CN112967241B - Hyperspectral image anomaly detection method based on local gradient guidance - Google Patents
Hyperspectral image anomaly detection method based on local gradient guidance Download PDFInfo
- Publication number
- CN112967241B CN112967241B CN202110216275.9A CN202110216275A CN112967241B CN 112967241 B CN112967241 B CN 112967241B CN 202110216275 A CN202110216275 A CN 202110216275A CN 112967241 B CN112967241 B CN 112967241B
- Authority
- CN
- China
- Prior art keywords
- gradient
- hyperspectral image
- image
- hyperspectral
- gradient profile
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30212—Military
-
- 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)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a hyperspectral image anomaly detection method based on local gradient guidance, which comprises the steps of firstly, positioning possible anomaly pixels of an input original hyperspectral remote sensing image; then according to the gradient contours of the possibly abnormal pixels obtained by rough selection, converting the local gradient contours and guiding and optimizing the original hyperspectral image to obtain the hyperspectral image with enhanced spatial structure information; and finally, performing anomaly detection on the enhanced hyperspectral image. The method and the device improve the accuracy of anomaly detection of the hyperspectral remote sensing image better.
Description
Technical Field
The invention belongs to the technical field of hyperspectral remote sensing image processing, and particularly relates to a hyperspectral image anomaly detection method based on local gradient guidance.
Background
The hyperspectral imager can obtain an image data cube, i.e. a three-dimensional hyperspectral image, with two-dimensional spatial information and one-dimensional spectral information. The hyperspectral image has the characteristics of high spectral resolution, integrated atlas, low spatial resolution and the like. The spectrum information can be used for inverting the component information of the substance, so that the hyperspectral image plays an important role in the problems of military reconnaissance, disaster early warning and the like. In these applications, anomaly detection does not require any prior information about the object, and directly detects an anomaly object in the scene, thus having strong practicability in various applications.
Anomaly detection refers to finding anomalous pixels of an unknown spectral signal in a hyperspectral image. Outliers generally refer to pixel locations where spectral and spatial features are significantly different from the surrounding environment. Specifically, first, their spectra are very different from the surrounding spectra. Meanwhile, the abnormal pixels are usually embedded in a local homogeneous background in the form of a plurality of pixels, i.e., are represented as spatial differences. Because of the lower spatial resolution of hyperspectral images, the spatial structure information inside the images tends to be smoother, which is different from the real scene. Considering that the number of abnormal pixels in a hyperspectral image is often only a small part, and the differences in spectrum and space are shown. Thus, in case of false alarms allowing a certain procedure, it is not difficult to completely detect abnormal pixels. The spatial structure information can be enhanced by carrying out gradient contour transformation on the possibly abnormal pixel points with certain false alarms. Due to the introduction of the space structure information, the abnormal detection precision of the enhanced hyperspectral image can be improved to a certain extent.
Disclosure of Invention
The invention aims to provide a hyperspectral image anomaly detection method based on local gradient guidance, which better improves the accuracy of hyperspectral remote sensing image anomaly detection.
The technical scheme adopted by the invention is that the hyperspectral image anomaly detection method based on local gradient guidance is implemented according to the following steps:
step 1, positioning possible abnormal pixel points of an input original hyperspectral remote sensing image;
step 2, according to the gradient contours of the possibly abnormal pixels obtained by rough selection in the step 1, converting the local gradient contours and guiding to optimize the original hyperspectral image to obtain the hyperspectral image with enhanced spatial structure information;
and 3, performing anomaly detection on the enhanced hyperspectral image.
The present invention is also characterized in that,
the step 1 is specifically implemented according to the following steps:
step 1.1, detecting an original input image by using an existing RX anomaly detection method, and obtaining a response value of each pixel point after detection;
step 1.2, sorting the response values of all the pixel points obtained in the step 1.1 from strong to weak;
step 1.3, in the sorting from strong to weak, temporarily marking the spatial position of the response in the previous K as a possible abnormal pixel.
K in step 1.3 is traversed 0.01 to 0.04 at intervals of 0.01.
The step 2 is specifically implemented according to the following steps:
step 2.1, describing the spatial layout of the original image gradient by using the gradient profile of the temporarily marked abnormal pixels, describing the gradient profile by using generalized Gaussian distribution parameterization, measuring fitting errors by using KL divergence and distance, and obtaining an optimal result by minimizing the average fitting errors on a gradient profile training set;
step 2.2, converting the observed gradient profile with parameters into an ideal gradient profile with parameters by multiplying the gradient profile of the temporary marked possible abnormal points by the transformation ratio between the ideal gradient profile and the actual gradient profile;
and 2.3, after the gradient profile of the ideal hyperspectral image is obtained, carrying out one-step iterative solution through a gradient descent method to obtain a minimized loss function, thereby restraining the gradient profile to be close to the transformed gradient profile, recovering the geometric structure of the hyperspectral image and achieving the purpose of enhancing the space information of the hyperspectral image.
In step 2.2, the shape parameter of the gradient profile of the original image is 1.6, and the shape parameter in the ideal hyperspectral image is 1.63.
The step 3 is specifically as follows:
and (3) detecting the hyperspectral image obtained in the step (2) by using the existing RX anomaly detection method to obtain a final detection result graph.
The hyperspectral image anomaly detection method based on local gradient guidance has the advantages that the hyperspectral image anomaly detection method is different from the existing anomaly detection method in that the original hyperspectral image is not directly processed, spatial information of the image is enhanced firstly through guidance of local gradient contours, and then the enhanced image is detected. Hyperspectral images typically have poor spatial detail due to limitations of the image sensor. Thereby reducing the detection accuracy of the hyperspectral image. According to the method, firstly, some possible abnormal pixels are selected from an original hyperspectral image, gradient sections of the abnormal pixels are described through generalized Gaussian distribution parameterization, the geometric structure of the hyperspectral image is restored through gradient contour conversion, and a hyperspectral image with enhanced local spatial information is obtained. The enhanced image and the original image are respectively detected by the existing RX anomaly detection method, and the validity of the method is verified.
Drawings
FIG. 1 is a flow chart of a hyperspectral image anomaly target detection method based on local gradient guidance in the present invention;
fig. 2 is a visual inspection of the present invention versus the comparison method for 4 hyperspectral images in the database of air.
Fig. 3 is a graph comparing the ROC curve of the detection effect of the present invention with that of the comparison method for 4 hyperspectral images in the database of air, and it is apparent that the proposed method always achieves the best detection performance.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
The invention discloses a hyperspectral image anomaly detection method based on local gradient guidance, which is implemented by a flow chart shown in figure 1 specifically according to the following steps:
step 1, positioning possible abnormal pixel points of an input original hyperspectral remote sensing image;
the step 1 is specifically implemented according to the following steps:
step 1.1, detecting an original input image by using an existing Reed Xiaoli (RX) anomaly detection method, and obtaining a response value of each pixel point after detection;
step 1.2, sorting the response values of all the pixel points obtained in the step 1.1 from strong to weak;
step 1.3, in the sorting from strong to weak, temporarily marking the spatial position of the response in the previous K as a possible abnormal pixel.
K in step 1.3 is traversed 0.01 to 0.04 at intervals of 0.01.
Step 2, according to the gradient contours of the possibly abnormal pixels obtained by rough selection in the step 1, converting the local gradient contours and guiding to optimize the original hyperspectral image to obtain the hyperspectral image with enhanced spatial structure information;
the step 2 is specifically implemented according to the following steps:
and 2.1, describing the spatial layout of the gradient of the original image by using the gradient profile of the temporarily marked abnormal pixels, and describing the gradient profile by using generalized Gaussian distribution parameterization, wherein a generalized Gaussian distribution function is used for fitting a gradient profile curve in the natural image. Using Kullback-Leibler (KL) divergence and distance measurement fitting error, obtaining an optimal result by minimizing the average fitting error on the gradient profile training set;
step 2.2, converting the observed gradient profile with parameters into an ideal gradient profile with parameters by multiplying the gradient profile of the temporary marked possible abnormal points by the transformation ratio between the ideal gradient profile and the actual gradient profile;
and 2.3, after the gradient profile of the ideal hyperspectral image is obtained, carrying out one-step iterative solution through a gradient descent method to obtain a minimized loss function, thereby restraining the gradient profile to be close to the transformed gradient profile, recovering the geometric structure of the hyperspectral image and achieving the purpose of enhancing the space information of the hyperspectral image.
In step 2.2, the shape parameter of the gradient profile of the original image is 1.6, and the shape parameter in the ideal hyperspectral image is 1.63.
And 3, performing anomaly detection on the enhanced hyperspectral image.
The step 3 is specifically as follows:
and (3) detecting the hyperspectral image obtained in the step (2) by using the existing RX anomaly detection method, namely performing a second-pass RX anomaly detection method to obtain a final detection result diagram.
In order to verify the effectiveness of the invention in anomaly detection of hyperspectral images, a comparison experiment is given through a simulation experiment. The experimental platform adopts MATLAB (R2018 a) on Windows with a processor of Intel core i5@2.30GHZ and a memory of 8.0 GB. The invention adopts RXD method as a detector to respectively detect the abnormality of the original image and the enhanced image, and obtains corresponding performance contrast to verify the effectiveness of the concept of local gradient guidance in the invention.
The experiment used a true hyperspectral image in the 4 airport-beach-city (ABU) dataset to verify the validity of the proposed method, airport-1, airport-2, airport-3 and airport-4, respectively, for the airport subset. These four data are all acquired by the airbone Visible/Infrared Imaging Spectrometer (aviis) sensor. Fig. 2 is a visual inspection diagram of 4 images by the RX method of the present invention, wherein the first column is a gray scale image of the 100 th band, the second column is a inspection reference diagram of 4 images, the third column is a inspection result diagram of the RXD method, and the fourth column is an inspection result diagram of the proposed invention. By looking at the visual inspection diagram in fig. 2, it can be found that the LGG-RX method has a higher detection rate for abnormal pixels.
In order to quantitatively evaluate the merits of the algorithm, the objective rating indexes adopted are: receiver operating characteristics ROC (Receiver Operating Characteristic) curve and AUC (Area Under Curve) value. The ROC curve establishes a one-to-one correspondence between the false alarm rate FP and the detection rate TP when different thresholds are taken in the detection process. The ROC curve takes the false alarm rate FP and the detection rate TP as the vertical axis and the horizontal axis respectively, the area of the area below the curve is an AUC value, and the larger the AUC value is, the better the detection performance of the algorithm is. Fig. 3 shows ROC curves for LGG-RX and comparative method RXD presented herein, with table 1 listing AUC statistics for different detectors. Clearly, the proposed method always achieves the best detection performance.
(1) Hyperspectral remote sensing image anomaly detection experiment:
table 1 is the AUC value obtained by detecting the classical RXD detection method and the hyperspectral image anomaly target detection algorithm based on local gradient guidance provided by the invention for 4 images in the hyperspectral remote sensing image dataset, which is an air point. LGG-RX means that the invention applies RXD method in gradient guidance.
As can be seen from the experimental results in Table 1, compared with the conventional anomaly detection method, the method provided by the invention has the advantages that some possible anomaly pixels are selected from the original hyperspectral image, the gradient profile of the anomaly pixels is described through generalized Gaussian distribution parameterization, the geometric structure of the hyperspectral image is recovered through gradient field conversion, and the anomaly detection performance is further improved.
Table 1 RXD detects the original image and the enhanced image, respectively, and the obtained corresponding AUC values
Claims (4)
1. The hyperspectral image anomaly detection method based on local gradient guidance is characterized by comprising the following steps of:
step 1, roughly selecting and positioning possible abnormal pixel points of an input original hyperspectral remote sensing image;
the step 1 is specifically implemented according to the following steps:
step 1.1, detecting an original input image by using an existing RX anomaly detection method, and obtaining a response value of each pixel point after detection;
step 1.2, sorting the response values of all the pixel points obtained in the step 1.1 from strong to weak;
step 1.3, temporarily marking the spatial position of the response in the previous K as a possible abnormal pixel in the sequence from strong to weak;
step 2, according to the gradient contours of the possibly abnormal pixels obtained by rough selection in the step 1, converting the local gradient contours and guiding to optimize the original hyperspectral image to obtain the hyperspectral image with enhanced spatial structure information;
the step 2 is specifically implemented according to the following steps:
step 2.1, describing the spatial layout of the original image gradient by using the gradient profile of the temporarily marked abnormal pixels, describing the gradient profile by using generalized Gaussian distribution parameterization, measuring fitting errors by using KL divergence and distance, and obtaining an optimal result by minimizing the average fitting errors on a gradient profile training set;
step 2.2, converting the observed gradient profile with parameters into an ideal gradient profile with parameters by multiplying the gradient profile of the temporary marked possible abnormal points by the transformation ratio between the ideal gradient profile and the actual gradient profile;
step 2.3, after obtaining the gradient profile of the ideal hyperspectral image, carrying out one-step iterative solution through a gradient descent method to obtain a minimized loss function, thereby restraining the gradient profile to be close to the transformed gradient profile, recovering the geometric structure of the hyperspectral image and achieving the purpose of enhancing the space information of the hyperspectral image;
and 3, performing anomaly detection on the enhanced hyperspectral image.
2. The method for detecting hyperspectral image anomaly based on local gradient guidance according to claim 1, wherein K in step 1.3 traverses 0.01 to 0.04 at intervals of 0.01.
3. The method for detecting hyperspectral image anomalies based on local gradient guidance according to claim 1, wherein in the step 2.2, the shape parameter of the gradient profile of the original image is 1.6, and the shape parameter in the ideal hyperspectral image is 1.63.
4. The hyperspectral image anomaly detection method based on local gradient guidance according to claim 1, wherein the step 3 is specifically as follows:
and (3) detecting the hyperspectral image obtained in the step (2) by using the existing RX anomaly detection method to obtain a final detection result graph.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110216275.9A CN112967241B (en) | 2021-02-26 | 2021-02-26 | Hyperspectral image anomaly detection method based on local gradient guidance |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110216275.9A CN112967241B (en) | 2021-02-26 | 2021-02-26 | Hyperspectral image anomaly detection method based on local gradient guidance |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112967241A CN112967241A (en) | 2021-06-15 |
CN112967241B true CN112967241B (en) | 2023-09-12 |
Family
ID=76276829
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110216275.9A Active CN112967241B (en) | 2021-02-26 | 2021-02-26 | Hyperspectral image anomaly detection method based on local gradient guidance |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112967241B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113657199B (en) * | 2021-07-28 | 2023-09-15 | 西安理工大学 | Hyperspectral image anomaly detection method based on space-spectrum extraction |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011051382A1 (en) * | 2009-10-29 | 2011-05-05 | Galderma Research & Development | Method and device for analysing hyper-spectral images |
CN105139412A (en) * | 2015-09-25 | 2015-12-09 | 深圳大学 | Hyperspectral image corner detection method and system |
CN108765359A (en) * | 2018-05-31 | 2018-11-06 | 安徽大学 | A kind of fusion method of target in hyperspectral remotely sensed image and full-colour image based on JSKF models and NSCT technologies |
CN111242910A (en) * | 2020-01-07 | 2020-06-05 | 西北工业大学 | Hyperspectral remote sensing image coarse-to-fine anomaly detection method based on tensor decomposition |
CN111986162A (en) * | 2020-07-28 | 2020-11-24 | 西安理工大学 | Hyperspectral abnormal point rapid detection method based on rough positioning and collaborative representation |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9940520B2 (en) * | 2015-05-01 | 2018-04-10 | Applied Research LLC. | Automatic target recognition system with online machine learning capability |
US10496883B2 (en) * | 2017-01-27 | 2019-12-03 | Signal Processing, Inc. | Method and system for enhancing predictive accuracy of planet surface characteristics from orbit |
-
2021
- 2021-02-26 CN CN202110216275.9A patent/CN112967241B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011051382A1 (en) * | 2009-10-29 | 2011-05-05 | Galderma Research & Development | Method and device for analysing hyper-spectral images |
CN105139412A (en) * | 2015-09-25 | 2015-12-09 | 深圳大学 | Hyperspectral image corner detection method and system |
CN108765359A (en) * | 2018-05-31 | 2018-11-06 | 安徽大学 | A kind of fusion method of target in hyperspectral remotely sensed image and full-colour image based on JSKF models and NSCT technologies |
CN111242910A (en) * | 2020-01-07 | 2020-06-05 | 西北工业大学 | Hyperspectral remote sensing image coarse-to-fine anomaly detection method based on tensor decomposition |
CN111986162A (en) * | 2020-07-28 | 2020-11-24 | 西安理工大学 | Hyperspectral abnormal point rapid detection method based on rough positioning and collaborative representation |
Non-Patent Citations (1)
Title |
---|
张炎 ; 华文深 ; 黄富瑜 ; 王强辉 ; 索文凯 ; .基于空谱联合异常度的高光谱异常目标检测.光谱学与光谱分析.2020,(第06期),全文. * |
Also Published As
Publication number | Publication date |
---|---|
CN112967241A (en) | 2021-06-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108389188B (en) | Sparse hyperspectral abnormal target detection method | |
CN111046800B (en) | Hyperspectral image abnormal target detection method based on low rank and sparse decomposition | |
US10032077B1 (en) | Vehicle track identification in synthetic aperture radar images | |
CN103716579A (en) | Video monitoring method and system | |
CN105913448B (en) | The high spectrum image object detection method of subspace is matched based on tensor | |
CN107704802B (en) | Performance prediction method for measurement of hyperspectral image clutter and anomaly detection algorithm | |
Ma et al. | A fast recursive collaboration representation anomaly detector for hyperspectral image | |
CN110991493B (en) | Hyperspectral anomaly detection method for collaborative representation and anomaly rejection | |
CN112967241B (en) | Hyperspectral image anomaly detection method based on local gradient guidance | |
US10650498B2 (en) | System, method, and non-transitory, computer-readable medium containing instructions for image processing | |
CN111563896A (en) | Image processing method for catenary anomaly detection | |
CN110706208A (en) | Infrared dim target detection method based on tensor mean square minimum error | |
CN107797110B (en) | System imaging method of synthetic aperture projection radiation and synthetic aperture projection radiometer | |
CN111462254A (en) | Multispectral monitoring method and system | |
CN107316009A (en) | High spectrum image object detection method based on tensor linear discriminant analysis dimensionality reduction | |
Zhang et al. | Feature-band-based unsupervised hyperspectral underwater target detection near the coastline | |
CN107895361A (en) | A kind of hyperspectral abnormity detection method based on local density's purifying background | |
CN112418156B (en) | Unsupervised classification-based optical remote sensing image sea surface ship target detection method | |
CN109190506A (en) | It is a kind of based on core is sparse and the EO-1 hyperion object detection method of space constraint | |
KR20150120805A (en) | Method and system for detecting human in range image | |
Lam | A fast approach for detecting human faces in a complex background | |
CN112686880A (en) | Method for detecting abnormity of railway locomotive component | |
KR101934317B1 (en) | System for automatic registration of images using association analysis of linear features | |
Guo et al. | Hyperspectral Target Detection with Target Prior Augmentation and Background Suppression-based Multi-Detector Fusion | |
CN111860616A (en) | General acquisition method for weak contrast collimation image target center of comprehensive diagnosis system |
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 |