CN116051519B - Method, device, equipment and storage medium for detecting double-time-phase image building change - Google Patents
Method, device, equipment and storage medium for detecting double-time-phase image building change Download PDFInfo
- Publication number
- CN116051519B CN116051519B CN202310077457.1A CN202310077457A CN116051519B CN 116051519 B CN116051519 B CN 116051519B CN 202310077457 A CN202310077457 A CN 202310077457A CN 116051519 B CN116051519 B CN 116051519B
- Authority
- CN
- China
- Prior art keywords
- clustering
- feature
- vector
- phase image
- characteristic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Abstract
The invention provides a method, a device, equipment and a storage medium for detecting the change of a double-time-phase image building, wherein the method comprises the following steps: inputting the double-phase images into a feature extraction model and acquiring splicing features; performing differential calculation on the spliced characteristic and dividing the spliced characteristic into a plurality of non-overlapping characteristic blocks; calculating the difference characteristic of each non-overlapping characteristic block and the average characteristic; performing pca calculation on the differential characteristic data and obtaining characteristic vectors and characteristic values of a covariance matrix; the feature vectors of the covariance matrix are arranged in a descending order according to the size of the feature values, and non-overlapping feature blocks are projected onto the feature vectors to obtain target vectors; and clustering the feature vector space of the target vector, and finally detecting the change of the double-time-phase image building according to the clustering result. According to the invention, the supervised method is converted into the unsupervised method, and the feature extraction pre-training model is adopted, so that a large number of annotations on the images are not needed manually, and the efficiency of building change detection is greatly improved.
Description
Technical Field
The present invention relates to the field of image detection technologies, and in particular, to a method, an apparatus, a device, and a storage medium for detecting a change of a building with dual-phase images.
Background
Along with the rapid development of economy in China in recent years, the application range of remote sensing images is also expanding, and the remote sensing images comprise multiple aspects of building change detection, urban data update, disaster prevention emergency, urban planning, population estimation, topographic map making and updating and the like. With the rapid development of deep learning technology, building change detection based on double-time-phase remote sensing images by using deep learning becomes a development trend and research hotspot in remote sensing scientific work in recent years.
Most of the current deep learning algorithms are mainly based on a supervised mode to train a deep learning network model to detect building changes based on double-time-phase remote sensing images. The building change detection based on the supervised mode needs to manually label a large number of labels, consumes more manpower and financial resources, has high time cost, and is closely related to the size of the labeled data set. Thus, there is a need for a solution that can quickly and accurately extract building changes.
Disclosure of Invention
The invention aims to provide a method, a device, equipment and a storage medium for detecting building changes by double-time-phase images, so as to solve the technical problems, and quickly and accurately extract the building changes.
In order to solve the technical problems, the invention provides a method for detecting the change of a building by double-time-phase image, which comprises the following steps:
inputting a first time phase image and a second time phase image into a pre-trained image feature extraction module, and respectively performing channel stitching on output features corresponding to the first time phase image and the second time phase image to obtain a first stitching feature and a second stitching feature;
performing differential calculation on the first splicing characteristic and the second splicing characteristic to obtain an image differential characteristic;
performing up-sampling operation on the image difference characteristics to obtain characteristic blocks to be segmented, wherein the size of the characteristic blocks to be segmented is the same as that of the first time phase image, and dividing the characteristic blocks to be segmented into a plurality of non-overlapping characteristic blocks;
calculating average characteristics of all non-overlapping characteristic blocks, and calculating differential characteristics of each non-overlapping characteristic block and the average characteristics to obtain differential characteristic data;
calculating the differential characteristic data by using a pca technology to obtain a covariance matrix, and obtaining a characteristic vector and a characteristic value of the covariance matrix;
the eigenvectors of the covariance matrix are arranged in a descending order according to the magnitude of eigenvalues, and vectors obtained by projecting non-overlapping eigenvectors to the eigenvectors of the covariance matrix according to pixels from left to right and from top to bottom are arranged in sequence to obtain target vectors;
clustering and marking the feature vector space of the target vector, and correcting a clustering label generated in the clustering process;
and determining a building change area of the double-phase image based on the corrected clustering result.
Further, the inputting the first time phase image and the second time phase image to the pre-trained image feature extraction module, and performing channel stitching on output features corresponding to the first time phase image and the second time phase image respectively to obtain a first stitching feature and a second stitching feature, including:
inputting a first time phase image into a pre-trained image feature extraction module, and obtaining two-stage features, three-stage features and four-stage features corresponding to the first time phase image output by the image feature extraction module;
performing up-sampling operation on two-stage features, three-stage features and four-stage features corresponding to the first time phase image by adopting a bilinear interpolation method, and splicing the stage features based on a preset axis to obtain a first splicing feature corresponding to the first time phase image;
inputting a second time phase image into a pre-trained image feature extraction module, and obtaining two-stage features, three-stage features and four-stage features which are output by the image feature extraction module and correspond to the second time phase image;
and respectively carrying out up-sampling operation on the two-stage features, the three-stage features and the four-stage features corresponding to the second time-phase image by adopting a bilinear interpolation method, and splicing the stage features based on a preset axis to obtain a second splicing feature corresponding to the second time-phase image.
Further, the image difference feature is an absolute value of a difference of the first stitching feature and the second stitching feature subtracted.
Further, the clustering marking is performed on the feature vector space of the target vector, and the clustering label generated in the clustering process is corrected, which comprises the following steps:
and clustering and marking the feature vector space of the target vector by adopting a k-medoids clustering algorithm, and correcting the clustering label generated in the clustering process.
Further, the clustering marking is performed on the feature vector space of the target vector by adopting a k-means clustering algorithm, and the clustering label generated in the clustering process is corrected, which comprises the following steps:
clustering the feature vector space of the target vector by adopting a k-means clustering algorithm to obtain a first clustering vector representing a building change area and a second clustering vector representing a building non-change area;
calculating a target average vector of the target vectors;
when the data corresponding to the non-overlapping feature blocks are marked as a first clustering vector and the average value of the data is smaller than the target average vector, marking the data corresponding to the non-overlapping feature blocks as the first clustering vector, otherwise marking the data corresponding to the non-overlapping feature blocks as a second clustering vector;
and when the data corresponding to the non-overlapping feature block is marked as a second cluster vector and the average value of the data is larger than the target average vector, marking the data corresponding to the non-overlapping feature block as the second cluster vector, otherwise marking the data corresponding to the non-overlapping feature block as the first cluster vector.
Further, the determining the building change area of the dual-phase image based on the corrected clustering result includes:
after marking correction, calculating a first clustering average vector corresponding to the first clustering vector and a second clustering average vector corresponding to the second clustering vector;
generating a building change binary image of a double-phase image based on the first clustering average vector and the second clustering average vector;
determining a building change area of the double-time-phase image based on the building change binary image of the double-time-phase image; the pixel with the value of 1 in the two-time-phase image building change binary image represents that a building is changed, and the pixel with the value of 0 in the two-time-phase image building change binary image represents that the building is not changed.
Further, the pre-trained image feature extraction module is a swin transformer network model pre-trained by using ImageNet.
The invention also provides a device for detecting the change of the building by double-time-phase images, which comprises:
the feature extraction module is used for inputting the first time phase image and the second time phase image into the pre-trained image feature extraction module, and respectively carrying out channel stitching on output features corresponding to the first time phase image and the second time phase image to obtain a first stitching feature and a second stitching feature;
the difference calculation module is used for carrying out difference calculation on the first splicing characteristic and the second splicing characteristic to obtain an image difference characteristic;
the feature segmentation module is used for carrying out up-sampling operation on the image difference features to obtain feature blocks to be segmented, the size of the feature blocks to be segmented is the same as that of the first time phase image, and the feature blocks to be segmented are segmented into a plurality of non-overlapping feature blocks;
the characteristic difference module is used for calculating the average characteristics of all the non-overlapping characteristic blocks and calculating the difference characteristics of each non-overlapping characteristic block and the average characteristics to obtain difference characteristic data;
the characteristic acquisition module is used for calculating the differential characteristic data by using a pca technology to obtain a covariance matrix and acquiring characteristic vectors and characteristic values of the covariance matrix;
the vector acquisition module is used for arranging the eigenvectors of the covariance matrix in a descending order according to the magnitude of the eigenvalues, and arranging vectors obtained by projecting non-overlapping eigenvectors to the eigenvectors of the covariance matrix according to pixels from left to right and from top to bottom in sequence to obtain target vectors;
the clustering marking module is used for carrying out clustering marking on the feature vector space of the target vector and correcting the clustering label generated in the clustering process;
and the change detection module is used for determining a building change area of the double-phase image based on the corrected clustering result.
The invention also provides a terminal device, which comprises a processor and a memory storing a computer program, wherein the processor realizes the dual-time-phase image building change detection method when executing the computer program.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the dual phase image building change detection method of any one of the above.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a method, a device, equipment and a storage medium for detecting the change of a double-time-phase image building, wherein the method comprises the following steps: inputting the double-phase images into a feature extraction model and acquiring splicing features; performing differential calculation on the spliced characteristic and dividing the spliced characteristic into a plurality of non-overlapping characteristic blocks; calculating the difference characteristic of each non-overlapping characteristic block and the average characteristic; performing pca calculation on the differential characteristic data and obtaining characteristic vectors and characteristic values of a covariance matrix; the feature vectors of the covariance matrix are arranged in a descending order according to the size of the feature values, and non-overlapping feature blocks are projected onto the feature vectors to obtain target vectors; and clustering the feature vector space of the target vector, and finally detecting the change of the double-time-phase image building according to the clustering result. According to the invention, the supervised method is converted into the unsupervised method, and the feature extraction pre-training model is adopted, so that a large number of annotations on the images are not needed manually, and the efficiency of building change detection is greatly improved.
Drawings
FIG. 1 is a flow chart of a method for detecting changes of a dual-phase image building;
FIG. 2 is a schematic diagram of a swin transformer according to the present invention;
fig. 3 is a schematic structural diagram of the dual-phase image building change detection device provided by the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, an embodiment of the present invention provides a method for detecting a change of a building by using a dual-phase image, which may include the steps of:
s1, inputting a first time phase image and a second time phase image into a pre-trained image feature extraction module, and respectively performing channel stitching on output features corresponding to the first time phase image and the second time phase image to obtain a first stitching feature and a second stitching feature;
s2, carrying out differential calculation on the first splicing characteristic and the second splicing characteristic to obtain an image differential characteristic;
s3, performing up-sampling operation on the image difference characteristics to obtain characteristic blocks to be segmented, wherein the size of the characteristic blocks to be segmented is the same as that of the first time phase image, and dividing the characteristic blocks to be segmented into a plurality of non-overlapping characteristic blocks;
s4, calculating average characteristics of all the non-overlapping characteristic blocks, and calculating differential characteristics of each non-overlapping characteristic block and the average characteristics to obtain differential characteristic data;
s5, calculating the differential characteristic data by using a pca technology to obtain a covariance matrix, and obtaining a characteristic vector and a characteristic value of the covariance matrix;
s6, arranging eigenvectors of the covariance matrix in a descending order according to the magnitude of eigenvalues, and arranging vectors obtained by projecting non-overlapping eigenvectors to eigenvectors of the covariance matrix according to pixels from left to right and from top to bottom in sequence to obtain target vectors;
s7, carrying out clustering marking on the feature vector space of the target vector, and correcting a clustering label generated in the clustering process;
and S8, determining a building change area of the double-phase image based on the corrected clustering result.
In the embodiment of the present invention, further, the inputting the first time phase image and the second time phase image to the pre-trained image feature extraction module, and performing channel stitching on output features corresponding to the first time phase image and the second time phase image respectively to obtain a first stitching feature and a second stitching feature, includes:
inputting a first time phase image into a pre-trained image feature extraction module, and obtaining two-stage features, three-stage features and four-stage features corresponding to the first time phase image output by the image feature extraction module;
performing up-sampling operation on two-stage features, three-stage features and four-stage features corresponding to the first time phase image by adopting a bilinear interpolation method, and splicing the stage features based on a preset axis to obtain a first splicing feature corresponding to the first time phase image;
inputting a second time phase image into a pre-trained image feature extraction module, and obtaining two-stage features, three-stage features and four-stage features which are output by the image feature extraction module and correspond to the second time phase image;
and respectively carrying out up-sampling operation on the two-stage features, the three-stage features and the four-stage features corresponding to the second time-phase image by adopting a bilinear interpolation method, and splicing the stage features based on a preset axis to obtain a second splicing feature corresponding to the second time-phase image.
In an embodiment of the present invention, further, the image difference feature is an absolute value of a difference subtracted from the first stitching feature and the second stitching feature.
In the embodiment of the present invention, further, the performing cluster marking on the feature vector space of the target vector and correcting the cluster label generated in the clustering process includes:
and clustering and marking the feature vector space of the target vector by adopting a k-medoids clustering algorithm, and correcting the clustering label generated in the clustering process.
In the embodiment of the present invention, further, the clustering marking is performed on the feature vector space of the target vector by adopting a k-means clustering algorithm, and the clustering label generated in the clustering process is corrected, including:
clustering the feature vector space of the target vector by adopting a k-means clustering algorithm to obtain a first clustering vector representing a building change area and a second clustering vector representing a building non-change area;
calculating a target average vector of the target vectors;
when the data corresponding to the non-overlapping feature blocks are marked as a first clustering vector and the average value of the data is smaller than the target average vector, marking the data corresponding to the non-overlapping feature blocks as the first clustering vector, otherwise marking the data corresponding to the non-overlapping feature blocks as a second clustering vector;
and when the data corresponding to the non-overlapping feature block is marked as a second cluster vector and the average value of the data is larger than the target average vector, marking the data corresponding to the non-overlapping feature block as the second cluster vector, otherwise marking the data corresponding to the non-overlapping feature block as the first cluster vector.
In an embodiment of the present invention, further, the determining a building change area of the dual-phase image based on the corrected clustering result includes:
after marking correction, calculating a first clustering average vector corresponding to the first clustering vector and a second clustering average vector corresponding to the second clustering vector;
generating a building change binary image of a double-phase image based on the first clustering average vector and the second clustering average vector;
determining a building change area of the double-time-phase image based on the building change binary image of the double-time-phase image; the pixel with the value of 1 in the two-time-phase image building change binary image represents that a building is changed, and the pixel with the value of 0 in the two-time-phase image building change binary image represents that the building is not changed.
In the embodiment of the present invention, further, the pre-trained image feature extraction module is a swin transformer network model pre-trained by using ImageNet.
Based on the above scheme, in order to better understand the method for detecting the change of the double-time-phase image building provided by the embodiment of the invention, the following details are described:
according to the embodiment of the invention, the image feature extraction module backbox of the swin transformer pre-trained by the ImageNet data set is adopted to extract the building features, and then a series of feature vector processing is carried out based on the extracted features, so that the extraction of the building change can be rapidly realized, and the workflow of the building change detection is greatly simplified.
The embodiment of the invention can be realized by the following steps:
1. referring to fig. 2, a T1 phase image is input into a swin transformer network model pre-trained by using ImageNet to respectively obtain a feature t1_stage2 output in a stage2, a feature t1_stage3 output in a stage3, and a feature t1_stage4 output in a stage4;
next, the T2 phase image is input into a swin transformer network model to obtain characteristics t2_stage2, t2_stage3, and t2_stage4 output in stages 2, stage3, and stage4, respectively.
It is understood that the T1 phase image and the T2 phase image have the same size; (the T1 time phase image and the T2 time phase image are images obtained by shooting the same area at different time points);
t1_stag2 and t2_stag2 have the same size and channel number, the size is 8 times of downsampling of the T1 phase image (T2 phase image);
t1_stag3 and t2_stag3 have the same size and channel number, and the size is 16 times of downsampling of the T1 phase image (T2 phase image);
t1_stage4 and t2_stage4 have the same size and channel number, and the size is 32 times of downsampling of the T1 phase image (T2 phase image).
1.1, up-sampling the T1_stag4 by 2 times by adopting a bilinear interpolation method, and then performing channel splicing with the T1_stag3 by using the axs=1 to obtain a T1_stag34;
the t1_stage34 is then upsampled 2 times by bilinear interpolation and then spliced with t1_stage2 at axe=2 to yield t1_stage23.
The bilinear interpolation method refers to applying a mathematical two-dimensional linear interpolation algorithm to an image to achieve the enlargement and reduction of the image. The scaling step of the image using bilinear interpolation is as follows: (1) Calculating a scaling factor (width and height directions) between the target picture and the original picture; (2) Back-pushing the virtual pixel position in the original picture from the target picture pixel position by using the scaling factor; (3) Finding four adjacent pixel points in the width and height directions from the virtual pixel positions; (4) And carrying out bilinear interpolation calculation by the four pixel points to obtain pixel values in the target image.
It can be understood that t1_stage2, t1_stage3, t1_stage4, t2_stage2, t2_stage3, t2_stage4 corresponding to the same are obtained, and their data formats are (batch, channel, height, width) respectively represent the number of batches, the number of channels, high and wide; axes=1 indicates that stitching is performed with channel as the axis. For example: assuming that the data format of t1_stage4 after upsampling is (2,128,100,100) and that of t1_stage3 is (2,256,100,100), the data format of t1_stage3 after splicing is (2,384,100,100) with axe=1 as an axis.
1.2, similarly, the steps of T2_stage4, T2_stage3 and T2_stage2 are operated according to the step sequence of (1), and finally T2_stage23 is obtained.
2. And carrying out differential calculation on the feature T1_stage23 and the feature T2_stage23 to obtain a differential feature F. For an image, the differential calculation can be defined simply as the absolute value of the difference of two features: f= |t1_stage23-t2_stage23|, although the difference can be defined as other forms, the simplest being the subtraction followed by the absolute value.
In this method, two images taken at the same place and at different times are obtained. There is a hypothesis here that if the front image has no building, the rear image has a building. After the two extracted features are subjected to differential operation, numerical differences exist between the place where the building changes and the place where the building does not change, and the data after differential calculation of the place where the building changes is larger in probability.
3. And (3) carrying out 2 times of up-sampling operation on the differential feature F, and carrying out 3 times of continuous up-sampling operation, wherein the step-by-step up-sampling is carried out to obtain the feature with the same size as the T1 time phase image, and the length of the feature is assumed to be H, and the width is assumed to be W.
4. Dividing X features into non-overlapping feature blocks of hxh, denoted as X i The length is H, the width is H, where i represents the index number of the non-overlapping feature block, the size range is 1.ltoreq.i.ltoreq.N= (H×W)/(h×h), and N represents the number of division of feature X into non-overlapping feature blocks of size hxh.
5. Calculating the average feature of all non-overlapping feature blocks, denoted as ψ N N represents the number of non-overlapping feature blocks of size hxh that divide feature X, the formula being:
6. calculating the difference characteristic of each non-overlapping characteristic block and the average characteristic, and recording as delta i I represents the index of the non-overlapping feature blocks:
Δ i ＝x i -Ψ N
7. for differential data delta i Performing pca calculation to obtain a data set, and supposing that the data set contains h 2 Vectors E orthogonal to each other s Corresponding to the associated best suited description delta i Scalar lambda for data distribution s . Accordingly, vector E s And a scalar lambda s Respectively covariance matrix C M Is described.
Wherein delta is i T Is a differential matrix delta i Transposed matrix of C M Is of size h 2 ×h 2 Has h 2 The feature vectors and the corresponding feature values.
It should be noted that the pca is actually high-dimensional dataAnd (5) reducing the dimension. The optimization objective of the data dimension reduction is assumed to be: assuming that a set of N-dimensional vectors is reduced to K dimensions, the goal is to select K unit orthogonal bases such that after the original data is transformed onto the set of bases, the covariance between each variable is 0, and the variance of the variables is as large as possible (under the constraint of orthogonality, the maximum K variances are taken); actually, PCa is acted on delta i The method is to perform dimension reduction on the model to obtain mutually orthogonal eigenvectors and eigenvalues.
8. C is C M The feature vectors of (2) are sorted in descending order according to the magnitude of the feature values.
9. Feature block x will not overlap i Each pixel on to C M In the feature vector of (1), the coordinates of the assumed pixel are (j, k), which are recorded asThen a new vector is obtained:
v i (j,k)＝[v 1 v 2 …v s …v S ] T
wherein the parameter S is a vector v i The number of feature vectors (j, k) (parameter S determines the dimension of the new vector), 1.ltoreq.S.ltoreq.h 2 ，S is more than or equal to 1 and less than or equal to S, S is C M The operation T represents the matrix transposition according to the index values of the eigenvectors ordered in descending order of eigenvalue sizes.
10. Feature block x will not overlap i The vectors obtained after projection of the pixels from left to right and from top to bottom are sequentially arranged together to obtain a vector v.
11. The v eigenvector space was clustered using a k-means clustering algorithm, setting k=2. The cluster generated by the k-medoids clustering algorithm is marked as k u And k c The building change area and the building non-change area are respectively represented.
The clustering process is specifically as follows: 1. k=2 data samples were randomly selected as particles (medoids) among the data samples v. (the criterion function of selecting the medoids is that the sum of the distances from all other points in the current data set to the center point is minimum; 2. the remaining sample points are repeatedly assigned to k cluster classes, according to the principle of closest distance to the medoids. 3. In each class, calculating a criterion function corresponding to each member point, and selecting a point corresponding to the minimum criterion function as a new medoids. 4. The process of 2-3 is repeated until all the merodids points no longer change or the set maximum number of iterations has been reached.
Then calculate and get the cluster k u Clustering of vectors average vector v u ；
Computing cluster k c Clustering of vectors average vector v c ；
Calculating the average vector v of the vectors v m ；
And further correcting cluster labels generated by k-medoids clusters:
when the same region of two different phases changes, the absolute value obtained by differentiating the images of the two phases is higher than that of the region which does not change.
When x is i The data at the upper position (j, k) is marked k by k-means c And its average value is greater than v m When it is marked as k c Otherwise marked as k u ；
When x is i The data at the upper position (j, k) is marked k by k-means u And its average value is less than v m When it is marked as k u Otherwise marked as k c ；
After correction, the cluster k is recalculated u Clustering of vectors average vector v u And cluster k c Clustering of vectors average vector v c 。
12. By v u And v c And generating a building change binary image of the double-phase image, wherein the building change binary image is expressed as CBM= { CBM (j, k) |1 is less than or equal to j is less than or equal to H,1 is less than or equal to k is more than or equal to W, when CBM (j, k) is 1, the pixel of the corresponding double-phase image at the position (j, k) changes, and when CBM (j, k) is 0, the pixel of the corresponding double-phase image at the position (j, k) does not change. The specific calculation formula is as follows:
|||| 2 representing euclidean distance (Euclidean distance).
It should be noted that, compared with other building change detection or semantic segmentation methods implemented in a deep learning manner, the embodiment of the invention has the following advantages: 1. the feature extraction modules of other tasks are easy to migrate: whether the tasks of image classification, target detection, semantic segmentation and other deep learning are realized, the trained feature extraction module can be directly used as the feature extraction module of the method, one feature extraction module does not need to be retrained for the method, a large number of pre-trained models are opened on the network at present, and the method can be directly used, and the method uses the image feature extraction module of the image Net pre-trained swin transformer. 2. Because the feature extraction pre-training model is used, the method does not need to manually label labels, and the working efficiency of building change detection is greatly improved; 3. in a certain sense, the method skillfully converts the supervised method into the unsupervised method, and avoids the resource and time waste caused by repeatedly manufacturing wheels.
It should be noted that, for simplicity of description, the above method or flow embodiments are all described as a series of combinations of acts, but it should be understood by those skilled in the art that the embodiments of the present invention are not limited by the order of acts described, as some steps may occur in other orders or concurrently in accordance with the embodiments of the present invention. Further, those skilled in the art will appreciate that the embodiments described in the specification are all alternative embodiments and that the actions involved are not necessarily required for the embodiments of the present invention.
Referring to fig. 3, an embodiment of the present invention further provides a dual-phase image building change detection device, including:
the feature extraction module 1 is used for inputting a first time phase image and a second time phase image into the pre-trained image feature extraction module, and respectively performing channel stitching on output features corresponding to the first time phase image and the second time phase image to obtain a first stitching feature and a second stitching feature;
the difference calculation module 2 is used for carrying out difference calculation on the first splicing characteristic and the second splicing characteristic to obtain an image difference characteristic;
the feature segmentation module 3 is used for performing up-sampling operation on the image difference features to obtain feature blocks to be segmented, which have the same size as the first time phase image, and segmenting the feature blocks to be segmented into a plurality of non-overlapping feature blocks;
the feature difference module 4 is used for calculating average features of all the non-overlapping feature blocks, and calculating difference features of each non-overlapping feature block and the average features to obtain difference feature data;
the feature acquisition module 5 is used for calculating the differential feature data by using a pca technology to obtain a covariance matrix and acquiring feature vectors and feature values of the covariance matrix;
the vector acquisition module 6 is used for arranging the eigenvectors of the covariance matrix in a descending order according to the magnitude of the eigenvalues, and arranging vectors obtained by projecting non-overlapping eigenvectors to the eigenvectors of the covariance matrix according to pixels from left to right and from top to bottom in sequence to obtain a target vector;
the clustering marking module 7 is used for carrying out clustering marking on the feature vector space of the target vector and correcting the clustering label generated in the clustering process;
and the change detection module 8 is used for determining a building change area of the double-phase image based on the corrected clustering result.
It can be understood that the embodiment of the device item corresponds to the embodiment of the method item of the present invention, and the dual-time-phase image building change detection device provided by the embodiment of the present invention may implement the dual-time-phase image building change detection method provided by any one of the embodiments of the method item of the present invention.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the dual phase image building change detection method of any one of the above.
It should be noted that the above-described apparatus embodiments are merely illustrative, and the units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the embodiment of the device provided by the invention, the connection relation between the modules represents that the modules have communication connection, and can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
It will be clear to those skilled in the art that, for convenience and brevity, the specific working process of the apparatus described above may refer to the corresponding process in the foregoing method embodiment, which is not described herein again.
The terminal device may be a computing device such as a desktop computer, a notebook computer, a palm computer, a cloud server, etc. The terminal device may include, but is not limited to, a processor, a memory.
The processor may be a central processing unit (Central Processing Unit, CPU), other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. The general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, which is a control center of the terminal device, and which connects various parts of the entire terminal device using various interfaces and lines.
The memory may be used to store the computer program, and the processor may implement various functions of the terminal device by running or executing the computer program stored in the memory and invoking data stored in the memory. The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data created according to the use of the cellular phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, smart Media Card (SMC), secure Digital (SD) Card, flash Card (Flash Card), at least one disk storage device, flash memory device, or other volatile solid-state storage device.
The storage medium is a computer readable storage medium, and the computer program is stored in the computer readable storage medium, and when executed by a processor, the computer program can implement the steps of the above-mentioned method embodiments. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth. It should be noted that the computer readable medium contains content that can be appropriately scaled according to the requirements of jurisdictions in which such content is subject to legislation and patent practice, such as in certain jurisdictions in which such content is subject to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.
While the foregoing is directed to the preferred embodiments of the present invention, it will be appreciated by those skilled in the art that changes and modifications may be made without departing from the principles of the invention, such changes and modifications are also intended to be within the scope of the invention.
Claims (6)
1. The method for detecting the change of the building by using the double-time-phase images is characterized by comprising the following steps of:
inputting a first time phase image and a second time phase image into a pre-trained image feature extraction module, and respectively performing channel stitching on output features corresponding to the first time phase image and the second time phase image to obtain a first stitching feature and a second stitching feature;
performing differential calculation on the first splicing characteristic and the second splicing characteristic to obtain an image differential characteristic;
performing up-sampling operation on the image difference characteristics to obtain characteristic blocks to be segmented, wherein the size of the characteristic blocks to be segmented is the same as that of the first time phase image, and dividing the characteristic blocks to be segmented into a plurality of non-overlapping characteristic blocks;
calculating average characteristics of all non-overlapping characteristic blocks, and calculating differential characteristics of each non-overlapping characteristic block and the average characteristics to obtain differential characteristic data;
calculating the differential characteristic data by using a pca technology to obtain a covariance matrix, and obtaining a characteristic vector and a characteristic value of the covariance matrix;
the eigenvectors of the covariance matrix are arranged in a descending order according to the magnitude of eigenvalues, and vectors obtained by projecting non-overlapping eigenvectors to the eigenvectors of the covariance matrix according to pixels from left to right and from top to bottom are arranged in sequence to obtain target vectors;
clustering and marking the feature vector space of the target vector, and correcting a clustering label generated in the clustering process;
determining a building change area of the double-phase image based on the corrected clustering result;
the method for obtaining the first splicing characteristic and the second splicing characteristic comprises the steps of:
inputting a first time phase image into a pre-trained image feature extraction module, and obtaining two-stage features, three-stage features and four-stage features corresponding to the first time phase image output by the image feature extraction module;
performing up-sampling operation on two-stage features, three-stage features and four-stage features corresponding to the first time phase image by adopting a bilinear interpolation method, and splicing the stage features based on a preset axis to obtain a first splicing feature corresponding to the first time phase image;
inputting a second time phase image into a pre-trained image feature extraction module, and obtaining two-stage features, three-stage features and four-stage features which are output by the image feature extraction module and correspond to the second time phase image;
performing up-sampling operation on two-stage features, three-stage features and four-stage features corresponding to the second time-phase image by adopting a bilinear interpolation method, and splicing the stage features based on a preset axis to obtain a second splicing feature corresponding to the second time-phase image;
the pre-trained image feature extraction module adopts a swin transducer network model which is pre-trained by using an ImageNet;
the clustering marking is carried out on the feature vector space of the target vector, and the clustering label generated in the clustering process is corrected, and the method comprises the following steps: clustering and marking the feature vector space of the target vector by adopting a k-means clustering algorithm, and correcting a clustering label generated in a clustering process, wherein the clustering label comprises the following specific steps:
clustering the feature vector space of the target vector by adopting a k-means clustering algorithm to obtain a first clustering vector representing a building change area and a second clustering vector representing a building non-change area;
calculating a target average vector of the target vectors;
when the data corresponding to the non-overlapping feature blocks are marked as a first clustering vector and the average value of the data is smaller than the target average vector, marking the data corresponding to the non-overlapping feature blocks as the first clustering vector, otherwise marking the data corresponding to the non-overlapping feature blocks as a second clustering vector;
and when the data corresponding to the non-overlapping feature block is marked as a second cluster vector and the average value of the data is larger than the target average vector, marking the data corresponding to the non-overlapping feature block as the second cluster vector, otherwise marking the data corresponding to the non-overlapping feature block as the first cluster vector.
2. The method of claim 1, wherein the image difference feature is an absolute value of a difference between the first stitching feature and the second stitching feature.
3. The method for detecting a building change in a dual-phase image according to claim 1, wherein determining a building change area of the dual-phase image based on the corrected clustering result comprises:
after marking correction, calculating a first clustering average vector corresponding to the first clustering vector and a second clustering average vector corresponding to the second clustering vector;
generating a building change binary image of a double-phase image based on the first clustering average vector and the second clustering average vector;
determining a building change area of the double-time-phase image based on the building change binary image of the double-time-phase image; the pixel with the value of 1 in the two-time-phase image building change binary image represents that a building is changed, and the pixel with the value of 0 in the two-time-phase image building change binary image represents that the building is not changed.
4. A dual-temporal image building change detection device, comprising:
the feature extraction module is used for inputting the first time phase image and the second time phase image into the pre-trained image feature extraction module, and respectively carrying out channel stitching on output features corresponding to the first time phase image and the second time phase image to obtain a first stitching feature and a second stitching feature;
the difference calculation module is used for carrying out difference calculation on the first splicing characteristic and the second splicing characteristic to obtain an image difference characteristic;
the feature segmentation module is used for carrying out up-sampling operation on the image difference features to obtain feature blocks to be segmented, the size of the feature blocks to be segmented is the same as that of the first time phase image, and the feature blocks to be segmented are segmented into a plurality of non-overlapping feature blocks;
the characteristic difference module is used for calculating the average characteristics of all the non-overlapping characteristic blocks and calculating the difference characteristics of each non-overlapping characteristic block and the average characteristics to obtain difference characteristic data;
the characteristic acquisition module is used for calculating the differential characteristic data by using a pca technology to obtain a covariance matrix and acquiring characteristic vectors and characteristic values of the covariance matrix;
the vector acquisition module is used for arranging the eigenvectors of the covariance matrix in a descending order according to the magnitude of the eigenvalues, and arranging vectors obtained by projecting non-overlapping eigenvectors to the eigenvectors of the covariance matrix according to pixels from left to right and from top to bottom in sequence to obtain target vectors;
the clustering marking module is used for carrying out clustering marking on the feature vector space of the target vector and correcting the clustering label generated in the clustering process;
the change detection module is used for determining a building change area of the double-phase image based on the corrected clustering result;
the method for obtaining the first splicing characteristic and the second splicing characteristic comprises the steps of:
inputting a first time phase image into a pre-trained image feature extraction module, and obtaining two-stage features, three-stage features and four-stage features corresponding to the first time phase image output by the image feature extraction module;
performing up-sampling operation on two-stage features, three-stage features and four-stage features corresponding to the first time phase image by adopting a bilinear interpolation method, and splicing the stage features based on a preset axis to obtain a first splicing feature corresponding to the first time phase image;
inputting a second time phase image into a pre-trained image feature extraction module, and obtaining two-stage features, three-stage features and four-stage features which are output by the image feature extraction module and correspond to the second time phase image;
performing up-sampling operation on two-stage features, three-stage features and four-stage features corresponding to the second time-phase image by adopting a bilinear interpolation method, and splicing the stage features based on a preset axis to obtain a second splicing feature corresponding to the second time-phase image;
the pre-trained image feature extraction module adopts a swin transducer network model which is pre-trained by using an ImageNet;
the clustering marking is carried out on the feature vector space of the target vector, and the clustering label generated in the clustering process is corrected, and the method comprises the following steps: clustering and marking the feature vector space of the target vector by adopting a k-means clustering algorithm, and correcting a clustering label generated in a clustering process, wherein the clustering label comprises the following specific steps:
clustering the feature vector space of the target vector by adopting a k-means clustering algorithm to obtain a first clustering vector representing a building change area and a second clustering vector representing a building non-change area;
calculating a target average vector of the target vectors;
when the data corresponding to the non-overlapping feature blocks are marked as a first clustering vector and the average value of the data is smaller than the target average vector, marking the data corresponding to the non-overlapping feature blocks as the first clustering vector, otherwise marking the data corresponding to the non-overlapping feature blocks as a second clustering vector;
and when the data corresponding to the non-overlapping feature block is marked as a second cluster vector and the average value of the data is larger than the target average vector, marking the data corresponding to the non-overlapping feature block as the second cluster vector, otherwise marking the data corresponding to the non-overlapping feature block as the first cluster vector.
5. A terminal device comprising a processor and a memory storing a computer program, wherein the processor, when executing the computer program, implements the dual phase image building change detection method of any one of claims 1 to 3.
6. A non-transitory computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the dual phase image building change detection method according to any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310077457.1A CN116051519B (en) | 2023-02-02 | 2023-02-02 | Method, device, equipment and storage medium for detecting double-time-phase image building change |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310077457.1A CN116051519B (en) | 2023-02-02 | 2023-02-02 | Method, device, equipment and storage medium for detecting double-time-phase image building change |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116051519A CN116051519A (en) | 2023-05-02 |
| CN116051519B true CN116051519B (en) | 2023-08-22 |
Family
ID=86133114
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310077457.1A Active CN116051519B (en) | 2023-02-02 | 2023-02-02 | Method, device, equipment and storage medium for detecting double-time-phase image building change |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116051519B (en) |
Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103456020A (en) * | 2013-09-08 | 2013-12-18 | 西安电子科技大学 | Remote sensing image change detection method based on treelet feature fusion |
| CN109871875A (en) * | 2019-01-21 | 2019-06-11 | 大连理工大学 | A kind of building change detecting method based on deep learning |
| CN111047551A (en) * | 2019-11-06 | 2020-04-21 | 北京科技大学 | Remote sensing image change detection method and system based on U-net improved algorithm |
| CN112819753A (en) * | 2021-01-12 | 2021-05-18 | 香港理工大学深圳研究院 | Building change detection method and device, intelligent terminal and storage medium |
| CN113971741A (en) * | 2020-07-21 | 2022-01-25 | Tcl科技集团股份有限公司 | Image labeling method, classification model training method and computer equipment |
| CN114067205A (en) * | 2021-11-11 | 2022-02-18 | 中山大学 | Light-weight arbitrary-scale double-time-phase image change detection method |
| CN114092833A (en) * | 2022-01-24 | 2022-02-25 | 长沙理工大学 | Remote sensing image classification method and device, computer equipment and storage medium |
| CN114842351A (en) * | 2022-04-11 | 2022-08-02 | 中国人民解放军战略支援部队航天工程大学 | Remote sensing image semantic change detection method based on twin transforms |
| CN114926746A (en) * | 2022-05-25 | 2022-08-19 | 西北工业大学 | SAR image change detection method based on multi-scale differential feature attention mechanism |
| CN115082293A (en) * | 2022-06-10 | 2022-09-20 | 南京理工大学 | Image registration method based on Swin transducer and CNN double-branch coupling |
| WO2022221719A2 (en) * | 2021-04-16 | 2022-10-20 | Strong Force Vcn Portfolio 2019, Llc | Systems, methods, kits, and apparatuses for digital product network systems and biology-based value chain networks |
| WO2022252799A1 (en) * | 2021-06-04 | 2022-12-08 | 成都数之联科技股份有限公司 | Model training method, woodland change detection method, system, and apparatus, and medium |
| CN115457385A (en) * | 2022-08-25 | 2022-12-09 | 浙江工业大学 | Building change detection method based on lightweight network |
| CN115457390A (en) * | 2022-09-13 | 2022-12-09 | 中国人民解放军国防科技大学 | Remote sensing image change detection method and device, computer equipment and storage medium |
| CN115471467A (en) * | 2022-08-31 | 2022-12-13 | 核工业北京地质研究院 | High-resolution optical remote sensing image building change detection method |
| CN115527121A (en) * | 2022-10-18 | 2022-12-27 | 国家林业和草原局中南调查规划院 | End-to-end forest land change detection method, system, medium, equipment and terminal |
| CN115601661A (en) * | 2022-10-31 | 2023-01-13 | 湖北工业大学（Cn） | Building change detection method for urban dynamic monitoring |
| CN115661652A (en) * | 2022-10-31 | 2023-01-31 | 沈阳建筑大学 | Object-oriented graph neural network unsupervised remote sensing image change detection method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016207875A1 (en) * | 2015-06-22 | 2016-12-29 | Photomyne Ltd. | System and method for detecting objects in an image |
| US11715060B2 (en) * | 2019-05-31 | 2023-08-01 | X Development Llc | Intelligent tracking system and methods and systems therefor |
-
2023
- 2023-02-02 CN CN202310077457.1A patent/CN116051519B/en active Active
Patent Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103456020A (en) * | 2013-09-08 | 2013-12-18 | 西安电子科技大学 | Remote sensing image change detection method based on treelet feature fusion |
| CN109871875A (en) * | 2019-01-21 | 2019-06-11 | 大连理工大学 | A kind of building change detecting method based on deep learning |
| CN111047551A (en) * | 2019-11-06 | 2020-04-21 | 北京科技大学 | Remote sensing image change detection method and system based on U-net improved algorithm |
| CN113971741A (en) * | 2020-07-21 | 2022-01-25 | Tcl科技集团股份有限公司 | Image labeling method, classification model training method and computer equipment |
| CN112819753A (en) * | 2021-01-12 | 2021-05-18 | 香港理工大学深圳研究院 | Building change detection method and device, intelligent terminal and storage medium |
| WO2022221719A2 (en) * | 2021-04-16 | 2022-10-20 | Strong Force Vcn Portfolio 2019, Llc | Systems, methods, kits, and apparatuses for digital product network systems and biology-based value chain networks |
| WO2022252799A1 (en) * | 2021-06-04 | 2022-12-08 | 成都数之联科技股份有限公司 | Model training method, woodland change detection method, system, and apparatus, and medium |
| CN114067205A (en) * | 2021-11-11 | 2022-02-18 | 中山大学 | Light-weight arbitrary-scale double-time-phase image change detection method |
| CN114092833A (en) * | 2022-01-24 | 2022-02-25 | 长沙理工大学 | Remote sensing image classification method and device, computer equipment and storage medium |
| CN114842351A (en) * | 2022-04-11 | 2022-08-02 | 中国人民解放军战略支援部队航天工程大学 | Remote sensing image semantic change detection method based on twin transforms |
| CN114926746A (en) * | 2022-05-25 | 2022-08-19 | 西北工业大学 | SAR image change detection method based on multi-scale differential feature attention mechanism |
| CN115082293A (en) * | 2022-06-10 | 2022-09-20 | 南京理工大学 | Image registration method based on Swin transducer and CNN double-branch coupling |
| CN115457385A (en) * | 2022-08-25 | 2022-12-09 | 浙江工业大学 | Building change detection method based on lightweight network |
| CN115471467A (en) * | 2022-08-31 | 2022-12-13 | 核工业北京地质研究院 | High-resolution optical remote sensing image building change detection method |
| CN115457390A (en) * | 2022-09-13 | 2022-12-09 | 中国人民解放军国防科技大学 | Remote sensing image change detection method and device, computer equipment and storage medium |
| CN115527121A (en) * | 2022-10-18 | 2022-12-27 | 国家林业和草原局中南调查规划院 | End-to-end forest land change detection method, system, medium, equipment and terminal |
| CN115601661A (en) * | 2022-10-31 | 2023-01-13 | 湖北工业大学（Cn） | Building change detection method for urban dynamic monitoring |
| CN115661652A (en) * | 2022-10-31 | 2023-01-31 | 沈阳建筑大学 | Object-oriented graph neural network unsupervised remote sensing image change detection method |
Non-Patent Citations (1)
| Title |
|---|
| 结合像元级和目标级的高分辨率遥感影像建筑物变化检测;张志强;张新长;辛秦川;杨晓羚;;测绘学报(第01期);102-112 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116051519A (en) | 2023-05-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10152655B2 (en) | Deep-learning network architecture for object detection | |
| CN107103326B (en) | Collaborative significance detection method based on super-pixel clustering | |
| CN109446889B (en) | Object tracking method and device based on twin matching network | |
| CN111768432A (en) | Moving target segmentation method and system based on twin deep neural network | |
| CN110866896B (en) | Image saliency target detection method based on k-means and level set super-pixel segmentation | |
| CN110852327A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
| CN114299343A (en) | Multi-granularity information fusion fine-granularity image classification method and system | |
| CN111178196B (en) | Cell classification method, device and equipment | |
| CN113920382A (en) | Cross-domain image classification method based on class consistency structured learning and related device | |
| WO2024060839A1 (en) | Object operation method and apparatus, computer device, and computer storage medium | |
| Hameed et al. | Content based image retrieval based on feature fusion and support vector machine | |
| CN112364916A (en) | Image classification method based on transfer learning, related equipment and storage medium | |
| CN116051519B (en) | Method, device, equipment and storage medium for detecting double-time-phase image building change | |
| CN110765917A (en) | Active learning method, device, terminal and medium suitable for face recognition model training | |
| CN110717405A (en) | Face feature point positioning method, device, medium and electronic equipment | |
| EP4220555A1 (en) | Training method and apparatus for image segmentation model, image segmentation method and apparatus, and device | |
| Dalara et al. | Entity Recognition in Indian Sculpture using CLAHE and machine learning | |
| CN115170854A (en) | End-to-end PCANetV 2-based image classification method and system | |
| CN114022458A (en) | Skeleton detection method and device, electronic equipment and computer readable storage medium | |
| CN114241411A (en) | Counting model processing method and device based on target detection and computer equipment | |
| Chen et al. | Large-scale image colorization based on divide-and-conquer support vector machines | |
| Huang et al. | A hybrid clustering approach for bag-of-words image categorization | |
| Szabó et al. | Distributed high dimensional information theoretical image registration via random projections | |
| US20230410465A1 (en) | Real time salient object detection in images and videos | |
| EP4125010A1 (en) | Adaptive learning based systems and methods for optimization of unsupervised clustering |
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 |