CN110008948B - Hyperspectral image target detection method based on variational self-coding network - Google Patents
Hyperspectral image target detection method based on variational self-coding network Download PDFInfo
- Publication number
- CN110008948B CN110008948B CN201910298153.1A CN201910298153A CN110008948B CN 110008948 B CN110008948 B CN 110008948B CN 201910298153 A CN201910298153 A CN 201910298153A CN 110008948 B CN110008948 B CN 110008948B
- Authority
- CN
- China
- Prior art keywords
- hyperspectral image
- detected
- coding network
- image
- target
- 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 76
- 230000003595 spectral effect Effects 0.000 claims abstract description 53
- 239000013598 vector Substances 0.000 claims abstract description 40
- 238000010586 diagram Methods 0.000 claims abstract description 21
- 238000000034 method Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 14
- 230000004927 fusion Effects 0.000 claims abstract description 6
- 238000001228 spectrum Methods 0.000 claims description 26
- 238000001914 filtration Methods 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 8
- 238000004088 simulation Methods 0.000 description 14
- 208000003569 Central serous chorioretinopathy Diseases 0.000 description 9
- 238000003331 infrared imaging Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 239000000126 substance Substances 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 229910052500 inorganic mineral Inorganic materials 0.000 description 2
- 239000011707 mineral Substances 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 239000013076 target substance Substances 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000003653 coastal water Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/143—Sensing or illuminating at different wavelengths
-
- 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
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a hyperspectral image target detection method based on a variational self-coding network, which mainly solves the technical problem of low detection precision in the prior art and comprises the following steps: acquiring a hyperspectral image to be detected and a real spectral vector of a target to be detected; constructing a variation self-coding network and training the variation self-coding network; acquiring a characteristic diagram of a hyperspectral image to be detected; calculating a spectral vector corresponding to the position of the maximum pixel value in each characteristic map in the hyperspectral image to be detected; calculating the spectral angle between each spectral vector and the true spectral vector; acquiring a fusion image; acquiring an initial detection image of a hyperspectral image to be detected; and acquiring a final detection target of the hyperspectral image to be detected. The method can reduce the frequency band interference in the hyperspectral image, reduce redundant information, better distinguish the target and the complex background in the hyperspectral image, improve the detection precision of the target point and reduce the complexity of data processing.
Description
Technical Field
The invention belongs to the technical field of image processing, relates to a hyperspectral image target detection method, and particularly relates to a hyperspectral image target detection method based on a variational self-coding network, which can be used for detecting a target similar to a known spectral curve from a hyperspectral image.
Background
The hyperspectral image of the object to be measured acquired by the hyperspectral imager contains abundant space, spectrum and radiation information of the object to be measured, the information not only shows the image characteristics of the space distribution of the object to be measured, but also can acquire the radiation intensity and the spectrum characteristics of the object to be measured by taking a certain pixel or pixel group as a target. Different substances contain unique spectral characteristics, the hyperspectral image contains spectral information of various substances, a known target substance spectral curve is matched with a spectral curve of each spectral vector in the hyperspectral image, the target substances in the background can be effectively identified, accurate detection of different substance targets is realized, however, the acquired real hyperspectral image is usually complex in background and easy to be interfered by noise, a large amount of redundancy exists in image data, the correlation among wave bands is high, and the method becomes an important problem to be overcome for improving the detection accuracy in the target detection process.
The variational self-coding network is an unsupervised learning network, can automatically learn characteristics from unmarked data, is a neural network which takes the reconstruction of data which is similar to input data as far as possible as a target, can give out better characteristic description than original data, has stronger characteristic learning capability, replaces the original data with the characteristics generated by the variational self-coding network in deep learning, reduces the interference among wave bands in the original data, reduces the dimensionality of the original data and obtains better effect; moreover, due to the fact that the hyperspectral image is difficult to acquire, the number of image sources is small, and the unsupervised network can train a network with good learning capacity by using less data which do not need to be marked, the variational self-coding network provides a good processing means in the aspect of hyperspectral image target detection.
The goal of hyperspectral image object detection is to identify objects that are spectrally identical or similar to the prior spectral curve, e.g., the objects may be rare vegetation species, vegetation of abnormal growth, illegal plants related to drug transactions, contaminated areas of coastal waters, adventures missing in deserts, buried archaeological structures, illegal crosses and vegetation covered vehicles, ships in marine backgrounds, etc. Therefore, the hyperspectral image target detection is effectively carried out, the hyperspectral image target detection method plays an important role in the fields of mineral detection, ocean research, agriculture, environmental protection and the like, and further improves social and economic benefits, and has incomparable advantages compared with other image target detection technologies.
According to background uncertainty, traditional hyperspectral target detection is roughly divided into a detection algorithm based on a probability statistical model, a detection algorithm based on a subspace model, a detection algorithm based on a kernel method and a detection algorithm based on spatial-spectral information combination. A paper published by Li W in Elsevier Science Inc in 2015 under the name of Combined space and collective representation for hyperspectral target detection discloses a hyperspectral target detection method based on combination of sparse representation and collaborative representation, and the method firstly represents a test sample through target atom sparse representation; then, cooperatively representing a test sample by using background atoms; and finally, the decision is realized by calculating the difference between the two representation residuals, and the detection result is obtained. Although the method can improve the target detection effect by utilizing the difference of sparse representation and collaborative representation, the method still has the defects that the target detection method is directly applied to the hyperspectral image with complex background and frequency band interference due to the complex background and more interference information of the hyperspectral image, the target is easy to miss detection, and the target detection precision is low; and because the original hyperspectral image is directly processed, the multiband characteristic of the hyperspectral image makes the data processing more complex.
Disclosure of Invention
The invention aims to provide a hyperspectral image target detection method based on a variational self-coding network aiming at the defects in the prior art, and the hyperspectral image target detection method is used for solving the technical problem of low detection precision in the prior art.
In order to achieve the purpose, the technical scheme adopted by the invention comprises the following steps:
(1) acquiring a hyperspectral image to be detected and a real spectral vector of a target to be detected:
selecting a hyperspectral image I to be detected with the size of W multiplied by H multiplied by L from a hyperspectral image library, and a real spectral vector d similar to a target spectral curve to be detected contained in the hyperspectral image I to be detected, wherein W, H, L respectively represents the number of the width, the height and the wave band of the hyperspectral image I to be detected, W is more than 0, H is more than 0, and L is more than or equal to 100;
(2) constructing a variational self-coding network, and training the variational self-coding network:
(2a) constructing a variational self-coding network comprising a coding network and a decoding network connected with the coding network, wherein the coding network comprises a first hidden layer and a first output layer connected with the first hidden layer, the total number of nodes of the first output layer is equal to the number m of wave bands of an output characteristic diagram of the first output layer, the size of the characteristic diagram is W multiplied by H multiplied by m, and m is more than or equal to 2 and less than or equal to L; the decoding network comprises a second hidden layer and a second output layer connected with the second hidden layer, the total number of nodes of the second output layer is equal to the number L of wave bands of the hyperspectral image I to be detected, and the size of a reconstructed hyperspectral image R output by the second output layer is W multiplied by H multiplied by L;
(2b) inputting a hyperspectral image I to be detected into a variational self-coding network for iterative training to obtain a trained variational self-coding network;
(3) acquiring a characteristic diagram of a hyperspectral image to be detected:
taking the hyperspectral image I to be detected as the input of the trained variational self-coding network in the coding network to obtain m characteristic maps { F) of the output end of the coding network1,F2,…Fs,…,Fm}，FsIs the s-th characteristic diagram, and s is more than or equal to 2 and less than or equal to m;
(4) calculating a spectral vector corresponding to the position of the maximum pixel value in each characteristic map in the hyperspectral image to be detected:
(4a) calculating each feature map FsCoordinate (i) corresponding to the middle maximum pixel values,js)：
(is,js)＝max{Ps(i,j)|0＜i≤W,0＜j≤H}
Where max denotes the maximum value operation, Ps(i, j) represents a feature map FsPixel values for each coordinate;
(4b) determining each of (i)s,js) Corresponding spectral vector I (I) in hyperspectral image I to be detecteds,js) Obtaining m spectral vectors;
(5) calculating the spectral angle of each spectral vector with the true spectral vector:
calculating each spectral vector I (I)s,js) Spectral angles sam(s) with the true spectral vector d, resulting in m spectral angles, wherein:
wherein arccos represents an inverse cosine operation, | ·| luminance2Representing a two-norm operation;
(6) acquiring a fusion image:
(6a) selecting K spectrum angles satisfying SAM(s) and less than or equal to tau from the m spectrum angles, wherein K is more than or equal to 2 and less than or equal to m, and extracting a feature map corresponding to each spectrum angle to obtain K feature maps, wherein tau represents a set threshold value;
(6b) and fusing the K characteristic graphs to obtain a fused image f:
wherein, FnRepresenting the nth characteristic diagram;
(7) acquiring an initial detection image of a hyperspectral image to be detected:
performing attribute filtering on the fused image f to obtain an opening operation sketch A, an original operation sketch C and a closing operation sketch E, and calculating an initial detection image D of the hyperspectral image I to be detected:
D＝|A-C|+|E-C|
wherein, | · | represents an absolute value taking operation;
(8) acquiring a final detection target of a hyperspectral image to be detected:
and (4) conducting guiding filtering on the initial detection image D of the hyperspectral image I to be detected to obtain a target similar to a known real target spectrum curve in the hyperspectral image I to be detected.
Compared with the prior art, the invention has the following advantages:
firstly, the invention adopts a variational self-coding network, the spectrum information of the hyperspectral image is represented by the hidden nodes of the network, the extracted characteristic graph reduces the interference between wave bands of the original hyperspectral image and the redundant information in the data, the spectrum characteristic of the target is effectively extracted, the target and the complex background in the hyperspectral image are better distinguished, the problems of complex background and low detection precision of the hyperspectral image with frequency band interference in the prior art are solved, the omission factor of the hyperspectral image target can be reduced, and the detection precision is effectively improved.
Secondly, the invention uses the spectral angle to select the features of the feature map extracted by the variational self-coding network, obtains the low-dimensional identification map by self-adaptive weighting fusion, reduces the data processing amount by performing the subsequent processing on the low-dimensional identification map, and simultaneously uses the attribute filtering and the guide filtering to perform the post-processing, thereby reducing the miss detection rate of the target point, solving the problems of complex processing and low detection precision of the hyperspectral image data in the prior art, greatly reducing the complexity of the data processing in the hyperspectral target detection process, improving the detection efficiency and further improving the detection precision.
Drawings
FIG. 1 is a flow chart of an implementation of the present invention;
FIG. 2 is a simulation diagram of the detection result of the hyperspectral target detection method based on the combination of sparse representation and collaborative representation of CSCR in the prior art;
FIG. 3 is a simulation of the test results using the method of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and the specific embodiments.
Referring to fig. 1, the present invention includes the steps of:
step 1) acquiring a hyperspectral image to be detected and a real spectral vector of a target to be detected:
selecting a hyperspectral image I to be detected with the size of WxHxL and a real spectral vector d similar to a target spectral curve to be detected contained in the hyperspectral image I to be detected from a hyperspectral image library, wherein W, H, L respectively represents the number of the width, the height and the wave band of the hyperspectral image I to be detected, W is more than 0, H is more than 0, L is more than or equal to 100, in the example, the hyperspectral image I to be detected is a real hyperspectral image collected by an airborne visible light/infrared imaging spectrometer (AVIRIS), the size of the hyperspectral image I to be detected is 80 x 100 x 189, and the real spectral vector d is a spectral vector of a certain mineral substance;
step 2), constructing a variational self-coding network, and training the variational self-coding network:
(2a) constructing a variation self-coding network comprising a coding network and a decoding network connected with the coding network, wherein the coding network comprises a first hidden layer and a first output layer connected with the first hidden layer, the total number of nodes of the first hidden layer is V epsilon [50,80], the total number of nodes of the first output layer is equal to the wave band number m of a characteristic diagram output by the first output layer, the size of the characteristic diagram is W multiplied by H multiplied by m, and m is more than or equal to 2 and less than or equal to L; the decoding network comprises a second hidden layer and a second output layer connected with the second hidden layer, the total number of nodes of the second hidden layer is equal to that of nodes of the first hidden layer, the total number of nodes of the second output layer is equal to the number L of wave bands of the hyperspectral image I to be detected, and the size of a reconstructed hyperspectral image R output by the second output layer is W multiplied by H multiplied by L; in this example, V is 70, m is 20;
(2b) inputting a hyperspectral image to be detected into a variational self-coding network for iterative training to obtain a trained variational self-coding network;
(2b1) setting the training iteration number as T, the training total iteration number as Y, and making T equal to 1, in this example, Y equal to 50;
(2b2) inputting a hyperspectral image I to be detected into a first hidden layer of a coding network for coding, storing neuron nodes of the coding network after updating, and outputting a coded feature map through a first output layer;
(2b3) inputting the coded feature map into a second hidden layer of a decoding network for decoding, storing neuron nodes of the updated decoding network, and outputting a decoded reconstructed hyperspectral image R through a second output layer;
(2b4) judging whether T is equal to Y, if yes, executing (2b 5); otherwise, let T be T +1, and perform step (2b 2);
(2b5) calculating an average spectrum angle of each spectrum vector I (u, v) of the hyperspectral image I to be detected and each spectrum vector R (u, v) of the reconstructed hyperspectral image R:
wherein mean represents the operation of taking the mean, arccos represents the operation of taking the inverse cosine, | ·| luminance2Operation for representing two normMaking;
(2b6) set the threshold to ξ and will satisfy
And the time-varying variational self-coding network is used as the variational self-coding network after training.
Step 3) obtaining a characteristic diagram of the hyperspectral image to be detected:
taking the hyperspectral image I to be detected as the input of the trained variational self-coding network in the coding network to obtain m characteristic maps { F) of the output end of the coding network1,F2,…Fs,…,Fm}，FsIs the s-th characteristic diagram, and s is more than or equal to 2 and less than or equal to m;
step 4) calculating a spectral vector corresponding to the position of the maximum pixel value in each characteristic map in the hyperspectral image to be detected:
(4a) calculating each feature map FsCoordinate (i) corresponding to the middle maximum pixel values,js)：
(is,js)＝max{Ps(i,j)|0＜i≤W,0＜j≤H}
Where max denotes the maximum value operation, Ps(i, j) represents a feature map FsPixel values for each coordinate;
(4b) determining each of (i)s,js) Corresponding spectral vector I (I) in hyperspectral image I to be detecteds,js) Obtaining m spectral vectors;
step 5) calculating the spectrum angle between each spectrum vector and the true spectrum vector:
calculating each spectral vector I (I)s,js) Spectral angles sam(s) with the true spectral vector d, resulting in m spectral angles, wherein:
wherein arccos represents an inverse cosine operation, | ·| luminance2Representing a two-norm operation;
step 6), acquiring a fusion image:
(6a) selecting K spectrum angles satisfying SAM(s) and less than or equal to tau from the m spectrum angles, wherein K is more than or equal to 2 and less than or equal to m, and extracting a feature map corresponding to each spectrum angle to obtain K feature maps, wherein tau represents a set threshold value;
(6b) and fusing the K characteristic graphs to obtain a fused image f:
wherein, FnRepresenting the nth characteristic diagram;
step 7) obtaining an initial detection image of the hyperspectral image to be detected:
performing attribute filtering on the fused image f to obtain an opening operation sketch A, an original operation sketch C and a closing operation sketch E, and calculating an initial detection image D of the hyperspectral image I to be detected:
D＝|A-C|+|E-C|
wherein, | · | represents an absolute value taking operation;
step 8) obtaining a final detection target of the hyperspectral image to be detected:
and (4) conducting guiding filtering on the initial detection image D of the hyperspectral image I to be detected to obtain a target similar to a known real target spectrum curve in the hyperspectral image I to be detected.
The effect of the present invention will be further described with reference to simulation experiments.
1. Simulation conditions are as follows:
the simulation experiment of the invention is carried out under the conditions of Intel (R) core (TM) i5-7200U CPU with main frequency of 2.50GHz x 8, hardware environment with internal memory of 8GB and software environment of MATLAB.
2. Simulation content and result analysis:
the simulation experiment of the invention is to adopt the method of the invention and the hyperspectral target detection method based on the combination of sparse representation and collaborative representation CSCR in the prior art to carry out simulation, and two simulation experiments are respectively carried out under the simulation conditions.
Referring to fig. 2, a detailed description will be given of a simulation experiment 1 performed by using a simulation diagram of a hyperspectral target detection method based on a combination of sparse representation and collaborative representation CSCR in the prior art. Fig. 2(a) is a real hyperspectral image acquired by an airborne visible light/infrared imaging spectrometer (AVIRIS), fig. 2(b) is a target point distribution diagram attached to the real hyperspectral image acquired by the airborne visible light/infrared imaging spectrometer (AVIRIS), and white bright spots in fig. 2(b) represent target points in the real hyperspectral image. Fig. 2(c) is a graph of the detection result of fig. 2(a) by using a hyperspectral target detection method based on combined sparse and collaborative representation CSCR in the prior art, and white bright spots in fig. 2(c) represent detected target points.
The simulation experiment 2 using the method of the present invention will be described in detail with reference to fig. 3. Fig. 3(a) is a real hyperspectral image acquired by an airborne visible light/infrared imaging spectrometer (AVIRIS), fig. 3(b) is a target point distribution diagram attached to the real hyperspectral image acquired by the airborne visible light/infrared imaging spectrometer (AVIRIS), and white bright spots in fig. 3(b) represent target points in the real hyperspectral image. Fig. 3(c) is a graph of the results of a simulation experiment using the method of the present invention, and the bright spots in fig. 3(c) represent detected target points.
Comparing fig. 2(c) and fig. 3(c) it can be seen that: compared with the hyperspectral target detection method based on the combined sparse representation and the collaborative representation of the CSCR in the prior art, the number of the target points detected by the hyperspectral target detection method based on the combined sparse representation and the collaborative representation of the CSCR in the prior art is 9, and the number of the target points detected by the hyperspectral target detection method based on the combined sparse representation and the collaborative representation of the CSCR in the prior art is only 6, so that the hyperspectral target detection method based on the combined sparse representation and the collaborative representation of the CSCR in the prior art has the advantages of more detected target points, reduced omission ratio and better detection effect.
In order to evaluate the detection performance of the two methods, the detection accuracy AUC value was calculated according to the following formula:
wherein eta represents the AUC value of the detection precision, alpha represents the number of target points in the detection result, and beta represents the number of target points in a target point distribution diagram attached to a real hyperspectral image acquired by an airborne visible light/infrared imaging spectrometer (AVIRIS). The two methods calculate the detection Accuracy (AUC) values as shown in the table below.
TABLE 1 comparison table of target point detection accuracy between the method of the present invention and the prior art method
| Method type | Detection accuracy AUC |
| Prior Art | 66.67％ |
| The invention | 100％ |
As can be seen from Table 1, compared with the hyperspectral target detection method based on combined sparse and collaborative representation CSCR in the prior art, the detection precision obtained by using the method is obviously improved.
In conclusion, the spectral vector dimension characteristics of the hyperspectral image to be detected are extracted by using the variational self-coding network, the prior target spectrum and the spectral vector dimension characteristics are used for feature selection, and then the targets are clustered and detected by using wave band fusion, attribute filtering and guide filtering to obtain the detection target of the hyperspectral image, so that the frequency band interference in the hyperspectral image is reduced, the redundant information is reduced, and the detection precision of the target is improved.
Claims (3)
1. A hyperspectral image target detection method based on a variational self-coding network is characterized by comprising the following steps:
(1) acquiring a hyperspectral image to be detected and a real spectral vector of a target to be detected:
selecting a hyperspectral image I to be detected with the size of W multiplied by H multiplied by L from a hyperspectral image library, and a real spectral vector d similar to a target spectral curve to be detected contained in the hyperspectral image I to be detected, wherein W, H, L respectively represents the number of the width, the height and the wave band of the hyperspectral image I to be detected, W is more than 0, H is more than 0, and L is more than or equal to 100;
(2) constructing a variational self-coding network, and training the variational self-coding network:
(2a) constructing a variational self-coding network comprising a coding network and a decoding network connected with the coding network, wherein the coding network comprises a first hidden layer and a first output layer connected with the first hidden layer, the total number of nodes of the first output layer is equal to the number m of wave bands of an output characteristic diagram of the first output layer, the size of the characteristic diagram is W multiplied by H multiplied by m, and m is more than or equal to 2 and less than or equal to L; the decoding network comprises a second hidden layer and a second output layer connected with the second hidden layer, the total number of nodes of the second output layer is equal to the number L of wave bands of the hyperspectral image I to be detected, and the size of a reconstructed hyperspectral image R output by the second output layer is W multiplied by H multiplied by L;
(2b) inputting a hyperspectral image I to be detected into a variational self-coding network for iterative training to obtain a trained variational self-coding network;
(3) acquiring a characteristic diagram of a hyperspectral image to be detected:
taking the hyperspectral image I to be detected as the input of the trained variational self-coding network in the coding network to obtain m characteristic maps { F) of the output end of the coding network1,F2,…Fs,…,Fm}，FsIs the s-th characteristic diagram, and s is more than or equal to 2 and less than or equal to m;
(4) calculating a spectral vector corresponding to the position of the maximum pixel value in each characteristic map in the hyperspectral image to be detected:
(4a) calculating each feature map FsCoordinate (i) corresponding to the middle maximum pixel values,js)：
(is,js)＝max{Ps(i,j)|0＜i≤W,0＜j≤H}
Where max denotes the maximum value operation, Ps(i, j) represents a feature map FsPixel of each coordinateA value;
(4b) determining each of (i)s,js) Corresponding spectral vector I (I) in hyperspectral image I to be detecteds,js) Obtaining m spectral vectors;
(5) calculating the spectral angle of each spectral vector with the true spectral vector:
calculating each spectral vector I (I)s,js) Spectral angles sam(s) with the true spectral vector d, resulting in m spectral angles, wherein:
wherein arccos represents an inverse cosine operation, | ·| luminance2Representing a two-norm operation;
(6) acquiring a fusion image:
(6a) selecting K spectrum angles satisfying SAM(s) and less than or equal to tau from the m spectrum angles, wherein K is more than or equal to 2 and less than or equal to m, and extracting a feature map corresponding to each spectrum angle to obtain K feature maps, wherein tau represents a set threshold value;
(6b) and fusing the K characteristic graphs to obtain a fused image f:
wherein, FnRepresenting the nth characteristic diagram;
(7) acquiring an initial detection image of a hyperspectral image to be detected:
performing attribute filtering on the fused image f to obtain an opening operation sketch A, an original operation sketch C and a closing operation sketch E, and calculating an initial detection image D of the hyperspectral image I to be detected:
D＝|A-C|+|E-C|
wherein, | · | represents an absolute value taking operation;
(8) acquiring a final detection target of a hyperspectral image to be detected:
and (4) conducting guiding filtering on the initial detection image D of the hyperspectral image I to be detected to obtain a target similar to a known real target spectrum curve in the hyperspectral image I to be detected.
2. The method for detecting the hyperspectral image object based on the variational self-coding network as claimed in claim 1, wherein the total number of nodes of the first hidden layer in the step (2a) is V e [50,80], and the total number of nodes of the second hidden layer is equal to the total number of nodes of the first hidden layer.
3. The hyperspectral image target detection method based on the variational self-coding network according to claim 1 is characterized in that the hyperspectral image to be detected is input into the variational self-coding network for iterative training in the step (2b), and the implementation steps are as follows:
(2b1) setting the training iteration number as T, setting the training total iteration number as Y, and setting T as 1;
(2b2) inputting a hyperspectral image I to be detected into a first hidden layer of a coding network for coding, storing neuron nodes of the coding network after updating, and outputting a coded feature map through a first output layer;
(2b3) inputting the coded feature map into a second hidden layer of a decoding network for decoding, storing neuron nodes of the updated decoding network, and outputting a decoded reconstructed hyperspectral image R through a second output layer;
(2b4) judging whether T is equal to Y, if yes, executing (2b 5); otherwise, let T be T +1, and perform step (2b 2);
(2b5) calculating an average spectrum angle of each spectrum vector I (u, v) of the hyperspectral image I to be detected and each spectrum vector R (u, v) of the reconstructed hyperspectral image R:
wherein mean represents the operation of taking the mean, arccos represents the operation of taking the inverse cosine, | ·| luminance2Representing a two-norm operation;
(2b6) setting the threshold to ξAnd will satisfy
And the time-varying variational self-coding network is used as the variational self-coding network after training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910298153.1A CN110008948B (en) | 2019-04-15 | 2019-04-15 | Hyperspectral image target detection method based on variational self-coding network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910298153.1A CN110008948B (en) | 2019-04-15 | 2019-04-15 | Hyperspectral image target detection method based on variational self-coding network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110008948A CN110008948A (en) | 2019-07-12 |
| CN110008948B true CN110008948B (en) | 2021-06-01 |
Family
ID=67171739
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910298153.1A Active CN110008948B (en) | 2019-04-15 | 2019-04-15 | Hyperspectral image target detection method based on variational self-coding network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110008948B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619373B (en) * | 2019-10-31 | 2021-11-26 | 北京理工大学 | Infrared multispectral weak target detection method based on BP neural network |
| CN111564188B (en) * | 2020-04-29 | 2023-09-12 | 核工业北京地质研究院 | Quantitative analysis method based on variation self-coding mineral information |
| CN111783884B (en) * | 2020-06-30 | 2024-04-09 | 山东女子学院 | Unsupervised hyperspectral image classification method based on deep learning |
| CN112906750B (en) * | 2021-01-25 | 2022-04-29 | 浙江大学 | Hyperspectral image-based material analysis method and system |
| CN112766223B (en) * | 2021-01-29 | 2023-01-06 | 西安电子科技大学 | Hyperspectral image target detection method based on sample mining and background reconstruction |
| CN113643364A (en) * | 2021-07-05 | 2021-11-12 | 珠海格力电器股份有限公司 | Image target detection method, device and equipment |
| CN114118308B (en) * | 2022-01-26 | 2022-05-20 | 南京理工大学 | Hyperspectral target detection method based on constrained energy minimization variational self-coding |
Family Cites Families (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6771798B1 (en) * | 1998-11-03 | 2004-08-03 | The United States Of America As Represented By The Secretary Of The Navy | Hyperspectral visualization extensible workbench |
| CN101710387B (en) * | 2009-10-29 | 2013-02-06 | 中国科学院对地观测与数字地球科学中心 | Intelligent method for classifying high-resolution remote sensing images |
| CN101894270B (en) * | 2010-07-26 | 2013-10-02 | 中国科学院遥感与数字地球研究所 | Method for full-automatic sample selection oriented to classification of remote-sensing images |
| US8897570B1 (en) * | 2011-03-31 | 2014-11-25 | Raytheon Company | Detection of targets from hyperspectral imagery |
| GB2506688A (en) * | 2012-10-08 | 2014-04-09 | Bae Systems Plc | Detection of a target in a scene using hyperspectral imaging |
| CN103714341B (en) * | 2014-01-21 | 2016-09-28 | 北京航空航天大学 | Hyper spectral reflectance data light spectrum signature extracting method based on overall situation sensitivity analysis |
| CN104298999B (en) * | 2014-09-30 | 2017-08-25 | 西安电子科技大学 | EO-1 hyperion feature learning method based on recurrence autocoding |
| CN105224960B (en) * | 2015-11-04 | 2018-07-20 | 江南大学 | Corn seed classification hyperspectral imagery identification model update method based on clustering algorithm |
| US10496883B2 (en) * | 2017-01-27 | 2019-12-03 | Signal Processing, Inc. | Method and system for enhancing predictive accuracy of planet surface characteristics from orbit |
| CN107368846A (en) * | 2017-06-22 | 2017-11-21 | 华南理工大学 | Hyperspectral image classification method based on wavelet transformation and rarefaction representation |
| CN109146890B (en) * | 2018-07-16 | 2020-07-31 | 西安电子科技大学 | Abnormal target detection method of hyperspectral image based on filter |
-
2019
- 2019-04-15 CN CN201910298153.1A patent/CN110008948B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110008948A (en) | 2019-07-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110008948B (en) | Hyperspectral image target detection method based on variational self-coding network | |
| CN111738124B (en) | Remote sensing image cloud detection method based on Gabor transformation and attention | |
| CN108389188B (en) | Sparse hyperspectral abnormal target detection method | |
| CN110298235B (en) | Hyperspectral anomaly detection method and system based on manifold constraint self-coding network | |
| CN113990384A (en) | Deep learning-based frozen electron microscope atomic model structure building method and system and application | |
| CN105989597B (en) | Hyperspectral image abnormal target detection method based on pixel selection process | |
| CN112766227B (en) | Hyperspectral remote sensing image classification method, device, equipment and storage medium | |
| CN111339827A (en) | SAR image change detection method based on multi-region convolutional neural network | |
| CN111008664B (en) | Hyperspectral sea ice detection method based on space-spectrum combined characteristics | |
| CN112766223B (en) | Hyperspectral image target detection method based on sample mining and background reconstruction | |
| CN113378971B (en) | Classification model training method and system for near infrared spectrum and classification method and system | |
| CN103150577A (en) | Method for selecting high spectrum remotely sensed image self-adaptive wave band based on particle swarm optimization | |
| CN114398948A (en) | Multispectral image change detection method based on space-spectrum combined attention network | |
| CN116402825B (en) | Bearing fault infrared diagnosis method, system, electronic equipment and storage medium | |
| CN111680579B (en) | Remote sensing image classification method for self-adaptive weight multi-view measurement learning | |
| CN113421198B (en) | Hyperspectral image denoising method based on subspace non-local low-rank tensor decomposition | |
| CN112784777B (en) | Unsupervised hyperspectral image change detection method based on countermeasure learning | |
| Zhang et al. | Feature-band-based unsupervised hyperspectral underwater target detection near the coastline | |
| CN110751144B (en) | Canopy plant hyperspectral image classification method based on sparse representation | |
| CN109460788B (en) | Hyperspectral image classification method based on low-rank-sparse information combination network | |
| CN113887656B (en) | Hyperspectral image classification method combining deep learning and sparse representation | |
| Chiu et al. | Evaluation of incorporating texture into wetland mapping from multispectral images | |
| CN118097313B (en) | Hyperspectral image classification method based on frequency domain active learning | |
| CN117115553B (en) | Hyperspectral remote sensing image classification method based on mask spectral space feature prediction | |
| Guo et al. | Image classification of solar radio spectrum based on deep learning |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221111 Address after: 710077 312-11, Block E, Science Park, Xi'an University of Technology, No. 26, Dengling Road, Zhangba Street Office, High tech Zone, Xi'an City, Shaanxi Province Patentee after: Shaanxi Silk Road Tiantu Satellite Technology Co.,Ltd. Address before: 710071 No. 2 Taibai South Road, Shaanxi, Xi'an Patentee before: XIDIAN University Patentee before: Xi'an Tongyuan Essen Enterprise Management Consulting Partnership (L.P.) Effective date of registration: 20221111 Address after: 710071 No. 2 Taibai South Road, Shaanxi, Xi'an Patentee after: XIDIAN University Patentee after: Xi'an Tongyuan Essen Enterprise Management Consulting Partnership (L.P.) Address before: 710071 No. 2 Taibai South Road, Shaanxi, Xi'an Patentee before: XIDIAN University |
|
| CI03 | Correction of invention patent | ||
| CI03 | Correction of invention patent |
Correction item: Patentee|Address Correct: Xi'an Electronic and Science University|710071 No. 2 Taibai South Road, Shaanxi, Xi'an False: Shaanxi Silk Road Tiantu Satellite Technology Co.,Ltd.|710077 312-11, Block E, Science Park, Xi'an University of Technology, No. 26, Dengling Road, Zhangba Street Office, High tech Zone, Xi'an City, Shaanxi Province Number: 47-02 Volume: 38 |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221226 Address after: 710071 Taibai South Road, Yanta District, Xi'an, Shaanxi Province, No. 2 Patentee after: XIDIAN University Patentee after: Xi'an Tongyuan Essen Enterprise Management Consulting Partnership (L.P.) Address before: 710071 No. 2 Taibai South Road, Shaanxi, Xi'an Patentee before: XIDIAN University Effective date of registration: 20221226 Address after: 710077 312-11, Block E, Science Park, Xi'an University of Technology, No. 26, Dengling Road, Zhangba Street Office, High tech Zone, Xi'an City, Shaanxi Province Patentee after: Shaanxi Silk Road Tiantu Satellite Technology Co.,Ltd. Address before: 710071 Taibai South Road, Yanta District, Xi'an, Shaanxi Province, No. 2 Patentee before: XIDIAN University Patentee before: Xi'an Tongyuan Essen Enterprise Management Consulting Partnership (L.P.) |