in the two remote sensing images as a reference image to establish a reference coordinate system, and using one image in the later historical period as an image to be registered to establish an image coordinate system to be registered;
obtaining characteristic points in the remote sensing image through a Forstner angular point operator, screening and filtering the characteristic points through a first-order polynomial fitting global transformation geometric model, matching the characteristic points by using a cross-correlation algorithm, and determining a connection point;
determining transformation model parameters required by registration of remote sensing images of different time sequences according to the relation between the connection points, and constructing a transformation model;
and according to the transformation model, carrying out geometric coordinate transformation and interpolation resampling operation on the remote sensing image to be registered to obtain a final registered image.
Optionally, the marking a change area in the target image according to a change between the acquired historical images of different time phases and the target image includes:
carrying out blocking processing on the multi-temporal remote sensing image, and filtering out regions without changes in the remote sensing image;
selecting a target double-time phase image, importing the target double-time phase image into labeling software, configuring the same reference space for the target double-time phase image in the labeling software, and then building a label vector layer;
comparing the new time phase image layer with the old time phase image layer, determining and marking a change area in the image, and constructing a label layer of the change area;
perfecting and supplementing an attribute table of a label layer in a change area;
and converting the completely supplemented label layer into a grid and outputting the grid to obtain a new double-time-phase remote sensing image, an old double-time-phase remote sensing image and a changed label grid map.
Optionally, before the step of training the sample data in the self-built sample library by using the deep learning network model to obtain the change detection model, the method further includes the following steps:
sample data in a self-built sample library is enhanced by carrying out sampling operation or geometric transformation operation on the sample data in the self-built sample library;
and carrying out normalization processing on the sample data in the self-built sample library.
Optionally, the method further includes a step of performing model quality evaluation on the change detection model, and retraining the change detection model according to a result of the model quality evaluation, where the step specifically includes:
calculating a precision evaluation index, a recall rate evaluation index, an F1 score evaluation index and an evaluation index of an IOU function of the change detection model;
according to the precision evaluation index, the recall rate evaluation index, the F1 score evaluation index and the evaluation index of the IOU function, when the numerical value of each evaluation index does not meet the quality requirement of the model, sample data in the self-built sample library is adjusted, and the change detection model is retrained until each evaluation index of the change detection model meets the quality requirement of the model.
In another aspect of the embodiments of the present invention, a device for processing a remote sensing image change detection type sample library is further provided, including:
the system comprises a first module, a second module and a third module, wherein the first module is used for acquiring a multi-source remote sensing image in an experimental area;
the second module is used for carrying out image preprocessing on the image to be processed according to the multi-source remote sensing image to obtain a preprocessed target image;
a third module, configured to mark a change area in the target image according to a change condition between the acquired historical images at different time phases and the target image;
the fourth module is used for carrying out sampling operation on the target image, adjusting the size and the sliding interval of an image block in the remote sensing image and constructing to obtain a self-constructed sample library;
the fifth module is used for training the sample data in the self-built sample library by adopting a deep learning network model to obtain a change detection model; the change detection model is used for carrying out change detection on the remote sensing image; and marking a changed area on the remote sensing image in the sample data.
Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;
the memory is used for storing programs;
the processor executes the program to implement the method as described above.
Still another aspect of embodiments of the present invention provides a computer-readable storage medium, which stores a program,
the program is executed by a processor to implement the method as described above.
The embodiment of the invention also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and the computer instructions executed by the processor cause the computer device to perform the foregoing method.
The embodiment of the invention obtains the multi-source remote sensing image in the experimental area; according to the multi-source remote sensing image, image preprocessing is carried out on the image to be processed, and a preprocessed target image is obtained; marking a change area in the target image according to the change condition between the acquired historical images in different time phases and the target image; sampling the target image, adjusting the size and the sliding interval of an image block in the remote sensing image, and constructing to obtain a self-constructed sample library; training sample data in the self-built sample library by adopting a deep learning network model to obtain a change detection model; the embodiment of the invention has high efficiency, and effectively utilizes the small sample library to fully extract the deep characteristics of the remote sensing image change class, thereby realizing the optimization and the update of the network parameters.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a flowchart illustrating the overall steps provided by an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
In view of the problems in the prior art, an aspect of the embodiments of the present invention provides a method for processing a remote sensing image change detection type sample library, including:
acquiring a multi-source remote sensing image in an experimental area;
according to the multi-source remote sensing image, image preprocessing is carried out on the image to be processed, and a preprocessed target image is obtained;
marking a change area in the target image according to the change condition between the acquired historical images in different time phases and the target image;
sampling the target image, adjusting the size and sliding interval of image blocks in the remote sensing image, and constructing to obtain a self-constructed sample library;
training sample data in the self-built sample library by adopting a deep learning network model to obtain a change detection model; the change detection model is used for carrying out change detection on the remote sensing image; and marking the remote sensing image in the sample data with a changed area.
Optionally, the image preprocessing is performed on the image to be processed according to the multi-source remote sensing image to obtain a preprocessed target image, and the method includes:
performing histogram matching processing on two remote sensing images in the same region at different time phases;
carrying out image registration processing on the two remote sensing images of the front time sequence and the rear time sequence after the histogram matching processing;
and denoising the remote sensing image subjected to the image registration processing.
Optionally, the performing image registration processing on the two remote sensing images of the time sequence before and after the histogram matching processing includes:
using the previous historical period image in the two remote sensing images as a reference image to establish a reference coordinate system, and using one image in the later historical period as an image to be registered to establish an image coordinate system to be registered;
obtaining characteristic points in the remote sensing image through a Forstner corner operator, screening and filtering the characteristic points through a first-order polynomial fitting global transformation geometric model, matching the characteristic points by using a cross-correlation algorithm, and determining a connection point;
determining transformation model parameters required by registration of remote sensing images of different time sequences according to the relation between the connection points, and constructing a transformation model;
and according to the transformation model, carrying out geometric coordinate transformation and interpolation resampling operation on the remote sensing image to be registered to obtain a final registered image.
Optionally, the marking out a change area in the target image according to a change condition between the acquired historical images of different time phases and the target image includes:
the multi-temporal remote sensing image is processed in a blocking mode, and areas which are not changed in the remote sensing image are filtered;
selecting a target double-time phase image, importing the target double-time phase image into labeling software, configuring the same reference space for the target double-time phase image in the labeling software, and then building a label vector layer;
comparing the new time phase image layer with the old time phase image layer, determining and marking a change area in the image, and constructing a label image layer of the change area;
perfecting and supplementing an attribute table of a label layer in a change area;
and converting the completely supplemented label layer into a grid and outputting the grid to obtain a new double-time-phase remote sensing image, an old double-time-phase remote sensing image and a changed label grid map.
Optionally, before the step of training the sample data in the self-built sample library by using the deep learning network model to obtain the change detection model, the method further includes the following steps:
sample data in a self-built sample library is enhanced by carrying out sampling operation or geometric transformation operation on the sample data in the self-built sample library;
and carrying out normalization processing on the sample data in the self-built sample library.
Optionally, the method further includes a step of performing model quality evaluation on the change detection model, and retraining the change detection model according to a result of the model quality evaluation, where the step specifically includes:
calculating a precision evaluation index, a recall rate evaluation index, an F1 score evaluation index and an evaluation index of an IOU function of the change detection model;
according to the precision evaluation index, the recall rate evaluation index, the F1 score evaluation index and the evaluation index of the IOU function, when the numerical value of each evaluation index does not meet the quality requirement of the model, sample data in the self-built sample library is adjusted, and the change detection model is retrained until each evaluation index of the change detection model meets the quality requirement of the model.
Another aspect of the embodiments of the present invention further provides a processing apparatus for a remote sensing image change detection type sample library, including:
the system comprises a first module, a second module and a third module, wherein the first module is used for acquiring a multi-source remote sensing image in an experimental area;
the second module is used for carrying out image preprocessing on the image to be processed according to the multi-source remote sensing image to obtain a preprocessed target image;
a third module, configured to mark a change area in the target image according to a change condition between the acquired historical images at different time phases and the target image;
the fourth module is used for carrying out sampling operation on the target image, adjusting the size and the sliding interval of an image block in the remote sensing image and constructing to obtain a self-constructed sample library;
the fifth module is used for training the sample data in the self-built sample library by adopting a deep learning network model to obtain a change detection model; the change detection model is used for carrying out change detection on the remote sensing image; and marking a changed area on the remote sensing image in the sample data.
Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;
the memory is used for storing programs;
the processor executes the program to implement the method as described above.
Still another aspect of embodiments of the present invention provides a computer-readable storage medium, which stores a program,
the program is executed by a processor to implement the method as described above.
The embodiment of the invention also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and the computer instructions executed by the processor cause the computer device to perform the foregoing method.
The following detailed description of the embodiments of the invention is provided in conjunction with the accompanying drawings:
as shown in fig. 1, the method for constructing and enhancing a remote sensing image change detection type sample library provided by the invention comprises the following steps:
step 1, acquiring a remote sensing image of an experimental area and preparing;
step 2, preprocessing the remote sensing image;
step 3, marking a change target;
step 4, sampling a data set;
step 5, dividing a data set;
step 6, enhancing the data of the sample library;
step 7, normalizing the data of the sample database;
step 8, training a model;
and 9, evaluating the precision.
Optionally, in step 1, the preparation work includes: acquiring multi-source remote sensing images in an experimental area, comparing images at different time phases to determine the change type of a detection target, and cleaning invalid and low-quality data;
optionally, in step 2, the remote sensing image preprocessing includes:
step 2.1, performing histogram matching on two remote sensing images in the same region at different time phases to enable the two remote sensing images to have similar colors;
step 2.2, carrying out image registration on the two remote sensing images of the front time sequence and the rear time sequence after the histogram matching;
step 2.3, denoising the remote sensing image;
optionally, in step 3, the step of labeling the variation target specifically includes:
3.1, partitioning the multi-temporal remote sensing image, and filtering out regions without changes;
step 3.2, selecting double time phase images to import into labeling software, setting the same reference space, and building a new label vector layer;
step 3.3, comparing the new time phase image layer with the old time phase image layer, and identifying and marking a change area;
step 3.4, perfecting an attribute table of the label layer;
and 3.5, converting the label vector layer into a grid and exporting.
At the moment, the data set comprises the remote sensing images of the new and old double time phases and the change label raster image.
Optionally, in step 4, a sliding window sampling mode is selected to perform non-overlapping and non-interval sampling on the data set.
Optionally, in step 5, the training set, the verification set and the test set are allocated in a certain manner according to a division ratio.
Optionally, in step 6, the sample database data enhancement method mainly adopts horizontal flipping, vertical flipping, transposition, random rotation by 90 ° and affine transformation.
In step 7, the sample data is normalized.
8, training a model, namely training samples in the sample library by using a deep learning network model to obtain a change detection model;
and 9, precision evaluation, namely performing data prediction test on the image data to be tested input into the change detection model, calculating an evaluation function of the test, jumping to the next step if the evaluation function values reach the standard, and adjusting the parameters of the deep learning network model to return to the model training step and repeating the iterative training if the evaluation function values do not reach the standard.
Taking a specific application scenario as an example, in the specific scenario of monitoring building changes, the method of the present invention is adopted to perform processing, and the method includes the following steps:
1, acquiring a remote sensing image of an experimental area and preparing; 2, preprocessing a remote sensing image; 3, marking a change target; 4, sampling a data set; 5, dividing a data set; 6, enhancing data of a sample library; 7, normalizing the sample data; 8, training a model; and 9, evaluating the precision.
Specifically, step 1, the preparation work includes: and acquiring multi-source remote sensing images in the experimental area, comparing the images at different time phases to determine the change type of the detection target, and cleaning invalid and low-quality data.
The data source selects a Pleiades satellite remote sensing image of city A in 12 months in 2019 to 2 months in 2022 and a Worldview-2 optical remote sensing image in 2010 to 2018. The areas in the Pleiades satellite images mainly consist of urban built-up areas, seas and mountainous areas, and the building changes mainly include old house renovation areas, newly built residential areas, business areas, small garages, a small number of large warehouses and factories. The area in the Worldview-2 satellite image is mainly a development area, and due to the fact that the image time interval is long, the change range of buildings in the acquisition area is large, a plurality of buildings which are already built or are under construction are provided, the types of the buildings comprise high-rise apartments, villa residences, travel facilities, commercial hotels, large factories and the like, and the mountainous area is not changed obviously.
Step 2, the remote sensing image preprocessing comprises the following steps:
step 2.1, the embodiment selects two remote sensing images in the same area, takes one image in the previous historical period as a reference image, and performs histogram matching on one image in the later historical period, so that the two high-resolution remote sensing images with different time sequences have similar colors, and the influence of detecting the accuracy of the building change area of the high-resolution remote sensing image is reduced;
the image histogram matching is to objectively reflect the gray level distribution of an image by using a gray level histogram of the image, make the image similar to the color of another image by converting the histogram of the image, and then reduce the radiation difference between the images, thereby achieving the purpose of radiation matching. The formula of the histogram matching algorithm is as follows:
G(z q )＝s k
z q ＝G -1 (s k ) Equation 1
Wherein, the gray-scale variables of the original image T and the target image G are r k And z q ，p r (r) and p z (z) representing a corresponding probability density function; where MN is the total number of pixels of the image, n j Is provided with a gray value r j L is the number of possible gray levels in the image.
Step 2.2, the embodiment registers the collected remote sensing images and comprises four steps:
(1) A reference coordinate system is established. And taking the image in the previous historical period of the two remote sensing images as a reference image to establish a reference coordinate system, and taking the image in the later historical period as an image to be registered to establish an image coordinate system to be registered.
(2) A connection point is selected. Because two images of the same geographic position and different time sequences to be registered reflect the same surface feature of the same region to a certain extent, some pixel points of the two images usually have a certain similarity relation, unchanging features (such as angular points in the images) in the images can be extracted first, and then a corresponding relation is established between the extracted features by using a matching algorithm. Since the precision of image registration is generally seriously affected by the accuracy, number and distribution of the selection of the connection points, in the registration process, firstly, a Forstner corner operator is used to obtain the feature points, then, a first-order polynomial is used to fit a global transformation geometric model to screen and filter the feature points, finally, a cross-correlation algorithm (see formula 2) is used to match the connection points, and the estimation of the parameters of the whole transformation model can be realized at a later stage.
Wherein E (S) i,j ) And E (T) represent the average gray-scale values of the subgraphs and templates at (i, j), respectively.
(3) And establishing a transformation model. And determining transformation model parameters required by the registration of the remote sensing images of different time sequences according to the relation between the connection points.
(4) Coordinate transformation and resampling. And (4) based on the transformation model obtained in the step (3), carrying out geometric coordinate transformation and interpolation resampling operation on the remote sensing image to be registered to obtain a final registered image.
In the embodiment, the resampling is carried out by using the cubic convolution firstly, and then the geometric coordinate transformation is carried out by using the polynomial model, so that the problem of the spatial position offset of the ground object in two remote sensing images with different time sequences can be solved well, and the ground object at the same position can be well corresponded;
the cubic convolution method reduces the amount of calculation and is separable in calculation in two directions. Describing cubic convolution interpolation of one-dimensional functions, assuming that the function in one-dimensional space is f and the interpolation function in the present embodiment is g, then for each interpolation point xk, there is g (xk) = f (xk). For data sampled at equal intervals, many interpolation functions can be expressed in the form:
for a certain interpolation point x, for which a value needs to be obtained, it must lie between two sample points xj and xj + 1. Let s = (x-xj)/h, have (x-xk)/h = (x-xj + xj-xk)/h = s + j-k. Thus, equation (3) can be written as:
g(x)＝∑ k c k u (s + j-k) equation 4
And since u is in the interval (-2, 2), 0 s are woven into the layer 1, equation (4) is simplified as:
g(x)＝c j-1 u(s+1)+c j u(s)+c j+1 u(s-1)+c j+2 u (s-2) equation 5
Another three convolution kernels (see equation 6) and boundary conditions (see equations 7, 8):
c -1 ＝f(x 2 )-3f(x 1 )+3f(x 0 ) Equation 7
c n ＝f(x n-3 )-3f(x n-2 )+3f(x n-1 ) Equation 8
And 2.3, the task of the embodiment does not detect all changed pixel points, but can accurately determine the changed area of the building. Therefore, it is necessary to effectively remove particularly prominent local features and noise to reduce errors in building change detection. The example is to process the collected remote sensing data by median filtering.
Median filtering is a non-linear spatial filtering method. The nonlinear filtering technology can effectively inhibit image noise and improve the signal-to-noise ratio of an image. The median filtering firstly sorts the gray levels of the neighborhood points, and then selects the middle value as an output gray level value. Compared with the mean filter and other linear filters, the median filter can well filter impulse Noise (impulse Noise) and Salt-and-Pepper Noise (Salt and Pepper Noise). Meanwhile, the details of the edge contour of the image can be well protected. The formula for median filtering is as follows:
g (x, y) = med { f (x-i, y-i) }, (i, j) ∈ S formula 9
Wherein g (x, y), f (x, y) are pixel gray values, and S is a template window.
In step 3, marking a change target in ArcGis software, repeatedly comparing changes between the double-time phase remote sensing images, and manually marking a changed area, wherein the steps specifically comprise:
3.1, the multi-temporal original remote sensing image is an RGB three-band image derived by software, and the original image is cut into 10 pairs of small-image images before labeling because the image size of the original image is too large, so that a labeling task is convenient to distribute, and because a part of the first-class data source is on the sea surface, the part is filtered out; in the second type of data source, the unchanged area is too large and the distribution is extremely unbalanced, so that the area with no change is partially filtered out to ensure that the proportion of positive and negative samples of the data set is seriously unbalanced.
Step 3.2, the double time phase image is guided into ArcGIS, then a new surface vector file, namely a label layer, is created, the same reference space is set, and the new surface vector file is placed on the uppermost layer;
3.3, the new image layer B is arranged on the upper layer of the old image layer A, the A-type images are checked back and forth by using a roller shutter tool, whether building changes exist or not is judged, the changed areas are marked, the marked areas are accurate, and only new parts are marked if buildings are overlapped; editing the label layer, creating a surface element, and delineating a change area;
step 3.4, adding a value field into the attribute table of the label layer, assigning the value of the element in the label layer to be 255, and preparing for deriving a binary image later;
and 3.5, converting the label vector layer into a grid and exporting. In order to ensure that the range size and the pixel size of the label grid graph and the image layer B image are consistent, a background image layer with all pixel values of 0 obtained by reclassification of the image B is preset in an experiment, and the image layer is used as an element to convert the output pixel size in the grid, and the processing range and the input of a capture grid.
At the moment, the data set comprises the remote sensing images of the new and old double time phases and the change label raster image.
Step 4, in this example, the size of the used high-resolution remote sensing image data cannot be directly input into the network model, and the images need to be spliced at a later stage, so that the sampling operation is performed in the sliding window mode in this example, the size of the image block is set to 256 × 256 pixels, and the sliding interval is also 256 pixels, that is, sampling without overlapping and without interval.
And sampling and cutting the remote sensing images of the new and old double time phases and the grid graph of the changed labels into sample files with set sizes to form a sample library. Compared with other streaming data sets, the self-built sample library (named as ZHset for the moment) has certain advantages in terms of sample quantity and labeling instances, as shown in Table 1.
TABLE 1
Table 1 above describes the results of comparing the self-constructed dataset with other popular datasets.
Step 5, the partition ratio used herein is 7.
Step 6, enhancing the data of the sample library;
it should be noted that, the data volume of the training set directly obtained by manual labeling is limited, so that the data set used for training the example network model is augmented, and the generalization capability of the deep neural network is improved by enriching sample data. Sampling is one way of data augmentation, and geometric transformation of sampled image data can also increase the number of data sets.
The API-allocations package specially written for data enhancement is used herein, and in consideration of the size of the data set of the pearl sea and the application scenario of the present example, the sample library data enhancement method in the present example mainly includes: horizontal flipping, vertical flipping, transposition, random rotation by 90 ° and affine transformation.
And 7, normalizing the sample data to normalize the pixel value of the remote sensing image between 0 and 1, so that the training sample data has similar distribution, the difference between the training sample data and the training sample data is reduced, and the convergence rate of the network model is correspondingly improved. This example chooses to use the z-score normalization method, i.e., to normalize the data by calculating the mean and variance of the sample data. The mean value of the processed input image data is 0, and the variance is 1, namely, the input image data conforms to the standard normal distribution. Can be formulated as:
x = (x- μ)/σ equation 10
Where μ and σ are the mean and variance of all sample data, respectively. For this example, the data range of all three channels of the image is reduced from interval [0, 255] to interval [0,1] by performing a normalization pre-process on the training and verification sample data.
Step 8, training a model; and training the samples in the sample library by using the deep learning network model to obtain a change detection model. In the embodiment, a twin neural network with full convolution is selected, and samples in a sample library are input into the network for iterative training to obtain a change detection model.
Step 9, evaluating the precision; in this example, the trained model is used to predict the test area and perform accuracy assessment, and four assessment indexes are used: precision (Precision), recall (Recall), F1 score (F1-score) and IOU function, if the Precision requirement is met, outputting the result, if the Precision requirement is not met, adding a sample or modifying the sample to continue iteration until the Precision requirement is met.
Where TP is the number of true positives, indicating the number of correctly detected changed pixels, TP is the number of false positives, indicating the number of false positives for false negatives for the number of changed pixels, and FN is the number of false negatives for the number of changed pixels for the false negatives.
In summary, the invention constructs a sample library which has certain advantages in terms of sample quantity and labeled examples, and the training model obtained through the sample library has a better prediction result on the test area.
In alternative embodiments, the functions/acts noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved. Furthermore, the embodiments presented and described in the flow charts of the present invention are provided by way of example in order to provide a more thorough understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of larger operations are performed independently.
Furthermore, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or software module, or one or more functions and/or features may be implemented in a separate physical device or software module. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be understood within the ordinary skill of an engineer, given the nature, function, and internal relationship of the modules. Accordingly, those skilled in the art can, using ordinary skill, practice the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative of and not intended to limit the scope of the invention, which is to be determined from the appended claims along with their full scope of equivalents.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. A processing method for a remote sensing image change detection type sample library is characterized by comprising the following steps:
acquiring a multi-source remote sensing image in an experimental area;
according to the multi-source remote sensing image, image preprocessing is carried out on the image to be processed, and a preprocessed target image is obtained;
marking a change area in the target image according to the change condition between the acquired historical images in different time phases and the target image;
sampling the target image, adjusting the size and sliding interval of image blocks in the remote sensing image, and constructing to obtain a self-constructed sample library;
training sample data in the self-built sample library by adopting a deep learning network model to obtain a change detection model; the change detection model is used for carrying out change detection on the remote sensing image; and marking a changed area on the remote sensing image in the sample data.
2. The method for processing the remote sensing image change detection type sample library according to claim 1, wherein the image preprocessing is performed on the image to be processed according to the multi-source remote sensing image to obtain a preprocessed target image, and the method comprises the following steps:
performing histogram matching processing on two remote sensing images in the same region at different time phases;
carrying out image registration processing on the two remote sensing images of the front time sequence and the rear time sequence after the histogram matching processing;
and denoising the remote sensing image subjected to the image registration processing.
3. The processing method of the remote sensing image change detection type sample library according to claim 2, wherein the image registration processing is performed on the two remote sensing images of the time sequence before and after the histogram matching processing, and comprises the following steps:
using the previous historical period image in the two remote sensing images as a reference image to establish a reference coordinate system, and using one image in the later historical period as an image to be registered to establish an image coordinate system to be registered;
obtaining characteristic points in the remote sensing image through a Forstner corner operator, screening and filtering the characteristic points through a first-order polynomial fitting global transformation geometric model, matching the characteristic points by using a cross-correlation algorithm, and determining a connection point;
determining transformation model parameters required by registration of remote sensing images of different time sequences according to the relation between the connection points, and constructing a transformation model;
and according to the transformation model, carrying out geometric coordinate transformation and interpolation resampling operation on the remote sensing image to be registered to obtain a final registered image.
4. The processing method of the remote sensing image change detection type sample library according to claim 1, wherein the step of marking a change area in the target image according to the change situation between the acquired historical images of different time phases and the target image comprises the steps of:
the multi-temporal remote sensing image is processed in a blocking mode, and areas which are not changed in the remote sensing image are filtered;
selecting a target double-time-phase image, importing the target double-time-phase image into labeling software, configuring the same reference space for the target double-time-phase image in the labeling software, and then building a label vector layer;
comparing the new time phase image layer with the old time phase image layer, determining and marking a change area in the image, and constructing a label image layer of the change area;
perfecting and supplementing an attribute table of a label layer in a change area;
and converting the label layer after the completion of the supplement into a grid and outputting the grid to obtain a new double-time-phase remote sensing image, an old double-time-phase remote sensing image and a change label grid map.
5. The processing method of the remote sensing image change detection sample library according to claim 1, wherein before the step of training the sample data in the self-built sample library by using the deep learning network model to obtain the change detection model, the processing method further comprises the following steps:
sample data in a self-built sample library is enhanced by carrying out sampling operation or geometric transformation operation on the sample data in the self-built sample library;
and carrying out normalization processing on the sample data in the self-built sample library.
6. The method for processing the remote sensing image change detection type sample library according to claim 1, further comprising the steps of performing model quality evaluation on the change detection model, and retraining the change detection model according to the result of the model quality evaluation, wherein the steps specifically include:
calculating a precision evaluation index, a recall rate evaluation index, an F1 score evaluation index and an evaluation index of an IOU function of the change detection model;
according to the precision evaluation index, the recall rate evaluation index, the F1 score evaluation index and the evaluation index of the IOU function, when the numerical value of each evaluation index does not meet the quality requirement of the model, sample data in the self-built sample library is adjusted, and the change detection model is retrained until each evaluation index of the change detection model meets the quality requirement of the model.
7. A processing device for a remote sensing image change detection type sample library is characterized by comprising:
the system comprises a first module, a second module and a third module, wherein the first module is used for acquiring a multi-source remote sensing image in an experimental area;
the second module is used for carrying out image preprocessing on the image to be processed according to the multi-source remote sensing image to obtain a preprocessed target image;
a third module, configured to mark a change area in the target image according to a change condition between the acquired historical images at different time phases and the target image;
the fourth module is used for carrying out sampling operation on the target image, adjusting the size and the sliding interval of an image block in the remote sensing image and constructing to obtain a self-constructed sample library;
the fifth module is used for training the sample data in the self-built sample library by adopting a deep learning network model to obtain a change detection model; the change detection model is used for carrying out change detection on the remote sensing image; and marking the remote sensing image in the sample data with a changed area.
8. An electronic device comprising a processor and a memory;
the memory is used for storing programs;
the processor executing the program realizes the method of any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that the storage medium stores a program, which is executed by a processor to implement the method according to any one of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that the computer program realizes the method according to any of claims 1 to 6 when executed by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211541397.6A CN115830297A (en) | 2022-12-02 | 2022-12-02 | Processing method of remote sensing image change detection sample library |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211541397.6A CN115830297A (en) | 2022-12-02 | 2022-12-02 | Processing method of remote sensing image change detection sample library |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115830297A true CN115830297A (en) | 2023-03-21 |
Family
ID=85543832
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211541397.6A Pending CN115830297A (en) | 2022-12-02 | 2022-12-02 | Processing method of remote sensing image change detection sample library |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115830297A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117094430A (en) * | 2023-07-19 | 2023-11-21 | 青海师范大学 | Crop distribution prediction method, system, equipment and medium |
| CN117392486A (en) * | 2023-12-12 | 2024-01-12 | 湖北珞珈实验室 | Method, device, equipment and storage medium for constructing natural resource element sample library |
| CN117575979A (en) * | 2023-08-01 | 2024-02-20 | 广东省国土资源测绘院 | Remote sensing image change detection method and device |
-
2022
- 2022-12-02 CN CN202211541397.6A patent/CN115830297A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117094430A (en) * | 2023-07-19 | 2023-11-21 | 青海师范大学 | Crop distribution prediction method, system, equipment and medium |
| CN117094430B (en) * | 2023-07-19 | 2024-04-26 | 青海师范大学 | Crop distribution prediction method, system, equipment and medium |
| CN117575979A (en) * | 2023-08-01 | 2024-02-20 | 广东省国土资源测绘院 | Remote sensing image change detection method and device |
| CN117392486A (en) * | 2023-12-12 | 2024-01-12 | 湖北珞珈实验室 | Method, device, equipment and storage medium for constructing natural resource element sample library |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Automatic building information model reconstruction in high-density urban areas: Augmenting multi-source data with architectural knowledge | |
| CN109800736B (en) | Road extraction method based on remote sensing image and deep learning | |
| CN115830297A (en) | Processing method of remote sensing image change detection sample library | |
| Weinmann et al. | Distinctive 2D and 3D features for automated large-scale scene analysis in urban areas | |
| CN103400151B (en) | The optical remote sensing image of integration and GIS autoregistration and Clean water withdraw method | |
| Malmgren-Hansen et al. | Convolutional neural networks for SAR image segmentation | |
| CN106845341A (en) | A kind of unlicensed vehicle identification method based on virtual number plate | |
| CN111091101B (en) | High-precision pedestrian detection method, system and device based on one-step method | |
| CN113569724B (en) | Road extraction method and system based on attention mechanism and dilation convolution | |
| CN112633140A (en) | Multi-spectral remote sensing image urban village multi-category building semantic segmentation method and system | |
| Lauko et al. | Local color and morphological image feature based vegetation identification and its application to human environment street view vegetation mapping, or how green is our county? | |
| Shu | Deep convolutional neural networks for object extraction from high spatial resolution remotely sensed imagery | |
| CN114913434A (en) | High-resolution remote sensing image change detection method based on global relationship reasoning | |
| CN111242134A (en) | Remote sensing image ground object segmentation method based on feature adaptive learning | |
| CN113378642B (en) | Method for detecting illegal occupation buildings in rural areas | |
| CN102609721B (en) | Remote sensing image clustering method | |
| CN115019174B (en) | Up-sampling remote sensing image target recognition method based on pixel recombination and attention | |
| CN114419465B (en) | Method, device and equipment for detecting change of remote sensing image and storage medium | |
| CN110310263A (en) | A kind of SAR image residential block detection method based on significance analysis and background priori | |
| CN115830322A (en) | Building semantic segmentation label expansion method based on weak supervision network | |
| CN105809187A (en) | Multi-manufacturer partial discharge data result diagnosis analysis method based on image identification | |
| CN116129280B (en) | Method for detecting snow in remote sensing image | |
| CN108256420B (en) | A method of the specific atural object in harbour is monitored using multispectral analysis | |
| CN115456957B (en) | Method for detecting change of remote sensing image by full-scale feature aggregation | |
| Dogru et al. | Fast-ICA Based Lane Detection Method for Autonomous Vehicles |
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 |