CN111833353B - Hyperspectral target detection method based on image segmentation - Google Patents
Hyperspectral target detection method based on image segmentation Download PDFInfo
- Publication number
- CN111833353B CN111833353B CN202010687080.8A CN202010687080A CN111833353B CN 111833353 B CN111833353 B CN 111833353B CN 202010687080 A CN202010687080 A CN 202010687080A CN 111833353 B CN111833353 B CN 111833353B
- Authority
- CN
- China
- Prior art keywords
- image
- hyperspectral
- clustering
- point set
- center point
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 26
- 238000003709 image segmentation Methods 0.000 title claims abstract description 25
- 230000011218 segmentation Effects 0.000 claims abstract description 55
- 238000012805 post-processing Methods 0.000 claims description 34
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 238000000034 method Methods 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 13
- 238000004891 communication Methods 0.000 claims description 10
- 238000003064 k means clustering Methods 0.000 claims description 9
- 230000003595 spectral effect Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 230000000877 morphologic effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000005260 corrosion Methods 0.000 claims description 3
- 230000007797 corrosion Effects 0.000 claims description 3
- 230000008961 swelling Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 10
- 238000013136 deep learning model Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 101100008049 Caenorhabditis elegans cut-5 gene Proteins 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- 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
- G06T2207/10036—Multispectral image; Hyperspectral image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Abstract
The invention relates to a hyperspectral target detection method based on image segmentation, belongs to the technical field of target detection, and solves the problem that in the prior art, certain difficulty exists in obtaining more accurate hyperspectral target information. The invention discloses a hyperspectral target detection method based on image segmentation, which comprises the following steps of: performing image segmentation on the hyperspectral image to obtain a category of pixels in the hyperspectral image and a two-dimensional hyperspectral segmentation image; based on the hyperspectral segmentation image, carrying out pixel clustering to obtain a clustering center point set corresponding to each category; based on the hyperspectral segmentation image and the clustering center point set, judging the image connectivity to obtain an optimized clustering center point set corresponding to each category; and outputting target information based on the hyperspectral segmentation image and the optimized clustering center point set. The accuracy of detecting and identifying the hyperspectral target information is improved, and the hyperspectral target information can be obtained more accurately and efficiently.
Description
Technical Field
The invention relates to the technical field of target detection, in particular to a hyperspectral target detection method based on image segmentation.
Background
The hyperspectral image technology images a target area simultaneously by tens of to hundreds of continuous and subdivided spectral wave bands, obtains target image information and spectral information thereof, and really combines the spectrum and the image for the first time.
The hyperspectral remote sensing is widely applied in the earth observation field in recent years, the ground object target classification research based on the hyperspectral remote sensing image is mature, and classified objects comprise crops, buildings and streets, minerals, rivers, wetland resources and the like.
The hyperspectral remote sensing image classification technology is a feasible scheme for hyperspectral target detection, particularly hyperspectral disguised target detection, but the technology is based on pixel-level classification, so that certain difficulty exists in obtaining accurate hyperspectral target information (such as center coordinates, target height and width).
Disclosure of Invention
In view of the foregoing analysis, embodiments of the present invention provide a hyperspectral target detection method based on image segmentation, so as to solve the problem that it is difficult to obtain relatively accurate hyperspectral target information (e.g., center coordinates, target height and width, etc.).
The embodiment of the invention provides a hyperspectral target detection method based on image segmentation, which comprises the following steps:
performing image segmentation on the hyperspectral image to obtain a category of pixels in the hyperspectral image and a two-dimensional hyperspectral segmentation image;
based on the hyperspectral segmentation image, carrying out pixel clustering to obtain a clustering center point set corresponding to each category;
based on the hyperspectral segmentation image and the clustering center point set, judging image connectivity to obtain an optimized clustering center point set corresponding to each category;
and outputting target information based on the hyperspectral segmentation image and the optimized clustering center point set.
Further, image segmentation is carried out on the hyperspectral image to obtain the category of the pixels in the hyperspectral image and a two-dimensional hyperspectral segmentation image, and the image segmentation method comprises the following steps:
performing down-sampling on the original hyperspectral image to obtain a down-sampled hyperspectral image;
and carrying out image segmentation on the down-sampling hyperspectral image to obtain the category of the pixels in the down-sampling hyperspectral image and a two-dimensional hyperspectral segmentation image.
Further, still include: denoising the hyperspectral segmentation image to obtain a post-processing segmentation image;
wherein, the performing pixel clustering based on the hyperspectral segmentation image to obtain a cluster center point set corresponding to each category comprises: based on the post-processing segmentation image, carrying out pixel clustering to obtain a clustering center point set corresponding to each category;
the method for judging the image connectivity based on the hyperspectral segmentation image and the clustering center point set to obtain the optimized clustering center point set corresponding to each category comprises the following steps: based on the post-processing segmentation image and the clustering center point set, carrying out image connectivity judgment to obtain an optimized clustering center point set corresponding to each category;
outputting target information based on the hyperspectral segmentation image and the optimized clustering center point set, wherein the target information comprises: and outputting target information based on the post-processing segmentation image and the optimized clustering center point set.
Further, the denoising processing of the hyperspectral segmented image to obtain a post-processing segmented image includes:
binarizing the hyperspectral segmentation image to obtain a binary image X;
performing a morphological opening operation on the binary image X according to
Obtaining an image
In the formula, Θ represents corrosion,
denotes swelling, S denotes a structural element;
the image
Computing images for undirected weighted graphs
Obtaining connectivity between adjacent pixels, and numbering the connectivity areas;
checking the number of pixels contained in each communication area according to the serial numbers, and deleting the communication areas with the number of pixels lower than a preset value to obtain an image
Comparing the images
And obtaining the post-processing segmentation image together with the hyperspectral segmentation image.
Further, the noise removed in the denoising process is at least one of isolated pixels predicted to be of a certain category, burrs or small bridges of a communication region predicted to be of a certain category, and a communication region predicted to be of a certain category and having a pixel number lower than a preset value.
Further, the performing pixel clustering on the image segmented based on the post-processing to obtain a cluster center point set corresponding to each category includes:
and clustering pixels with the same category in the post-processing segmentation image to obtain a cluster central point set corresponding to each category, wherein the cluster central point set comprises one or more cluster central points.
Further, the pixel clustering algorithm is a K-means clustering algorithm, and the obtaining of the cluster center point set corresponding to each of the categories includes:
inputting coordinates of all pixels of the category;
determining the number K of the optimal clustering center points of the categories by utilizing an elbow rule;
and obtaining K clustering central points and forming a clustering central point set.
Further, the post-processing segmentation image is an undirected weight graph, and each node in the undirected weight graph corresponds to each pixel point in the post-processing image; the image connectivity judgment is performed based on the post-processing segmentation image and the clustering center point set, and obtaining an optimized clustering center point set corresponding to each category comprises the following steps:
s1: selecting one of the categories and the set of clustered centroids corresponding to the category;
s2: when the cluster center point set comprises a cluster center point, the cluster center point set is an optimized cluster center point set, and step S5 is executed; or
When the cluster center point set comprises at least two cluster center points, arbitrarily selecting two cluster center points from the cluster center point set, and executing step S3;
s3: when the two clustering central points are communicated, the midpoint of the two clustering central points is used as a new clustering central point and is added into the clustering central point set, and the two clustering central points are deleted from the clustering central point set; or
When the two clustering central points are not communicated, putting the two clustering central points back to the clustering central point set;
s4: traversing all the clustering center points in the clustering center point set, and repeating S2-S3 until any two clustering center points are not communicated, so as to obtain an optimized clustering center point set;
s5: traversing the category, repeatedly executing S1-S4;
s6: outputting the optimized cluster center point sets for all the categories.
Further, the spatial resolution and the spectral resolution of the down-sampled hyperspectral image are less than or equal to the spatial resolution and the spectral resolution of the original hyperspectral image.
Further, the height and the width of the hyperspectral split image are respectively equal to the height and the width of the downsampling hyperspectral image.
Compared with the prior art, the invention can realize at least one of the following beneficial effects:
1. the accuracy of detecting and identifying the hyperspectral target information is improved, and the hyperspectral target information can be more accurately obtained;
2. the calculated amount of data is reduced by downsampling the original hyperspectral image;
3. by denoising the hyperspectral segmentation image, the interference of noise (such as false alarm target, namely wrongly classified pixels) is reduced while the data calculation amount is further reduced, and the target identification precision is improved.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a flowchart of a hyperspectral target detection method based on image segmentation in an embodiment of the invention;
FIG. 2 is a flowchart of another hyperspectral target detection method based on image segmentation in the embodiment of the invention;
FIG. 3 is a schematic diagram of down-sampling an original hyperspectral image in an embodiment of the invention;
FIG. 4 is a diagram illustrating a deep learning model for predicting pixel classes according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of denoising a hyperspectral segmentation image in the embodiment of the invention;
FIG. 6 is a schematic flow chart of a denoising algorithm in the embodiment of the present invention;
FIG. 7 is a schematic flow chart of a clustering algorithm in an embodiment of the present invention;
fig. 8 is a schematic diagram of the positions of the clustering center points before and after the graph connectivity determination in the embodiment of the present invention.
Reference numerals:
1. 2, 3-class; 4-background area; 5. 6-noise;
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
The scale of the partial graph, the position of the point coordinate in the space, and the like in the embodiment of the present invention are only for convenience of explaining the embodiment of the present invention, and do not represent the precise absolute size and the absolute position.
The embodiment of the invention discloses a hyperspectral target detection method based on image segmentation, which comprises the following steps: performing image segmentation on the hyperspectral image to obtain a category of pixels in the hyperspectral image and a two-dimensional hyperspectral segmentation image; based on the hyperspectral segmentation image, carrying out pixel clustering to obtain a clustering center point set corresponding to each category; based on the hyperspectral segmentation image and the clustering center point set, judging the image connectivity to obtain an optimized clustering center point set corresponding to each category; target information is output based on the hyperspectral segmentation image and the optimized clustering center point set, as shown in fig. 1.
The accuracy of detecting and identifying the hyperspectral target information is improved by using image segmentation, pixel clustering and graph connectivity judgment for hyperspectral target detection, and the hyperspectral target information can be obtained more accurately.
Referring to fig. 2, fig. 2 is a flowchart of another hyperspectral target detection method based on image segmentation according to an embodiment of the present invention, including the following steps:
s100: performing down-sampling on the original hyperspectral image to obtain a down-sampled hyperspectral image;
s200: performing image segmentation on the downsampled hyperspectral image to obtain a category to which pixels in the downsampled hyperspectral image belong and a two-dimensional hyperspectral segmentation image;
s300: denoising the hyperspectral segmentation image to obtain a post-processing segmentation image;
s400, based on the post-processing segmentation image, carrying out pixel clustering to obtain a clustering center point set corresponding to each category;
s500: based on the post-processing segmentation image and the clustering center point set, carrying out image connectivity judgment to obtain an optimized clustering center point set corresponding to each category;
s600: and outputting target information based on the post-processing segmentation image and the optimized clustering center point set.
The original image is subjected to down-sampling, image segmentation, segmented image denoising, pixel clustering and image connectivity judgment to be used for hyperspectral target detection, so that the accuracy of detecting and identifying hyperspectral target information is improved, the data calculation amount is reduced, the noise interference is reduced, the hyperspectral target information (such as the hyperspectral target center coordinate, height and width) can be more accurately obtained, and the method has strong adaptability to the size of a hyperspectral target.
Referring to fig. 3, fig. 3 is a schematic diagram illustrating a down-sampling of an original hyperspectral image according to an embodiment of the invention. Fig. 3 (left) is an original hyperspectral image, and a downsampled hyperspectral image is obtained through the downsampling processing in step S100 (as shown in fig. 3 (right)). The size of the input original hyperspectral image is H multiplied by W multiplied by C, wherein H, W, C respectively represents the height, width and channel number of the original hyperspectral image; obtaining a down-sampled hyperspectral image through down-sampling processing, wherein the size of the down-sampled hyperspectral image is h multiplied by w multiplied by C, and h, w and C respectively represent the height, width and channel number of the down-sampled hyperspectral image; h is more than H, and W is more than W.
In one embodiment, the down-sampling method is a direct down-sampling method with multiple times in space, namely, in the row vector pixels of the original hyperspectral image, one pixel is selected as the row vector pixel of the down-sampled hyperspectral image every (W/W-1) pixels; in the column vector pixels of the original hyperspectral image, one pixel is selected as the column vector pixel of the downsampled hyperspectral image every (H/H-1) pixels.
In an embodiment, the spatial resolution and the spectral resolution of the down-sampled hyperspectral image are less than or equal to the spatial resolution and the spectral resolution of the original hyperspectral image.
Referring to fig. 4 and 5, based on step S100, a down-sampled hyperspectral image is obtained, and then the process continues to step S200, so as to obtain a hyperspectral segmentation image (as shown in fig. 5 (left)). Specifically, the category to which each pixel of the downsampled hyperspectral image belongs is predicted, the predicted category is obtained, and finally the hyperspectral segmented image is obtained. In an embodiment, the dimensionality of the down-sampled hyperspectral image is h × w × C, and the C channel pixels corresponding to the same spatial position belong to the same category, so that the dimensionality of the hyperspectral image is h × w.
In an embodiment, the height and width of the hyperspectral ly segmented image are equal to the height and width of the downsampled hyperspectral image, respectively.
In an embodiment, a deep learning model based on a convolutional neural network is adopted to predict pixel categories, the structure of the deep learning model is shown in fig. 4, the classification mechanism of the method is that a certain number of sample pixels are randomly selected as training data, the rest pixels are used as test data, and the trained deep learning model can predict the category to which each pixel in the downsampling hyperspectral image belongs.
With continued reference to fig. 4, in this model, 5 × 5 × C image blocks need to be cut out from the downsampled hyperspectral image as input data, where C is the number of channels of the downsampled hyperspectral image, and 128 and 384 represent the number of channels of corresponding data in the depth learning model. In one embodiment, the activation employs a non-negative linear correction unit, ReLU; in the training phase, the learning rate is set to 0.001, the batch size is 180, and the stride (stride) of both the convolutional layer and the maximum pooling layer is set to 1; the 5 x 5 xc input data is first passed through a multi-scale filter module that extracts spatial information, which has three parallel branches: branch one is the 1 × 1 × 128 convolutional layer plus the 5 × 5 maximum pooling layer, branch two is the 3 × 3 × 128 convolutional layer and one 3 × 3 maximum pooling layer, and branch three is the 5 × 5 × 128 convolutional layer. The output dimension of each branch is 5 × 5 × 128, and finally, the outputs of the three branches are spliced on the channel to obtain a data block with the dimension of 5 × 5 × 384.
In the next 9 convolutional layers, the convolutional filter size is 1 × 1 × 128, and the output data block dimension is 5 × 5 × 128. The convolution filter size in the last convolution layer is 1 × 1 × M, and the output data dimension is 5 × 5 × M, where M represents the number of prediction classes. Predicting the category of the pixels of the down-sampled hyperspectral image in a classifier, wherein a classifier function adopts softmax according to a formula
In the formula yiAn output value representing a sample class i, and M is the number of sample classes. And finally, obtaining the position of the maximum probability value, namely the category of the corresponding pixel, by adopting an argmax function, and outputting 5 multiplied by 5 data. And inputting all the cut 5 multiplied by C data blocks into a depth learning model, so that the predicted category of each pixel in the downsampled hyperspectral image can be obtained.
Please refer to fig. 5, wherein fig. 5 is a schematic diagram of denoising a hyperspectral segmentation image according to an embodiment of the invention. Based on step S100 and step S200, a hyperspectral divided image (as shown in fig. 5 (left)) is obtained, and the process proceeds to step S300 to obtain a post-processing divided image (as shown in fig. 5 (right)). In one embodiment, the hyperspectral segmented image includes class 1, class 2, class 3, background region 4, noise 5, and noise 6; the gray-scale values corresponding to the category 1, the category 2, the category 3 and the background area 4 are 1, 2, 3 and 0, respectively, the noise 5 and the noise 6 are pixels that are misclassified, the noise 5 is a pixel of the background area 4 but is misclassified as the category 1, and the noise 6 is a pixel of the background area 4 but is misclassified as the category 3. In order to reduce the interference of noise (false alarm signal, i.e. misclassified signal), based on step S300, the hyperspectral segmented image is denoised to obtain a post-processed segmented image.
In an embodiment, please refer to fig. 6, fig. 6 is a schematic flow chart of a denoising algorithm in an embodiment of the invention. The denoising algorithm core is a morphological open operation and deletes a communication area lower than a threshold value, and comprises the following steps:
s301: binarizing the two-dimensional hyperspectral segmentation image to obtain a binarized image X; optionally, the threshold is 0.5.
S302: performing morphological open operation on the binary image X according to the formula
Obtaining an image
In the formula, Θ represents corrosion,
denotes swelling, S denotes a structural element; the step can reduce burrs or small bridges of part of isolated pixels and connected regions; optionally, the size of S is 3 × 3.
S303: image of a person
Computing images for undirected weighted graphs
And the connectivity between adjacent pixels in the image is obtained and numbered.
S304: checking the number of pixels contained in each communication area according to the serial number, and deleting the communication areas with the number of pixels lower than a preset value to obtain an image
Optionally, deleting means setting the number of the connected region as a background number or an adjacent connected region number; further, the deleting means that the number of the communication area lower than a preset value is set as a background number, wherein the preset value is the number of pixels; optionally, the preset value is 10.
S305: comparing images
Performing hyperspectral segmentation on the image to obtain a post-processing segmentation image; image processing method
And resetting the gray value of the pixel which is not the background in the hyperspectral image to be the gray value of the pixel at the corresponding spatial position in the hyperspectral image to finally obtain the post-processing image.
In one embodiment, the noise removed by the denoising process is at least one of isolated pixels predicted to be of a certain category, burrs or bridges of a connected region predicted to be of a certain category, and a connected region predicted to be of a certain category and having a number of pixels lower than a preset value.
By denoising the hyperspectral segmentation image, the interference of noise (such as false alarm target, namely wrongly classified pixels) is reduced while the data calculation amount is further reduced, and the hyperspectral target information can be more accurately obtained.
On the basis of completing the steps S100, S200 and S300, performing pixel clustering based on the post-processing segmentation image to obtain a clustering center point set corresponding to each category; that is, the image is divided based on the post-processing, and pixels other than the background in the image are clustered to obtain a cluster center point set corresponding to each category.
In an embodiment, pixels with the same category in the post-processing segmented image are clustered, and a cluster center point set corresponding to each category is obtained, wherein the cluster center point set comprises one or more cluster center points.
In an embodiment, the clustering algorithm of the target pixel cluster is a K-means clustering algorithm, the steps of the K-means clustering algorithm are shown in fig. 7, and the cluster center points obtained by the K-means clustering algorithm are shown in fig. 8 (left). The K-means clustering algorithm comprises the following steps:
s401: inputting coordinates of all pixels in a certain category;
s402: determining the number K of the optimal clustering center points by utilizing an elbow rule (Elbowmethod);
s403: and obtaining K clustering central points and forming a clustering central point set.
Referring to fig. 8, fig. 8 is a schematic diagram illustrating a graph connectivity determination front and back clustering center point position according to an embodiment of the present invention; fig. 8 (left) is a schematic diagram of the clustering center points obtained by the K-means clustering algorithm, and fig. 8 (right) is a schematic diagram of the clustering center points in the optimized clustering center point set in each category obtained through graph connectivity judgment. And (4) obtaining a clustering center point set based on the post-processing segmentation image and the step 400, judging the connectivity of any two clustering center points in the clustering center set, and further obtaining an optimized clustering center point set.
In an embodiment, the post-processing segmented image is an undirected weighted graph, each node in the undirected weighted graph corresponds to each pixel point in the post-processing image, and the graph connectivity determination process is as follows:
s1: selecting a category and a cluster center point set corresponding to the category;
s2: when the cluster center point set comprises a cluster center point, the cluster center point set is an optimized cluster center point set, and step S5 is executed; or
When the cluster center point set comprises at least two cluster center points, randomly selecting two cluster center points from the cluster center point set, and executing the step S3;
s3: when the two clustering central points are communicated, the midpoint of the two clustering central points is used as a new clustering central point and is added into a clustering central point set, and meanwhile, the two clustering central points are deleted from the clustering central point set; or
When the two clustering central points are not communicated, putting the two clustering central points back to the clustering central point set;
s4: traversing all the clustering center points in the clustering center point set, and repeating S2-S3 until any two clustering center points are not communicated, so as to obtain an optimized clustering center point set;
s5: traversing all categories, and repeatedly executing S1-S4;
s6: and outputting the optimized clustering center point set of all the categories.
Referring to fig. 8 again, fig. 8 (right) is a schematic diagram of the cluster center point in the optimized cluster center point set obtained by the K-means clustering algorithm and the graph connectivity judgment in fig. 8 (left), and the black dots in fig. 8 represent the positions of the corresponding cluster center points in the space.
In an embodiment, the cluster center point numbers of the category 1, the category 2, and the category 3 obtained by the K-means clustering algorithm are respectively 2, 4, and 3 (as shown in fig. 8 (left)), and the cluster center point numbers in the optimized cluster center point set obtained by the graph connectivity judgment are all 1 (as shown in fig. 8 (right)).
Based on steps S100 to S500, step S600 is performed, namely target information is output based on the post-processed image and the optimized clustering center point set.
In an embodiment, the output information comprises coordinates of cluster center points in the set of optimized cluster center points of all classes, a distance between a leftmost pixel and a rightmost pixel that belong to the same cluster center point (i.e. a horizontal distance of the leftmost pixel and the rightmost pixel in the original hyperspectral image), and a distance between an uppermost pixel and a lowermost pixel that belong to the same cluster center point (i.e. a vertical distance of the uppermost pixel and the lowermost pixel in the original hyperspectral image).
The number of the hyperspectral targets is equal to the sum of the number of the clustering central points in all the optimized clustering central point sets; the hyperspectral target central point coordinate is a clustering central point coordinate in the optimized clustering central point set; the hyperspectral target width is the horizontal distance between the leftmost pixel and the rightmost pixel which belong to the same clustering center point in the original hyperspectral image; the hyperspectral target height is the vertical distance between the uppermost pixel and the lowermost pixel which belong to the same clustering center point in the original hyperspectral image.
Therefore, the coordinates of the center point, the target width and the target height of each hyperspectral target can be obtained according to the coordinates of the cluster center points in the optimized cluster center point sets of each category, the distance between the leftmost pixel and the rightmost pixel which belong to the same cluster center point and the distance between the uppermost pixel and the lowermost pixel which belong to the same cluster center point.
Compared with the prior art, the hyperspectral target detection method based on image segmentation improves the accuracy of detecting and identifying hyperspectral target information, reduces the calculated amount of data and the interference of noise, can more accurately obtain the hyperspectral target information, has strong adaptivity to the size of the hyperspectral target, and further has great advantages in the aspects of detecting and identifying the hyperspectral target with a camouflage object.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program, which is stored in a computer readable storage medium, to instruct related hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (8)
1. A hyperspectral target detection method based on image segmentation is characterized by comprising the following steps:
performing image segmentation on the hyperspectral image to obtain a category of pixels in the hyperspectral image and a two-dimensional hyperspectral segmentation image;
denoising the hyperspectral segmentation image to obtain a post-processing segmentation image; the post-processing segmentation image is an undirected weight graph, and each node in the undirected weight graph corresponds to each pixel point in the post-processing image;
based on the post-processing segmentation image, carrying out pixel clustering to obtain a clustering center point set corresponding to each category;
based on the post-processing segmentation image and the clustering center point set, carrying out image connectivity judgment to obtain an optimized clustering center point set corresponding to each category, and comprising the following steps of:
s1: selecting one of the categories and the set of clustered centroids corresponding to the category;
s2: when the cluster center point set comprises a cluster center point, the cluster center point set is an optimized cluster center point set, and step S5 is executed; alternatively, the first and second electrodes may be,
when the cluster center point set comprises at least two cluster center points, arbitrarily selecting two cluster center points from the cluster center point set, and executing step S3;
s3: when the two clustering central points are communicated, the midpoint of the two clustering central points is used as a new clustering central point and is added into the clustering central point set, and the two clustering central points are deleted from the clustering central point set; alternatively, the first and second electrodes may be,
when the two clustering central points are not communicated, putting the two clustering central points back to the clustering central point set;
s4: traversing all the clustering center points in the clustering center point set, and repeating S2-S3 until any two clustering center points are not communicated, so as to obtain an optimized clustering center point set;
s5: traversing the category, repeatedly executing S1-S4;
s6: outputting the optimized cluster center point sets of all the categories;
and outputting target information based on the post-processing segmentation image and the optimized clustering center point set.
2. The hyperspectral target detection method according to claim 1, wherein the image segmentation is performed on the hyperspectral image to obtain a category to which a pixel in the hyperspectral image belongs and a two-dimensional hyperspectral segmented image, and the method comprises the following steps:
performing down-sampling on the original hyperspectral image to obtain a down-sampled hyperspectral image;
and carrying out image segmentation on the down-sampling hyperspectral image to obtain the category of the pixels in the down-sampling hyperspectral image and a two-dimensional hyperspectral segmentation image.
3. The hyperspectral target detection method according to claim 1, wherein the denoising processing is performed on the hyperspectral segmented image to obtain a post-processing segmented image, and the method comprises the following steps:
binarizing the hyperspectral segmentation image to obtain a binary image X;
performing a morphological opening operation on the binary image X according to
Obtaining an image
In the formula, Θ represents corrosion,
denotes swelling, S denotes a structural element;
the image
Computing images for undirected weighted graphs
Obtaining connectivity between adjacent pixels, and numbering the connectivity areas;
checking the number of pixels contained in each communication area according to the serial numbers, and deleting the communication areas with the number of pixels lower than a preset value to obtain an image
Comparing the images
With the high spectrumAnd cutting the image to obtain the post-processing segmentation image.
4. The hyperspectral target detection method according to claim 1, wherein the noise removed in the denoising process is at least one of isolated pixels predicted to be of a certain class, burrs or small bridges of a connected region predicted to be of a certain class, and connected regions predicted to be of a certain class and having a number of pixels lower than a preset value.
5. The hyperspectral target detection method according to claim 1, wherein the pixel clustering based on the post-processing segmented image to obtain a cluster center point set corresponding to each of the categories comprises:
and clustering pixels with the same category in the post-processing segmentation image to obtain a cluster central point set corresponding to each category, wherein the cluster central point set comprises one or more cluster central points.
6. The hyperspectral target detection method according to claim 1, wherein the clustering algorithm of the pixel clustering is a K-means clustering algorithm, and the obtaining the cluster center point set corresponding to each of the categories comprises:
inputting coordinates of all pixels of the category;
determining the number K of the optimal clustering center points of the categories by utilizing an elbow rule;
and obtaining K clustering central points and forming a clustering central point set.
7. The hyperspectral target detection method of claim 2, wherein the spatial resolution and spectral resolution of the downsampled hyperspectral image are less than or equal to the spatial resolution and spectral resolution of the original hyperspectral image.
8. The hyperspectral target detection method according to claim 2, wherein the height and width of the hyperspectral split image are equal to the height and width of the downsampled hyperspectral image, respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010687080.8A CN111833353B (en) | 2020-07-16 | 2020-07-16 | Hyperspectral target detection method based on image segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010687080.8A CN111833353B (en) | 2020-07-16 | 2020-07-16 | Hyperspectral target detection method based on image segmentation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111833353A CN111833353A (en) | 2020-10-27 |
| CN111833353B true CN111833353B (en) | 2022-04-12 |
Family
ID=72923430
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010687080.8A Active CN111833353B (en) | 2020-07-16 | 2020-07-16 | Hyperspectral target detection method based on image segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111833353B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112668582A (en) * | 2020-12-31 | 2021-04-16 | 北京迈格威科技有限公司 | Image recognition method, device, equipment and storage medium |
| CN113516784B (en) * | 2021-07-27 | 2023-05-23 | 四川九洲电器集团有限责任公司 | Tooth segmentation modeling method and device |
| CN114882520B (en) * | 2022-07-08 | 2022-11-11 | 成都西交智汇大数据科技有限公司 | Method, system and equipment for detecting circuit diagram and readable storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880872A (en) * | 2012-08-28 | 2013-01-16 | 中国科学院东北地理与农业生态研究所 | Classification and construction method for semi-supervised support vector machine (SVM) remote sensing image |
| CN108009559A (en) * | 2016-11-02 | 2018-05-08 | 哈尔滨工业大学 | A kind of Hyperspectral data classification method based on empty spectrum united information |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101930547A (en) * | 2010-06-24 | 2010-12-29 | 北京师范大学 | Method for automatically classifying remote sensing image based on object-oriented unsupervised classification |
| CN104537067A (en) * | 2014-12-30 | 2015-04-22 | 广东电网有限责任公司信息中心 | Box separation method based on k-means clustering |
| CN104992454B (en) * | 2015-08-11 | 2017-07-04 | 辽宁工程技术大学 | A kind of compartmentalization becomes the image partition method of class automatically |
| WO2018095516A1 (en) * | 2016-11-22 | 2018-05-31 | Universidad De Las Palmas De Gran Canaria | Method of non-invasive detection of tumour and/or healthy tissue and hyperspectral imaging apparatus |
| CN107194936B (en) * | 2017-05-24 | 2021-02-02 | 哈尔滨工业大学 | Hyperspectral image target detection method based on superpixel combined sparse representation |
| CN110135432A (en) * | 2019-05-24 | 2019-08-16 | 哈尔滨工程大学 | A kind of high-spectrum remote sensing dividing method based on K-means cluster |
-
2020
- 2020-07-16 CN CN202010687080.8A patent/CN111833353B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880872A (en) * | 2012-08-28 | 2013-01-16 | 中国科学院东北地理与农业生态研究所 | Classification and construction method for semi-supervised support vector machine (SVM) remote sensing image |
| CN108009559A (en) * | 2016-11-02 | 2018-05-08 | 哈尔滨工业大学 | A kind of Hyperspectral data classification method based on empty spectrum united information |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111833353A (en) | 2020-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084292B (en) | Target detection method based on DenseNet and multi-scale feature fusion | |
| Chen et al. | Pothole detection using location-aware convolutional neural networks | |
| CN107424159B (en) | Image semantic segmentation method based on super-pixel edge and full convolution network | |
| Jiao et al. | A configurable method for multi-style license plate recognition | |
| CN108830285B (en) | Target detection method for reinforcement learning based on fast-RCNN | |
| CN111833353B (en) | Hyperspectral target detection method based on image segmentation | |
| CN108681693B (en) | License plate recognition method based on trusted area | |
| CN111709416B (en) | License plate positioning method, device, system and storage medium | |
| CN110287826B (en) | Video target detection method based on attention mechanism | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| CN108492298B (en) | Multispectral image change detection method based on generation countermeasure network | |
| Tao et al. | Scene context-driven vehicle detection in high-resolution aerial images | |
| CN109934216B (en) | Image processing method, device and computer readable storage medium | |
| CN111950488B (en) | Improved Faster-RCNN remote sensing image target detection method | |
| CN111274926B (en) | Image data screening method, device, computer equipment and storage medium | |
| CN113887459A (en) | Open-pit mining area stope change area detection method based on improved Unet + | |
| CN112232371A (en) | American license plate recognition method based on YOLOv3 and text recognition | |
| US20150356350A1 (en) | unsupervised non-parametric multi-component image segmentation method | |
| Hoang et al. | A novel approach for detection of pavement crack and sealed crack using image processing and salp swarm algorithm optimized machine learning | |
| CN112906816A (en) | Target detection method and device based on optical differential and two-channel neural network | |
| CN114419151A (en) | Multi-target tracking method based on contrast learning | |
| CN114332644B (en) | Large-view-field traffic density acquisition method based on video satellite data | |
| CN111597875A (en) | Traffic sign identification method, device, equipment and storage medium | |
| CN112819753B (en) | Building change detection method and device, intelligent terminal and storage medium | |
| CN114299383A (en) | Remote sensing image target detection method based on integration of density map and attention mechanism |
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 |