CN111598786B - Hyperspectral image unmixing method based on depth denoising self-coding network - Google Patents
Hyperspectral image unmixing method based on depth denoising self-coding network Download PDFInfo
- Publication number
- CN111598786B CN111598786B CN201911188298.2A CN201911188298A CN111598786B CN 111598786 B CN111598786 B CN 111598786B CN 201911188298 A CN201911188298 A CN 201911188298A CN 111598786 B CN111598786 B CN 111598786B
- Authority
- CN
- China
- Prior art keywords
- network
- denoising self
- image data
- hyperspectral image
- training
- 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
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 39
- 239000011159 matrix material Substances 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims abstract description 6
- 230000008569 process Effects 0.000 claims description 29
- 239000013598 vector Substances 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 230000003416 augmentation Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims 1
- 230000006870 function Effects 0.000 description 22
- 238000004422 calculation algorithm Methods 0.000 description 17
- 238000004088 simulation Methods 0.000 description 7
- 230000003595 spectral effect Effects 0.000 description 6
- 238000005457 optimization Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 238000001228 spectrum Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- IOCMOSUFEBBJOB-UHFFFAOYSA-K diethyl-bis[10-(triethylazaniumyl)decyl]azanium triiodide Chemical compound [I-].[I-].[I-].C(C)[N+](CCCCCCCCCC[N+](CC)(CC)CC)(CCCCCCCCCC[N+](CC)(CC)CC)CC IOCMOSUFEBBJOB-UHFFFAOYSA-K 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Classifications
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/002—Image coding using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- 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
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20084—Artificial neural networks [ANN]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a hyperspectral image unmixing method based on a depth denoising self-coding network. The invention comprises the following steps: designing a depth denoising self-coding network; inputting a group of training image data into a depth denoising self-coding network; extracting main characteristics of original data by encoding, and reconstructing the original data by decoding; continuously training to obtain optimized network parameters, so that the reconstructed data is more similar to the original data; after training, inputting test data, obtaining abundance coefficient of hyperspectral image by hidden layer, and outputting the obtained result by weight of the last layer of decoder as the obtained end member matrix. According to the invention, the weights of the hidden layer and the decoding layer are limited to be nonnegative on the basis of the traditional denoising encoder, the hidden layer is added and added as a constraint, and the L21 constraint is added as a regular term to the objective function, so that the joint sparsity between adjacent pixels is well utilized, and the precision of abundance estimation is improved.
Description
Technical Field
The invention belongs to the technical field of image processing and machine learning, and further relates to a hyperspectral image unmixing method based on a depth denoising self-coding network in the technical field of sparse unmixing.
Background
Hyperspectral images are characterized by very high spectral resolution, but by lower spatial resolution. The mixed pixels exist in a large amount in hyperspectral data under the influence of factors such as the atmospheric transmission mixed effect, the ground feature complexity, the low space resolution of a hyperspectral imager and the like, so that the improvement of hyperspectral image processing precision is restricted, and the mixed pixels become a main obstacle for impeding the deep development of hyperspectral remote sensing technology. Therefore, effective decomposition of the mixed pixels has become an important premise for wide application of hyperspectral images. The hybrid pixel can be regarded as a group of basis vectors combined in a certain proportion. The basis vector is the end member, and the certain proportion is the abundance.
The models aiming at the hyperspectral mixed pixels can be divided into a linear mixed model and a nonlinear mixed model, and the linear mixed model is simple in modeling and clear in physical meaning, so that at present, more researches at home and abroad adopt the linear mixed model. Conventional unmixing algorithms are statistical-based and geometric-based hyperspectral unmixing algorithms. With the vigorous development of compressed sensing and sparse representation theory, the Iordache et al introduce spectral sparsity into a unmixing model, and replace an end member set with a known spectral library to propose a sparse unmixing algorithm.
At present, sparse unmixing algorithms are mainly divided into convex optimization algorithms and greedy algorithms. The convex optimization algorithm mainly comprises SUnSAL, CL-SUnSAL, SUnSAL-TV, a weighted L1 regularization method and the like, and the convex optimization algorithm utilizes L1 norms to describe sparsity of abundance coefficients under certain conditions to carry out efficient solving. But the convex optimization algorithm solves slower than the greedy algorithm. Greedy algorithms such as Orthogonal Matching Pursuit (OMP) and matching pursuit algorithm (MP) are mainly based on a single observation vector (SMV) model, and do not consider the similarity between adjacent pixels when extracting end members, and are easy to fall into local optimum. The joint sparse solution mixing algorithm based on the multi-observation vector (MMV) model such as joint orthogonal matching pursuit (SOMP) and Subspace Matching Pursuit (SMP) adopts the joint sparse model and combines the blocking strategy to extract the end members, and compared with the algorithms such as OMP and MP, the global optimal solution can be obtained more accurately, and the defect is that excessive redundant end members exist in the end member set, so that the accuracy of abundance reconstruction is influenced.
Disclosure of Invention
The invention aims to provide a hyperspectral image unmixing method based on a depth denoising self-coding network, so as to improve the sparse unmixing precision of hyperspectral images.
The technical scheme of the invention is as follows: a hyperspectral image unmixing method based on a depth denoising self-coding network comprises the following steps:
(1) Adding a regular term into a network objective function according to the physical characteristics of the abundance coefficient and the characteristics of a constraint and a non-negative constraint on the basis of the denoising self-encoder, so as to form a depth denoising self-encoding network;
(2) Inputting a group of training image data into a depth denoising self-coding network, training the depth denoising self-coding network to obtain optimized network parameters, and obtaining a hyperspectral unmixed network model;
(3) Inputting the existing hyperspectral image data to be processed into a depth denoising self-coding network, extracting the characteristics of the hyperspectral image data through the coding process, and obtaining reconstructed original image data through decoding.
Further, in step (1), the depth denoising self-coding network includes an input layer, an implicit layer, and an output layer; the method comprises the steps of taking an encoding process from an input layer to an implicit layer, wherein the encoding process is to perform feature extraction on input hyperspectral image data to obtain a preliminary abundance coefficient, and taking a decoding process from the implicit layer to an output layer, wherein the decoding process is to decode the obtained preliminary abundance coefficient to obtain reconstructed original image data, namely;
adding a regular term to the network objective function to reduce redundant end members of the encoder, and introducing joint sparsity between adjacent pixels; selecting ReLu as an activation function can extract nonlinear characteristics of data and meet the non-negativity of an abundance coefficient; the objective function expression of the depth denoising self-coding network is as follows:
where X represents the input hyperspectral data, W represents the encoder weights, σ (X) represents the hidden layer activation function,and->Representing reconstructed image data +.>And an augmentation matrix of decoder weights a:
representing a squaring operation taking the F-norm, I.I 2,1 Represents a 2-norm addition operation taking each row vector in the matrix, λ represents a lagrangian coefficient, and the value of λ is set to 10e-6.
Further, in the step (2), training image data with the size of w×h×c is preprocessed to obtain w×h training sets with the size of 1×c, and then the training sets are input into a depth denoising self-coding network to perform multiple training so as to obtain optimized network parameters.
Furthermore, before the hyperspectral image data to be processed is input into the depth denoising self-coding network, the step (3) loads the depth denoising self-coding network model trained in the step (2), and then updates the network parameters into the network parameters trained in the step (2); then inputting hyperspectral image data, obtaining abundance coefficients through a network coding process, and obtaining an end member matrix through a decoding process.
Further, the training image data in the step (2) refers to hyperspectral image data as a training set, and the hyperspectral image to be processed in the step (3) refers to hyperspectral image data as a test set.
The method comprises the following steps: the depth denoising self-coding network is designed, the weight of an implicit layer and a decoding layer is limited to be nonnegative on the basis of a traditional denoising encoder, the constraint is added and becomes a constraint on the basis of the traditional denoising self-coding encoder, and the constraint L21 is added as a regular term to an objective function, so that the joint sparsity between adjacent pixels is utilized; inputting a group of training image data into a depth denoising self-coding network; extracting main characteristics of original data through network coding, and reconstructing the original data through decoding; continuously training to obtain optimized network parameters, so that the reconstructed data is more similar to the original data;
after training, inputting test data, obtaining abundance coefficient of hyperspectral image by hidden layer, and outputting the obtained result by weight of the last layer of decoder as the obtained end member matrix.
The invention has the beneficial effects that: the invention applies the denoising self-encoder to the hyperspectral unmixing network model by utilizing the characteristics of the denoising self-encoder, adds the L21 regular term into the objective function, reduces redundant rows of the encoder, well utilizes the joint sparsity between adjacent pixels, and solves the problem of low unmixing precision in the hyperspectral image unmixing process in the prior art, so that the invention has the advantage of high hyperspectral unmixing precision.
Drawings
FIG. 1 is a schematic diagram of the network model of the present invention;
FIG. 2 is a flow chart of the network model of the present invention;
FIG. 3 is a schematic representation of the raw abundance image corresponding to 9 end members in simulated data 1 of the present invention;
FIG. 4 is a schematic representation of an image of raw abundance in simulated data 2 of the present invention;
FIG. 5 is a schematic representation of an estimated abundance image of end members 1,5,9 of analog data 1 at 20dB Gaussian noise in the present invention;
FIG. 6 is a schematic representation of an estimated abundance image of end members 1,3,5 of analog data 2 at 20dB Gaussian noise in the present invention;
FIG. 7 is a schematic representation of abundance images of three different elements and reconstructed abundance images of real data in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the problems involved in the technical solutions of the present invention will be further described with reference to the accompanying drawings.
A hyperspectral image unmixing method based on a depth denoising self-coding network comprises the following steps:
(1) Adding a regular term into a network objective function according to the physical characteristics of the abundance coefficient and the characteristics of a constraint and a non-negative constraint on the basis of the denoising self-encoder, so as to form a depth denoising self-encoding network;
(2) Inputting a group of training image data into a depth denoising self-coding network, training the depth denoising self-coding network to obtain optimized network parameters, and obtaining a hyperspectral unmixed network model;
(3) Inputting the existing hyperspectral image data to be processed into a depth denoising self-coding network, extracting the characteristics of the hyperspectral image data through the coding process, and obtaining reconstructed original image data through decoding.
Further, in step (1), the depth denoising self-coding network includes an input layer, an implicit layer, and an output layer; the method comprises the steps of taking an encoding process from an input layer to an implicit layer, wherein the encoding process is to perform feature extraction on input hyperspectral image data to obtain a preliminary abundance coefficient, and taking a decoding process from the implicit layer to an output layer, wherein the decoding process is to decode the obtained preliminary abundance coefficient to obtain reconstructed original image data, namely;
adding a regular term to the network objective function to reduce redundant end members of the encoder, and introducing joint sparsity between adjacent pixels; selecting ReLu as an activation function can extract nonlinear characteristics of data and meet the non-negativity of an abundance coefficient; the objective function expression of the depth denoising self-coding network is as follows:
where X represents the input hyperspectral data, W represents the encoder weights, σ (X) represents the hidden layer activation function,and->Representing reconstructed image data +.>And an augmentation matrix of decoder weights a:
representing a squaring operation taking the F-norm, I.I 2,1 Represents a 2-norm addition operation taking each row vector in the matrix, λ represents a lagrangian coefficient, and the value of λ is set to 10e-6.
Further, in the step (2), training image data with the size of w×h×c is preprocessed to obtain w×h training sets with the size of 1×c, and then the training sets are input into a depth denoising self-coding network to perform multiple training so as to obtain optimized network parameters.
Furthermore, before the hyperspectral image data to be processed is input into the depth denoising self-coding network, the step (3) loads the depth denoising self-coding network model trained in the step (2), and then updates the network parameters into the network parameters trained in the step (2); then inputting hyperspectral image data, obtaining abundance coefficients through a network coding process, and obtaining an end member matrix through a decoding process.
Further, the training image data in the step (2) refers to hyperspectral image data as a training set, and the hyperspectral image to be processed in the step (3) refers to hyperspectral image data as a test set.
As in fig. 1-2, step S1: the weights of the hidden layer and the decoding layer are limited to be nonnegative on the basis of a traditional denoising self-encoder, the hidden layer is added and added to be a constraint, and the L21 constraint is added as a regular term to be an objective function, so that joint sparsity between adjacent pixels is utilized.
Step S11: the denoising self-encoder, recorded as DAE, can be regarded as a three-layer neural network, and comprises an input layer, an hidden layer and an output layer; the encoding process is from the input layer to the hidden layer, and the decoding process is from the hidden layer to the output layer.
Note that the encoder is f (x), the hidden layer activation function is σ (x), and the decoder is g (x), then the encoding process is expressed as:
S=f(X)=σ(WX)
where X represents the input hyperspectral data, W represents the encoder weights connecting the input layer and the hidden layer, and S is the hidden layer output (i.e., abundance ratio).
The decoding process is expressed as:
where a represents the decoder weights (i.e. end members) connecting the hidden layer and the output layer,representing the reconstructed data.
In the unmixing problem, decoder weight A and hidden layer S correspond to end member matrix and abundance coefficient respectively; the network learns the weights and hidden layer representations of the reconstructed data by minimizing the average reconstruction error:
where Σ represents the summation operation, i represents the number of columns of vectors in the hyperspectral data, the value range of i is an integer from 1 to n,representing a squaring operation taking a 2-norm; no bias is used in the network design because the bias tends to be a very negative value during training.
Step S12: the depth denoising self-encoder is recorded as DDAE, and the network model is shown in figure 1; for the unmixed problem, the sum of abundance is an important constraint on the reconstruction data in order to satisfy the constraintAnd the weight A is amplified by a constant, and the amplification matrices are respectively applied by +.>And->The representation is:
according to the decoding processAvailable->The column vectors of the abundance matrix satisfy a constraint.
In practical application, data often have noise, and the unmixed performance is drastically reduced due to the existence of noise and misestimation of the number of end members; to solve this problem, a regularization term W is introduced T || 2,1 To reduce redundant rows of the encoder, but this does not show joint sparsity between adjacent picture elements, it is improved to ||σ (WX) || 2,1 Thus, redundant end members are reduced, joint sparsity is introduced, and the performance of abundance estimation is improved; the objective function of W is defined as:
wherein the method comprises the steps ofRepresenting a squaring operation taking the F-norm, I.I 2,1 Representing a 2-norm addition operation of taking each row vector in the matrix, lambda represents a Lagrangian coefficient, and the value of lambda is set to 10e-6; according to the requirement of linear unmixingThe code function should be a linear mapping function, i.e. +.>Thus, to achieve an optimal solution, the coding function should also be linear; meanwhile, the hyperspectral unmixing requires that the abundance ratio satisfies non-negativity, so that the activation function also ensures that the hidden layer (abundance S) is non-negative; reLu is therefore chosen as the activation function, i.e., σ (x) =max (x, 0); when S is more than or equal to 0, the encoder is equivalent to a linear mapping function.
According to the unmixing requirement, the decoder weight also meets the non-negativity, namely A is more than or equal to 0; we solve this problem by the ReLu function, which guarantees the non-negativity of a during the optimization process.
Step S2: inputting a group of training image data into a depth denoising self-coding network;
inputting training data X, initializing encoder weight W and decoder weight A, and performing end member estimation by SMP algorithm to obtain A 0 And corresponding abundance ratio S 0 The method comprises the steps of carrying out a first treatment on the surface of the The encoder weights are initialized to w=s 0 X -1 Decoder weights are initialized to a=a 0 。
Step S3: extracting main characteristics of original data by encoding, and reconstructing the original data by decoding;
encoding the training data to obtain s=f (X) =σ (WX), S being the extracted feature; decoding S to obtain I.e. the reconstructed raw data.
Step S4: continuously training to obtain optimized network parameters, so that the reconstructed data is more similar to the original data;
setting the training times as 100 times and the training rate as 10e-4; the network parameters are continuously optimized during the training process by using an optimizer (RMSPropOptimizer) so that the obtained reconstructed data gradually approaches the original data.
Step S5: after training, inputting test data, obtaining S through a coding process, wherein the obtained S is the abundance coefficient of the hyperspectral image, the weight A of the decoder is the end member matrix, and finally outputting the obtained result S.
The effects of the present invention will be further described with reference to simulation experiments.
1. Simulation conditions:
simulation was performed on a system with CPU of Intel (R) Core (TM) i7-6700HQ 2.60GHz, 8GB memory, and Windows 10.
2. The simulation content:
in the simulation data experiment, a spectrum library splib06 of the United states geological survey (United States Geological Survey, USGS) is adopted, and the spectrum curves of 498 different substances under 224 spectrum bands are contained; simulation data 1 is generated from 9 end members randomly selected in a spectrum library, and comprises 100×100 pixels and 224 wave bands, and fig. 3 (a) - (i) respectively show original abundance images corresponding to the 9 end members; the added zero-mean Gaussian white noise SNR is 10, 20 and 30dB respectively.
Simulation data 2 is generated from 5 end members randomly selected in the spectral library based on a hyperspectral linear mixture model, with 75 x 75 pixels and 224 bands; FIG. 4 (a) shows simulated hyperspectral images, (b) - (f) show abundance images of five end members, respectively; the added zero-mean Gaussian white noise SNR is 10, 20 and 30dB respectively.
The data used in the real image is aviis cube data, the image subblock size is 250×191 pixels, wherein each pixel contains 188 spectral segments (the spectral segments with the atmospheric moisture absorption and low signal-to-noise ratio removed); FIG. 7 is a graph of the abundance profile generated by the Tricord 3.3 software and reconstructed abundance images of various techniques.
The invention is compared with SUnSAL, SUnSAL-TV and SMP in the prior art; in the experiment, the parameters of each technology are adjusted to be optimal, and the performance of the invention is fully verified.
The signal reconstruction error SRE is used as a measure of the unmixed precision of the analog data experiment, and the calculation formula is as follows:
wherein S represents the original abundance matrix, +.>Represents the reconstructed abundance matrix, expressed in dB as SRE (dB) =10log 10 (SRE)。
For the SRE index, a larger value indicates a smaller error between the abundance estimate and the true value of abundance, the better the abundance reconstruction performance.
The true image experiment adopts sparseness of abundance images and root mean square error of reconstructed images to evaluate the performance of an algorithm; sparsity is defined as: the number of non-zero values in the abundance matrix of the hyperspectral image; an abundance value greater than 0.001 is defined as a non-zero abundance to avoid calculating negligible values.
Root Mean Square Error (RMSE) is defined as follows:
table 1 results of the unmixed with different noise for analog data 1 and analog data 2.
As can be seen from table 1, in most cases, the inventive unmixed precision is highest among all techniques, and the unmixed precision is best at snr=10 dB, indicating that the inventive method has good denoising performance.
FIGS. 5 and 6 show abundance images obtained by unmixing analog data 1 and analog data 2, respectively, under 20dB Gaussian noise; as can be seen from fig. 5, the three comparison techniques have more noise points, but the noise points of the invention are fewer, the similarity with the original abundance image is higher, and the visual effect is better; although the SUnSAL-TV in FIG. 6 has almost no noise points, the image is too smooth and lacks part of the feature information; compared with SMP and SUnSAL, the invention has fewer noise points and better visual effect.
Table 2 abundance image sparsity and hyperspectral image reconstruction errors:
techniques for | SMP | SUnSAL | SUnSAL-TV | The invention is that |
Sparseness degree | 15.102 | 17.5629 | 20.472 | 10.0954 |
Reconstruction errors | 0.0034 | 0.0051 | 0.0038 | 0.0018 |
As can be seen from Table 2, the sparsity and reconstruction errors of the invention are both minimal and far superior to other algorithms, demonstrating that the invention has higher unmixed performance on real hyperspectral images.
It can be seen from fig. 7 that the reconstructed abundance image of the present invention has fewer noise points and retains the edge and feature information of the abundance image, more closely approaching the abundance profile generated by the software.
Claims (4)
1. The hyperspectral image unmixing method based on the depth denoising self-coding network is characterized by comprising the following steps of:
(1) Adding a regular term into a network objective function according to the physical characteristics of the abundance coefficient and the characteristics of a constraint and a non-negative constraint on the basis of the denoising self-encoder, so as to form a depth denoising self-encoding network;
in the step (1), the depth denoising self-coding network comprises an input layer, an implicit layer and an output layer; the method comprises the steps of taking an encoding process from an input layer to an implicit layer, wherein the encoding process is to perform feature extraction on input hyperspectral image data to obtain a preliminary abundance coefficient, and taking a decoding process from the implicit layer to an output layer, wherein the decoding process is to decode the obtained preliminary abundance coefficient to obtain reconstructed original image data, namely;
adding a regular term to the network objective function to reduce redundant end members of the encoder, and introducing joint sparsity between adjacent pixels; selecting ReLu as an activation function can extract nonlinear characteristics of data and meet the non-negativity of an abundance coefficient; the objective function expression of the depth denoising self-coding network is as follows:
where X represents the input hyperspectral data, W represents the encoder weights, σ (X) represents the hidden layer activation function,and->Representing reconstructed image data +.>And an augmentation matrix of decoder weights a:
representing a squaring operation taking the F-norm, I.I 2,1 Representing a 2-norm addition operation of taking each row vector in the matrix, lambda represents a Lagrangian coefficient, and the value of lambda is set to 10e-6;
(2) Inputting a group of training image data into a depth denoising self-coding network, training the depth denoising self-coding network to obtain optimized network parameters, and obtaining a hyperspectral unmixed network model;
(3) Inputting the existing hyperspectral image data to be processed into a depth denoising self-coding network, extracting the characteristics of the hyperspectral image data through the coding process, and obtaining reconstructed original image data through decoding.
2. The hyperspectral image unmixing method based on the depth denoising self-coding network according to claim 1, wherein: and (2) preprocessing training image data with the size of w multiplied by h multiplied by c to obtain w multiplied by h training sets with the size of 1 multiplied by c, inputting the training sets into a depth denoising self-coding network, and performing multiple training to obtain optimized network parameters.
3. The hyperspectral image unmixing method based on the depth denoising self-coding network according to claim 1, wherein: before hyperspectral image data to be processed are input into a depth denoising self-coding network, loading the depth denoising self-coding network model trained in the step (2); updating the network parameters into the trained network parameters in the step (2); then inputting hyperspectral image data, obtaining abundance coefficients through a network coding process, and obtaining an end member matrix through a decoding process.
4. The hyperspectral image unmixing method based on the depth denoising self-coding network according to claim 1, wherein: the training image data in the step (2) refers to hyperspectral image data as a training set, and the hyperspectral image to be processed in the step (3) refers to hyperspectral image data as a test set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911188298.2A CN111598786B (en) | 2019-11-28 | 2019-11-28 | Hyperspectral image unmixing method based on depth denoising self-coding network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911188298.2A CN111598786B (en) | 2019-11-28 | 2019-11-28 | Hyperspectral image unmixing method based on depth denoising self-coding network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111598786A CN111598786A (en) | 2020-08-28 |
CN111598786B true CN111598786B (en) | 2023-10-03 |
Family
ID=72183265
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911188298.2A Active CN111598786B (en) | 2019-11-28 | 2019-11-28 | Hyperspectral image unmixing method based on depth denoising self-coding network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111598786B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112699838B (en) * | 2021-01-13 | 2022-06-07 | 武汉大学 | Hyperspectral mixed pixel nonlinear blind decomposition method based on spectral diagnosis characteristic weighting |
CN113049530B (en) * | 2021-03-17 | 2022-07-26 | 北京工商大学 | Single-seed corn seed moisture content detection method based on near-infrared hyperspectrum |
CN113804657B (en) * | 2021-09-03 | 2022-11-29 | 中国科学院沈阳自动化研究所 | Sparse self-encoder spectral feature dimension reduction method based on multiple regression combination |
CN113486869B (en) * | 2021-09-07 | 2021-12-07 | 中国自然资源航空物探遥感中心 | Method, device and medium for lithology identification based on unsupervised feature extraction |
CN116091832B (en) * | 2023-02-16 | 2023-10-20 | 哈尔滨工业大学 | Tumor cell slice hyperspectral image classification method based on self-encoder network |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919864A (en) * | 2019-02-20 | 2019-06-21 | 重庆邮电大学 | A kind of compression of images cognitive method based on sparse denoising autoencoder network |
CN109978162A (en) * | 2017-12-28 | 2019-07-05 | 核工业北京地质研究院 | A kind of mineral content spectra inversion method based on deep neural network |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10902577B2 (en) * | 2017-06-19 | 2021-01-26 | Apeel Technology, Inc. | System and method for hyperspectral image processing to identify object |
US10861143B2 (en) * | 2017-09-27 | 2020-12-08 | Korea Advanced Institute Of Science And Technology | Method and apparatus for reconstructing hyperspectral image using artificial intelligence |
-
2019
- 2019-11-28 CN CN201911188298.2A patent/CN111598786B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109978162A (en) * | 2017-12-28 | 2019-07-05 | 核工业北京地质研究院 | A kind of mineral content spectra inversion method based on deep neural network |
CN109919864A (en) * | 2019-02-20 | 2019-06-21 | 重庆邮电大学 | A kind of compression of images cognitive method based on sparse denoising autoencoder network |
Non-Patent Citations (1)
Title |
---|
邢会欣.基于非负自编码器及非负矩阵分解的高光谱解混.《CNKI优秀硕士论文全文库》.2017,全文. * |
Also Published As
Publication number | Publication date |
---|---|
CN111598786A (en) | 2020-08-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111598786B (en) | Hyperspectral image unmixing method based on depth denoising self-coding network | |
CN109035142B (en) | Satellite image super-resolution method combining countermeasure network with aerial image prior | |
CN105163121B (en) | Big compression ratio satellite remote sensing images compression method based on depth autoencoder network | |
CN109671029B (en) | Image denoising method based on gamma norm minimization | |
CN113177882B (en) | Single-frame image super-resolution processing method based on diffusion model | |
CN103810755B (en) | Compressed sensing spectrum picture method for reconstructing based on documents structured Cluster rarefaction representation | |
CN104199627B (en) | Gradable video encoding system based on multiple dimensioned online dictionary learning | |
CN102542542A (en) | Image denoising method based on non-local sparse model | |
CN107154064B (en) | Natural image compressed sensing method for reconstructing based on depth sparse coding | |
CN104463223B (en) | Hyperspectral image group sparse unmixing method based on space spectrum information abundance constraint | |
CN106097278A (en) | The sparse model of a kind of multidimensional signal, method for reconstructing and dictionary training method | |
CN104506752B (en) | A kind of similar image compression method based on residual error compressed sensing | |
CN111915518B (en) | Hyperspectral image denoising method based on triple low-rank model | |
CN112967210A (en) | Unmanned aerial vehicle image denoising method based on full convolution twin network | |
CN107292316A (en) | A kind of method of the improving image definition based on rarefaction representation | |
CN111754430B (en) | Color image denoising method based on pure quaternion dictionary learning | |
CN105184742B (en) | A kind of image de-noising method of the sparse coding based on Laplce's figure characteristic vector | |
CN106447632A (en) | RAW image denoising method based on sparse representation | |
CN105590296B (en) | A kind of single-frame images Super-Resolution method based on doubledictionary study | |
CN107292855B (en) | Image denoising method combining self-adaptive non-local sample and low rank | |
Zou et al. | A nonlocal low-rank regularization method for fractal image coding | |
CN113436101B (en) | Method for removing rain by Dragon lattice tower module based on efficient channel attention mechanism | |
CN117236201B (en) | Diffusion and ViT-based downscaling method | |
Wen et al. | The power of complementary regularizers: Image recovery via transform learning and low-rank modeling | |
CN105846826B (en) | Compressed sensing signal reconfiguring method based on approximate smooth L0 norm |
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 |