CN113269693A - Power equipment hyperspectral image tag noise reduction method based on spectral space diagram - Google Patents
Power equipment hyperspectral image tag noise reduction method based on spectral space diagram Download PDFInfo
- Publication number
- CN113269693A CN113269693A CN202110654101.0A CN202110654101A CN113269693A CN 113269693 A CN113269693 A CN 113269693A CN 202110654101 A CN202110654101 A CN 202110654101A CN 113269693 A CN113269693 A CN 113269693A
- Authority
- CN
- China
- Prior art keywords
- value
- power equipment
- training
- lof
- training sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000010586 diagram Methods 0.000 title claims abstract description 19
- 230000003595 spectral effect Effects 0.000 title claims description 22
- 238000012549 training Methods 0.000 claims abstract description 95
- 238000001228 spectrum Methods 0.000 claims abstract description 11
- 230000011218 segmentation Effects 0.000 claims description 13
- 239000000126 substance Substances 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000000717 retained effect Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 16
- 238000005259 measurement Methods 0.000 abstract description 4
- 238000004458 analytical method Methods 0.000 abstract description 3
- 238000012706 support-vector machine Methods 0.000 description 12
- 238000002474 experimental method Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000011121 hardwood Substances 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003331 infrared imaging Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G06T5/70—
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of 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
- G06T2207/10048—Infrared 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]
Abstract
The application belongs to the technical field of power equipment information analysis, and relates to a power equipment hyperspectral image tag noise reduction method based on a spectrum space diagram. The label noise removing method based on the spectrum space diagram is applied to the technical field of power equipment, and the situation of misjudgment caused by inaccurate data information is simply generated by using a general algorithm. The application provides an improved LOF algorithm based on an SALOF algorithm, and the improved LOF algorithm, namely the SALOF algorithm, is used as a noise anomaly detection method and applied to HSI noise label detection. The Euclidean distance of the LOF is replaced by the SAM, the SALOF algorithm can effectively detect and remove the error standard sample, the quality of the training set is improved, and SAM measurement with higher classification precision is obtained. The label applied to the hyperspectral image of the power equipment has good technical advantages of high efficiency, short time and high precision when being applied to noise reduction, and is convenient to popularize and apply in the industry.
Description
Technical Field
The application relates to the technical field of power equipment information analysis, in particular to a power equipment hyperspectral image tag noise reduction method based on a spectrum space diagram.
Background
In the long-time operation process of the power equipment, various fault problems are inevitably generated, and in order to accurately detect fault information in real time, data analysis by applying hyperspectrum in the power equipment becomes an important detection means.
In the hyperspectral image analysis process of the power equipment, a type of information is called label information, and the label information reflects the attribute characteristics of the equipment, so that the label information belongs to a type of important data information. However, the tag information is also a relatively scarce resource, and is very susceptible to human and non-human factors, and a great amount of tag noise is generated along with the acquisition of the tag information, so how to reduce the adverse effect of the noise becomes a hot point for research by professional engineers.
Noise is reduced, weakened or eliminated, and people can more accurately acquire tag information, so that attribute characteristics and state characteristics of the power equipment are more practically judged and recognized, good and stable operation of the power equipment is facilitated, and the development of the power industry is facilitated. In recent years, a label noise removing method based on a spectrum space diagram is applied to the technical field of power equipment, but when the method is actually applied to the power industry, the situation that data information is not accurate enough to cause misjudgment is generated by simply using a general algorithm. How to optimize the power equipment hyperspectral image label noise reduction method based on the spectrum space diagram to enable the method to reflect the actual attribute characteristics of the power equipment more accurately becomes a technical problem to be solved urgently.
Disclosure of Invention
The application provides a label denoising method for a hyperspectral image of power equipment based on a spectral space diagram, and aims to solve the technical problem that a label denoising method in power equipment detection is not optimized enough.
The technical scheme adopted by the application is as follows:
a power equipment hyperspectral image tag noise reduction method based on a spectrum space diagram comprises the following steps:
acquiring a training sample of an original training set, and calculating a KNN (K-nearest neighbor) spectral angle value of the training sample;
obtaining an reachable distance value of each training sample according to the Kth neural network and the Kth minimum value of each training sample, and then calculating a local reachable density value of each training sample according to the reachable distance value;
obtaining an LOF value of the training sample by using the local reachability density value and the KNN spectral angle value of the training sample;
and setting a preset segmentation threshold value to remove the noise value of the original training set according to the LOF value of each training sample corresponding to different categories, and obtaining the denoised label information.
Optionally, after the step of setting a preset segmentation threshold to remove the noise value of the original training set according to the LOF values of different classes corresponding to each training sample, and obtaining the denoised label information, the method further includes:
calculating the spectrum angle between each type of different training samples by the following specific process:
wherein the content of the first and second substances,refers to the spectral angle between the jth and nth training samples in the c-th class, j ≠ 1,2At the upper partUnder the constraint of (2), the stretching distance is:
wherein the content of the first and second substances,is thatThe spectral angle of the Kth NN, wherein K is the number of NNs;
then, training samples are obtained from the local reachable density valuesThe LOF of (1) is determined,
the segmentation threshold is set to be mu,is defined as the probability of an anomaly of the training sample,is obtained by the following formula:
improved training setAnd (5) utilizing an SVM classifier to test the classification result of the improved training set A'.
Optionally, SAM is used in KNN of LOF algorithm instead of Ed.
Optionally, the parameter K is selected from the interval {1,2,3,4,5,6,7,8,9,10} and the parameter μ is selected from the interval {1.0,1.2,1.4, …,3.0 }.
Optionally, the default parameters are set as: k is 4, and the preset segmentation threshold μ is 2.0.
The technical scheme of the application has the following beneficial effects:
the utility model provides a spectral space diagram-based power equipment hyperspectral image label noise reduction method, provides an improved algorithm based on LOF algorithm, and applies the improved LOF algorithm, namely SALOF algorithm, as a noise anomaly detection method in HSI noise label detection. The Euclidean distance of the LOF is replaced by the SAM, the SALOF algorithm can effectively detect and remove the error standard sample, the quality of the training set is improved, and SAM measurement with higher classification precision is obtained. The label applied to the hyperspectral image of the power equipment has good technical advantages of high efficiency, short time and high precision when being applied to noise reduction, and is convenient to popularize and apply in the industry.
Drawings
In order to more clearly explain the technical solution of the present application, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious to those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a block flow diagram of one embodiment of the present application;
FIG. 2 is a schematic diagram of the local density between one training sample and all training samples of class i in the embodiment of the present application;
fig. 3 is a schematic view of LOF values of i-th different training samples in this embodiment.
Detailed Description
Reference will now be made in detail to embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following examples do not represent all embodiments consistent with the present application. But merely as exemplifications of systems and methods consistent with certain aspects of the application, as recited in the claims.
Referring to fig. 1, a flow chart of an embodiment of the present application is shown to facilitate understanding of technical solutions of the following embodiments.
The application provides a power equipment hyperspectral image tag noise reduction method based on a spectrum space diagram, which comprises the following steps:
acquiring a training sample of an original training set, and calculating a KNN (K-nearest neighbor) spectral angle value of the training sample;
obtaining an reachable distance value of each training sample according to the Kth neural network and the Kth minimum value of each training sample, and then calculating a local reachable density value of each training sample according to the reachable distance value;
obtaining an LOF value of the training sample by using the local reachability density value and the KNN spectral angle value of the training sample;
and setting a preset segmentation threshold value to remove the noise value of the original training set according to the LOF value of each training sample corresponding to different categories, and obtaining the denoised label information.
Optionally, after the step of setting a preset segmentation threshold to remove the noise value of the original training set according to the LOF values of different classes corresponding to each training sample, and obtaining the denoised label information, the method further includes:
calculating the spectrum angle between each type of different training samples by the following specific process:
wherein the content of the first and second substances,refers to the spectral angle between the jth and nth training samples in the c-th class, j ≠ 1,2At the upper partUnder the constraint of (2), the stretching distance is:
wherein the content of the first and second substances,is thatThe spectral angle of the Kth NN, wherein K is the number of NNs;
then theObtaining training samples from the local achievable density valuesThe LOF of (1) is determined,
the segmentation threshold is set to be mu,is defined as the probability of an anomaly of the training sample,is obtained by the following formula:
improved training setAnd (5) utilizing an SVM classifier to test the classification result of the improved training set A'.
In the anomaly detection, the LOF algorithm is based on the concepts of the reachable distance and the local reachable density, the locality is given by kNNs, and the distance of the LOF is used for estimating the density obtained by the LOF. By comparing the local reachability density of the target training sample with the proximity values of the target training sample, regions of similar density can be identified, and the training samples are much less dense than their neighbors. These samples are considered outliers. The local density is estimated from the typical distance that a training sample can "arrive" from a neighboring region. As shown in fig. 2, the kth NN of the i-th class of training samples may be obtained from euclidean distance. The reachable distance and the reachable density of the training samples are then calculated, respectively. Finally, the abnormal probability of each type of different training samples is determined. Typically, the reachable distance and the local reachable density in the LOF are additional measures to produce more stable results in the detection of misregistered training samples. Fig. 3 illustrates the LOF values for different training samples.
Optionally, SAM is used in KNN of LOF algorithm instead of Ed.
Optionally, the parameter K is selected from the interval {1,2,3,4,5,6,7,8,9,10} and the parameter μ is selected from the interval {1.0,1.2,1.4, …,3.0 }.
Optionally, the default parameters are set as: k is 4, and the preset segmentation threshold μ is 2.0.
In order to verify the technical effect of the application, relevant experiments are carried out, and the specific experiment results are as follows:
the saluf algorithm proposed in this application was evaluated using a KSC dataset obtained from an on-board visible infrared imaging spectrometer of KSC, florida, usa. The image has 224 bands (every 0.01 μm wide), a size of 512 × 614, a spatial resolution of 18 m/pixel, and a center wavelength of 0.4 to 2.5 μm, of which 48 bands are removed as water absorption and low signal-to-noise ratio bands. In the experiment, noise signatures were detected and removed.
Firstly, the generation method of the noise training set is as follows:
1) randomly extracting some training samples from one class, and randomly extracting different numbers of training samples from other classes (the other classes are some classes randomly extracted from the other classes);
2) the labels of the above samples correspond to a class of labels.
Second, a set of objective indices such as Overall Accuracy (OA), Average Accuracy (AA), and Kappa coefficient (Kappa) are used. Of these three indices, OA measures the percentage of correctly classified pixels, AA is the average of the percentage of correctly classified pixels for each class, and in order to make the measurement more objective, Kappa estimates the percentage of correctly classified pixels, which are corrected by the amount of protocol expected purely by chance.
Third, training samples were randomly drawn and the experiment was repeated 50 times to obtain mean and standard deviation of OA, AA, and Kappa.
Fourthly, in the classifier of the support vector machine, the influence of the training set with different numbers of the sample with error samples and the noise training set corrected by the SALOF method on the HSI classification is utilized.
Here we first analyzed the effect of different metrics in kNN on the detection of noise labels in table 1, such as Spectral Information Divergence (SID), euclidean distance (Ed), orthogonal subspace projection divergence (OPD), and SAM. In table 1, SVMs were trained with 20 true samples and 6 mis-labeled samples. As can be seen from table 1, table 1 analyzes the influence of different metrics in kNN on noise mark detection, and on the KSC data set, the SALOF method proposed in the present application can obtain SAM metrics with the highest classification accuracy. Therefore, Ed is replaced by SAM in KNN of LOF algorithm. The impact of K and the segmentation threshold μ in KNNs on the SALOF algorithm performance was then analyzed. The number of noise labels is set to 3 for the training set of each class in the KSC data set. The reported accuracy of this experiment is also the average result of a randomly selected noisy training set run. The parameter K is selected from the interval {1,2,3,4,5,6,7,8,9,10} and the parameter μ is selected from the interval {1.0,1.2,1.4, …,3.0}, and when K increases, the performance of the SVM classifier changes.
TABLE 1 analysis of the impact of different metrics on noise signature detection in kNN
The SALOF algorithm of the application can effectively detect and remove the noise label of the noise training set. With increasing μ, the classification accuracy shows a tendency to increase first and then decrease. Since different classes of HSI pixels may be composed of similar materials, the obtained LOF values do not differ significantly in local reachability density. Therefore, K-4 and μ -2.0 are used as default parameter settings that are recommended.
The present application also analyzes the effect of different numbers of false mark samples on the SVM classifier to be examined on the KSC image. The parameters of all training sample sets of different numbers of mislabeled samples are consistent with the experimental default parameters. The classification map is obtained by performing the SVM on a noisy training set consisting of 20 real training samples and a different number of false labeled samples. The classification map is obtained by the saluf algorithm of the present application. Experiments have shown that classification maps obtained with a support vector machine that wrongly labels samples always show more severe labeling errors in hardwood swamps and muds than classification maps obtained with the proposed saluf method. Therefore, according to the experimental result, when the training sample of each class is mixed with the error standard samples of different numbers of other classes, the support vector machine trains the error model, and the classification precision is reduced. Experimental results show that the SALOF algorithm can effectively detect and remove the error standard samples, and the quality of the training set is improved. Table 2 shows the classification accuracy of the KSC data set. The classification accuracy obtained by the SVM and the proposed method is trained by using 20 real samples and different numbers of off-set samples. The method records two time values: detection time and support vector machine time. Table 2 shows the SVM classification accuracy for different error labeled samples. Under three groups of experimental indexes (OA, AA and Kappa), the classification precision of the method is superior to that of the SVM trained by the original error marked pixels. This means that the method can effectively correct the noisy training set. The bottom of table 2 shows that the support vector machine algorithm consumes a lot of computation time compared to the proposed method, which indicates that the proposed method can efficiently detect certain noisy labels. However, because the training samples contain a small number of noise labels, the support vector machine with real training samples consumes less time than the SALOF method.
TABLE 2 Classification accuracy of KSC datasets
The above analyzes the influence of different noise label numbers on the SVMs trained by the original training set and the SVMs trained by the improved SALOFs training set. When the number of noise labels increases from 3 to 12, the classification performance of SVMs trained using the original training set always achieves a lower OA than SVMs trained using the improved training set. In particular, the saluf algorithm improves the detection performance for noisy labels when there are 5 noisy labels. The support vector machine was trained with the original training set and the improved training set with performance of 84.09% and 86.05%, respectively. Generally, the SALOF detection method for noise labels proposed in the present application is effective.
The utility model provides a spectral space diagram-based power equipment hyperspectral image label noise reduction method, provides an improved algorithm based on LOF algorithm, and applies the improved LOF algorithm, namely SALOF algorithm, as a noise anomaly detection method in HSI noise label detection. The Euclidean distance of the LOF is replaced by the SAM, the SALOF algorithm can effectively detect and remove the error standard sample, the quality of the training set is improved, and SAM measurement with higher classification precision is obtained. The label applied to the hyperspectral image of the power equipment has good technical advantages of high efficiency, short time and high precision when being applied to noise reduction, and is convenient to popularize and apply in the industry.
The embodiments provided in the present application are only a few examples of the general concept of the present application, and do not limit the scope of the present application. Any other embodiments extended according to the scheme of the present application without inventive efforts will be within the scope of protection of the present application for a person skilled in the art.
Claims (5)
1. A power equipment hyperspectral image tag noise reduction method based on a spectrum space diagram is characterized by comprising the following steps:
acquiring a training sample of an original training set, and calculating a KNN (K-nearest neighbor) spectral angle value of the training sample;
obtaining an reachable distance value of each training sample according to the Kth neural network and the Kth minimum value of each training sample, and then calculating a local reachable density value of each training sample according to the reachable distance value;
obtaining an LOF value of the training sample by using the local reachability density value and the KNN spectral angle value of the training sample;
and setting a preset segmentation threshold value to remove the noise value of the original training set according to the LOF value of each training sample corresponding to different categories, and obtaining the denoised label information.
2. The method for denoising the hyperspectral image label of the electrical equipment based on the spectral space diagram according to claim 1, wherein after the step of setting a preset segmentation threshold value to remove the noise value of the original training set according to the LOF value of each training sample corresponding to different categories and obtaining denoised label information, the method further comprises:
calculating the spectrum angle between each type of different training samples by the following specific process:
wherein the content of the first and second substances,refers to the spectral angle between the jth and nth training samples in the c-th class, j ≠ 1,2At the upper partUnder the constraint of (2), the stretching distance is:
wherein the content of the first and second substances,is thatThe spectral angle of the Kth NN, wherein K is the number of NNs;
then, training samples are obtained from the local reachable density valuesThe LOF of (1) is determined,
the segmentation threshold is set to be mu,is defined as the probability of an anomaly of the training sample,is obtained by the following formula:
3. The electric power equipment hyperspectral image label noise reduction method based on the spectral space diagram of claim 1, characterized in that SAM is adopted in KNN of LOF algorithm to replace Ed.
4. The spectral space diagram-based power equipment hyperspectral image label noise reduction method according to claim 1 or 2, wherein the parameter K is selected from {1,2,3,4,5,6,7,8,9,10} interval, and the parameter μ is selected from {1.0,1.2,1.4, …,3.0} interval.
5. The electric power equipment hyperspectral image tag noise reduction method based on the spectral space diagram according to claim 1, characterized in that default parameters are set as: k is 4, and the preset segmentation threshold μ is 2.0.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110654101.0A CN113269693A (en) | 2021-06-11 | 2021-06-11 | Power equipment hyperspectral image tag noise reduction method based on spectral space diagram |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110654101.0A CN113269693A (en) | 2021-06-11 | 2021-06-11 | Power equipment hyperspectral image tag noise reduction method based on spectral space diagram |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113269693A true CN113269693A (en) | 2021-08-17 |
Family
ID=77234883
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110654101.0A Pending CN113269693A (en) | 2021-06-11 | 2021-06-11 | Power equipment hyperspectral image tag noise reduction method based on spectral space diagram |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113269693A (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596244A (en) * | 2018-04-20 | 2018-09-28 | 湖南理工学院 | A kind of high spectrum image label noise detecting method based on spectrum angle density peaks |
CN110046639A (en) * | 2019-01-10 | 2019-07-23 | 湖南理工学院 | A kind of Hyperspectral imaging noise label detection method based on super-pixel weight density |
-
2021
- 2021-06-11 CN CN202110654101.0A patent/CN113269693A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596244A (en) * | 2018-04-20 | 2018-09-28 | 湖南理工学院 | A kind of high spectrum image label noise detecting method based on spectrum angle density peaks |
CN110046639A (en) * | 2019-01-10 | 2019-07-23 | 湖南理工学院 | A kind of Hyperspectral imaging noise label detection method based on super-pixel weight density |
Non-Patent Citations (1)
Title |
---|
BING TU, CHENGLE ZHOU, WENLAN KUANG, LONGYUAN GUO, XIAOFENG OU: "Hyperspectral Imagery Noise Label Detection by Spectral Angle Local Outlier Factor", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11586913B2 (en) | Power equipment fault detecting and positioning method of artificial intelligence inference fusion | |
CN110363182B (en) | Deep learning-based lane line detection method | |
US20190087638A1 (en) | Analyzing digital holographic microscopy data for hematology applications | |
Tu et al. | Hyperspectral imagery noisy label detection by spectral angle local outlier factor | |
Ye et al. | Classification based on 3-D DWT and decision fusion for hyperspectral image analysis | |
Jia et al. | A two-stage feature selection framework for hyperspectral image classification using few labeled samples | |
Cui et al. | Locality preserving genetic algorithms for spatial-spectral hyperspectral image classification | |
Hati et al. | Plant recognition from leaf image through artificial neural network | |
Yang et al. | A feature-metric-based affinity propagation technique for feature selection in hyperspectral image classification | |
US20120328197A1 (en) | Identifying matching images | |
CN109409389B (en) | Multi-feature-fused object-oriented change detection method | |
CN110705722A (en) | Diagnostic model for industrial equipment fault diagnosis and construction method and application thereof | |
CN106778680A (en) | A kind of hyperspectral image band selection method and device extracted based on critical bands | |
CN111242050A (en) | Automatic change detection method for remote sensing image in large-scale complex scene | |
CN103310200A (en) | Face recognition method | |
CN111639587A (en) | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network | |
CN111639697B (en) | Hyperspectral image classification method based on non-repeated sampling and prototype network | |
Ma et al. | Multiscale 2-D singular spectrum analysis and principal component analysis for spatial–spectral noise-robust feature extraction and classification of hyperspectral images | |
Cao et al. | Does normalization methods play a role for hyperspectral image classification? | |
Zhang et al. | An automatic recognition method for PCB visual defects | |
CN110020674B (en) | Cross-domain self-adaptive image classification method for improving local category discrimination | |
CN113111969B (en) | Hyperspectral image classification method based on mixed measurement | |
Siméoni et al. | Unsupervised object discovery for instance recognition | |
Shambulinga et al. | Supervised hyperspectral image classification using SVM and linear discriminant analysis | |
CN114863291B (en) | Hyperspectral image band selection method based on MCL and spectrum difference measurement |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210817 |