CN108765460A - Space-time joint method for detecting abnormality based on high spectrum image and electronic equipment - Google Patents
Space-time joint method for detecting abnormality based on high spectrum image and electronic equipment Download PDFInfo
- Publication number
- CN108765460A CN108765460A CN201810493266.2A CN201810493266A CN108765460A CN 108765460 A CN108765460 A CN 108765460A CN 201810493266 A CN201810493266 A CN 201810493266A CN 108765460 A CN108765460 A CN 108765460A
- Authority
- CN
- China
- Prior art keywords
- high spectrum
- spectrum image
- frame
- image sequence
- anomaly
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The space-time joint method for detecting abnormality and electronic equipment, method that the invention discloses a kind of based on high spectrum image include：For high spectrum image sequence, the free air anomaly figure of each frame image in the high spectrum image sequence is obtained；Obtain the time anomaly figure of each frame image in the high spectrum image sequence；According to the joint Abnormal Map of previous frame image, the trajectory predictions figure of the target detection of present frame is obtained；Previous frame is the frame adjacent with present frame；According to the free air anomaly figure, time anomaly figure and trajectory predictions figure of each frame, the joint Abnormal Map of target in the high spectrum image cube is obtained.The method of the present invention, which is applied, has lower false alarm rate and higher detection probability in the detection scene of Aircraft Targets in the air.
Description
Technical field
The invention belongs to image recognition technology more particularly to a kind of space-time joint abnormality detection sides based on high spectrum image
Method and electronic equipment.
Background technology
With the development of bloom spectrum sensor, high spectrum image have been applied to many classics image processing problems it
In.Compared to single band sensor, bloom spectrum sensor can collect the spectral signature information and space characteristics of image simultaneously
Information.This is also considerable advantage of the high spectrum image in image processing field.
Weak target refers to small-sized, intensity is very weak, signal-to-noise ratio is very low interesting target in the picture.Weak target
Detection is all widely used in military and civilian field, has attracted the interest of Many researchers.Due to being deposited in high spectrum image
The signal-to-noise ratio of complex background, the decaying of the interference of noise clutter and long distance transmission, interested target can be very low.And
And different target has different signal-to-noise ratio, will also result in the omission or erroneous judgement of target detection.Researcher can not be from single wave
Accurately and reliably testing result is obtained in the infrared image of section.High spectrum image contains the spectral information of target, can often obtain
To better testing result.Therefore, how to detect and track Weak target as in Hyperspectral imagery processing field it is current urgently
One of the technical issues of solution.
Invention content
For the problems of the prior art, the present invention provides a kind of space-time joint abnormality detection side based on high spectrum image
Method and electronic equipment, this method can detect small and weak moving target, accuracy rate of testing result from high spectrum image sequence
Height, false alarm rate are low.
In a first aspect, the present invention provides a kind of space-time joint method for detecting abnormality based on high spectrum image, including：
101, it is directed to high spectrum image sequence, obtains the free air anomaly of each frame image in the high spectrum image sequence
Figure；
102, the time anomaly figure of each frame image in the high spectrum image sequence is obtained；
103, according to the object detection results of previous frame image, the trajectory predictions figure of present frame target detection is obtained；Upper one
Frame is the frame adjacent with present frame, and the object detection results are free air anomaly figure, the time anomaly figure by previous frame image
It is generated with trajectory predictions figure；
104, according to the free air anomaly figure, time anomaly figure and trajectory predictions figure of each frame, the EO-1 hyperion is obtained
The joint Abnormal Map of target in image sequence.
Optionally, before the step 101, the method further includes：
100, dimension-reduction treatment is carried out to the high spectrum image sequence, obtains the high spectrum image sequence after dimensionality reduction；
Correspondingly, the step 101 is specially：
Obtain the free air anomaly figure of each frame in the high spectrum image sequence after dimensionality reduction；
The step 102 is specially：
Obtain the time anomaly figure of each frame in the high spectrum image sequence after dimensionality reduction.
Optionally, the step 101 includes：
Using formula S (x, y, t)=(Vt-μt)T·(Φt)-1·(Vt-μt) obtain free air anomaly figure S (x, y, t)；
Wherein, Vt∈R1×kRefer to the feature vector of the pixel to be detected in PC (x, y, k, t), μt∈R1×kFinger PC (x, y, k,
T) mean value, ΦtRefer to the auto-covariance of PC (x, y, k, t)；PC (x, y, k, t) is the high spectrum image sequence or original after dimensionality reduction
High spectrum image sequence.
Optionally, the step 102 includes：
According to formula T (x, y, t)=(Vt-μt)T·(Φt+1)-1·(Vt-μt) obtain time anomaly figure T (x, y, t)；
Wherein, Vt∈R1×kRefer to the feature vector of the pixel to be detected in PC (x, y, k, t), μt∈R1×kFinger PC (x, y, k,
T) mean value, Φt+1Refer to the auto-covariance of PC (x, y, k, t+1)；
PC (x, y, k, t) be dimensionality reduction after high spectrum image sequence or original high spectrum image sequence present frame, PC (x,
Y, k, t+1) be dimensionality reduction after high spectrum image sequence or original high spectrum image sequence in present frame next frame.
Optionally, the step 103 includes：
According to formulaObtain present frame target detection trajectory predictions figure P (x, y,
t)；
STP (x, y, t-1) identifies the joint Abnormal Map of previous frame；
STP (x, y, t)=N (ST (x, y, t)+C) N (P (x, y, t)+C), N () refer to normalization operation, and C is a warp
Test constant, ST (x, y, t)=N (S (x, y, t)) N (T (x, y, t))
Optionally, the step 104 includes：
The high spectrum image sequence is obtained according to formula S TP (x, y, t)=N (ST (x, y, t)+C) N (P (x, y, t)+C)
The joint Abnormal Map STP (x, y, t) of target in row；
Wherein, ST (x, y, t)=N (S (x, y, t)) N (T (x, y, t)), P (x, y, t) be trajectory predictions figure, T (x, y,
T) be time anomaly figure, S (x, y, t) is free air anomaly figure.
Optionally, the method further includes：
105, the joint Abnormal Map of target in the high spectrum image sequence of acquisition is post-processed；
The post-processing includes：The processing of anti-interference process and adaptive threshold.
Or；
The anti-interference process includes：Using formula S TP (x, y, t)=(STP (x, y, t)) r to combining Abnormal Map STP
(x, y, t) enhances into row index, and r is an empirical；
Using formula Th=μSTP+k·σSTPAdaptive threshold Th is obtained, will be combined using adaptive threshold every in Abnormal Map
One point is handled, obtain treated joint Abnormal Map；
Wherein, μSTPIt is the mean value of joint Abnormal Map STP (x, y, t), σSTPTo combine the variance of Abnormal Map STP (x, y, t),
K is empirical.
Optionally, the spectral region of the high spectrum image sequence includes following whole wave bands or subband：
0.39~0.7 μm of visible light, 0.76~2.5 μm of short infrared, 3~5 μm of MID INFRARED and long-wave infrared
8~12 μm.
The subband of the present embodiment can be that the subband, the subband of short infrared, medium wave of visible light are red
Outside line, long-wave infrared subband, can also be selection the subbands such as visible light wave range and short infrared, can basis
Actual needs selection, does not limit its wavelength-division wave band.
Optionally, the step 100 includes：
Dimension-reduction treatment is carried out to the high spectrum image sequence using principal component analysis mode.
The device have the advantages that as follows：
High spectrum image sequence not only contains spatial information, further comprises temporal information and spectral information, of the invention
Method can solve the problems, such as weak moving target detection, and target is utilized in space, time and characteristic information spectrally in method,
Calculate separately free air anomaly figure, time anomaly figure.In order to detect the Movement consistency feature of target, it is pre- that this method generates track
Mapping.Free air anomaly figure, time anomaly figure and trajectory predictions figure are merged, interested mesh can be easily detected from background
Mark.This method is applied to cloud clutter background to get off the plane the test data set of target.The experimental results showed that this method is with relatively low
False alarm rate and higher detection probability.Further, small target deteection suffers from important work in civilian and military field
With.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention without having to pay creative labor, may be used also for those of ordinary skill in the art
To obtain other attached drawings according to these attached drawings.
Fig. 1 is the schematic diagram of the generating process for the hyperspectral datacube that one embodiment of the invention provides；
Fig. 2 is the reduction process schematic diagram for the hyperspectral datacube that one embodiment of the invention provides；
Fig. 3 is the schematic diagram for the space-time joint method for detecting abnormality that one embodiment of the invention provides；
Fig. 4 is the schematic diagram of a frame in the high spectrum image sequence that one embodiment of the invention provides；
Fig. 5 is tripleplane's schematic diagram in Fig. 4；
Fig. 6 is the schematic diagram of the experimental result of the method for the present invention.
Specific implementation mode
In order to preferably explain the present invention, in order to understand, below in conjunction with the accompanying drawings, by specific implementation mode, to this hair
It is bright to be described in detail.
In the following description, by multiple and different aspects of the description present invention, however, for common skill in the art
For art personnel, the present invention can be implemented just with some or all structures or flow of the present invention.In order to explain
Definition for, specific number, configuration and sequence are elaborated, however, it will be apparent that these specific details the case where
Under can also implement the present invention.It in other cases, will no longer for some well-known features in order not to obscure the present invention
It is described in detail.
The method of the embodiment of the present invention is detecting the small dim moving target of high spectrum image sequence.It is different that space is calculated first
Chang Tu.The spatial saliency information that free air anomaly figure can generally extract high spectrum image sequence by RXD algorithms obtains.So
Afterwards, based on adjacent two field pictures have similar background it is assumed that time domain EO-1 hyperion Abnormal Map can by calculate two frames it is continuous
High spectrum image obtains.In addition, trajectory predictions figure is also introduced in the present invention, to extract the motion continuity feature of target.Most
Afterwards, the joint Abnormal Map of target is merged to obtain by free air anomaly figure, time domain Abnormal Map and trajectory predictions figure.
The high spectrum image that bloom spectrum sensor generates can include tens spectral bands, such as visible light (0.39~0.7
μm), short infrared (SWIR:0.76~2.5 μm), MID INFRARED (MWIR:3~5 μm) and long-wave infrared (LWIR:8
~12 μm).The spectral region of the high spectrum image sequence used in the embodiment of the present invention includes visible light and part short-wave infrared
Line (0.76~0.96 μm).The image that this group generates, referred to as high spectrum image cube, by EO-1 hyperion camera in corresponding spectrum
Shooting in range, the sampling interval is about 10nm.Each cube includes 25 wave bands, by EO-1 hyperion camera in entire light
25 width images are generated in spectral limit, as shown in Figure 1.
The data volume for the high spectrum image that bloom spectrum sensor generates, several times usually higher than common infrared image.Therefore, after
Continuous image analysis system must improve operational capability, to meet the requirement of image transmitting, storage, calculating etc..In order to subtract
Lack calculation amount, dimensionality reduction pretreatment has been carried out to high spectrum image in the embodiment of the present invention.For example, principal component analysis can be used
(PCA) dimension of hyperspectral image data is reduced.In Hyperspectral imagery processing, PCA provides a kind of simple method, no
The dimension that data can only be reduced, can also inhibit noise.In addition, the data that PCA is obtained can also regard original EO-1 hyperion as
The feature of image data.Following part embodiment uses hyperspectral image data, subregion to use high spectrum image, refers to
Meaning is identical.
As illustrated in fig. 2, it is assumed that each hyperspectral image data includes L wave band, and the image I of each wave bandn(n=
1,2 ..., L) suffer from identical size m × n.So each hyperspectral image data may be expressed as I=[I1；
I2；...；IL], the purpose that the size of this image data is m × n × L.PCA is that the dimension of data is reduced to k dimensions simultaneously from L dimensions
And a dimensionality reduction matrix PC (x, y, k) is generated, size is m × n × k, as shown in Figure 2.After PCA dimension-reduction treatment, high-spectrum
Each pixel of picture is owned by the feature vector that a size is 1 × k.
Based on high spectrum image/hyperspectral image data/high spectrum image sequence after above-mentioned dimension-reduction treatment, the present invention is real
The method for applying example may include following steps：
101, the free air anomaly figure of each frame image in high spectrum image after dimensionality reduction is obtained.
As shown in figure 3, in figure 3, (x, y) refers to image pixel point coordinates, t refers to time coordinate.
Since free air anomaly figure is suitble to excavated space singularity characteristics.For the high spectrum image after dimensionality reduction, due to not having
Priori, the detection of target are that have strong different pixel from ambient background based on searching.RX algorithms are a kind of abnormality detections
Algorithm.RX algorithms can indicate as follows:
In formula, x is detected pixel,It is the mean value of image, Φ is the auto-covariance of image.
In the present embodiment, in order to find abnormal pixel in spatial domain, RX algorithms be used to calculate free air anomaly figure S
(x,y,t).Free air anomaly figure S (x, y, t) can be expressed from the next：
S (x, y, t)=(Vt-μt)T·(Φt)-1·(Vt-μt) (2)
In formula, Vt∈R1×kRefer to the feature vector of the pixel to be detected in PC (x, y, k, t), μt∈R1×kFinger PC (x, y, k,
T) mean value, ΦtRefer to the auto-covariance of PC (x, y, k, t).
Free air anomaly figure described in the present embodiment can be regarded as the matrix for m × n, and the value of each element exists in matrix
0, between 1.Free air anomaly figure is used for describing the intensity of anomaly of each pixel in high spectrum image.It is a certain in free air anomaly figure
The numerical value of a pixel is smaller, illustrates that the degree of the point exception is lower, more tends to the background in target detection；The numerical value of the point
It is bigger, illustrate that the intensity of anomaly of the point is higher, more tends to the target in target detection.Free air anomaly figure is according to high spectrum image
Space characteristics obtain, be calculated in the present embodiment by RXD algorithms.
102, the time anomaly figure of each frame image in high spectrum image after dimensionality reduction is obtained.
Time anomaly figure is used for detection time singularity characteristics in the present embodiment.The background sample of the previous frame of hypothesis is being worked as
In previous frame still effectively, and in this case, a point-device background estimating can be established over time.
Principal component of the calculating of time anomaly figure T (x, y, t) depending on present frame high spectrum image and next frame high spectrum image.With sky
Between Abnormal Map it is different, the information of present frame and next frame image is utilized in time anomaly figure simultaneously, as shown in formula (3):
T (x, y, t)=(Vt-μt)T·(Φt+1)-1·(Vt-μt) (3)
Wherein, Vt∈R1×kRefer to the feature vector of the pixel to be detected in PC (x, y, k, t), μt∈R1×kFinger PC (x, y, k,
T) mean value, Φt+1Refer to the auto-covariance of PC (x, y, k, t+1).
Time anomaly figure described in the present embodiment can be regarded as the matrix of m × n, in matrix the value of each element 0,
Between 1.Similar to free air anomaly figure, the time anomaly figure in the step is equally used for describing each pixel in high spectrum image
The intensity of anomaly of point.The numerical value of some pixel is smaller in time anomaly figure, illustrates that the degree of the point exception is lower, is more inclined to
Background in target detection；The numerical value of the point is bigger, illustrates that the intensity of anomaly of the point is higher, more tends in target detection
Target.Different from free air anomaly figure, time anomaly figure considers the singularity of present frame and next frame image, and space characteristics figure
Only only account for the singularity of current frame image.Next frame image is considered as background by time anomaly figure, is examined on current frame image
Survey abnormal point.Free air anomaly figure is obtained according to high spectrum image temporal characteristics, is calculated in the present embodiment by a kind of improved RXD
Method is calculated.
103, according to the object detection results of previous frame image, the trajectory predictions figure of present frame target detection is obtained；Upper one
Frame is the frame adjacent with present frame.
The object detection results of previous frame image are different by the free air anomaly figure of previous frame image, time in the present embodiment
Often figure and trajectory predictions figure generate.
It will be appreciated that in the present embodiment the trajectory predictions map generalization of present frame contain incessantly previous frame space,
Time anomaly figure, the trajectory predictions figure for further comprising previous frame.Trajectory predictions map generalization is the mistake of a continuous iteration in fact
Journey.The trajectory predictions figure of first frame can be preset in high spectrum image sequence, and the trajectory predictions figure of subsequent frames can be
It is obtained by iterative manner.
For example, for most of high spectrum image sequence, even if target size is small, intensity is weak, but target
Still there is certain similitude between frames, and the noise in image sequence is often discontinuous.Target is being schemed
As the property continuously occurred in sequence consecutive frame can become interframe continuity.Based on above-mentioned interframe continuity it is assumed that this implementation
The motion continuity feature of trajectory predictions figure extraction target can be used in example.Assuming that the target in image sequence will not move on a large scale
It is dynamic, then trajectory predictions figure can be calculated by the convolution kernel of a ζ × ζ.The calculating of trajectory predictions figure P (i, j, t)
Need the joint Abnormal Map using previous frame.Trajectory predictions figure P (i, j, t) can be indicated such as following formula：
In formula, STP (x, y, t-1) refers to the joint Abnormal Map that previous frame is calculated.The specific calculating of STP (x, y, t-1)
Following formula (6).
104, according to the free air anomaly figure, time anomaly figure and trajectory predictions figure of each frame, high spectrum image is obtained
The joint Abnormal Map of middle target.
Specifically, free air anomaly figure, time anomaly and trajectory predictions figure are merged；
First, free air anomaly figure and time anomaly figure are merged, is shown below:
ST (x, y, t)=N (S (x, y, t)) N (T (x, y, t)) (5)
In formula, N () normalizes operation, and ST (x, y, t) refers to the image singularity characteristics in space-time joint domain.
In addition, in order to continuously detected Weak target, final joint Abnormal Map has also merged trajectory predictions figure P
(x, y, t), is shown below：
STP (x, y, t)=N (ST (x, y, t)+C) N (P (x, y, t)+C) (6)
In formula, N () refers to normalization operation, and C is an empirical, and is set as 1 × 10 in this experiment-8.Often
Number C enhances the robustness of this method in an iterative process, and ensures Abnormal Map ST (i, j, t) and trajectory diagram P (i, j, t)
In the equal non-zero of each element.
Combine Abnormal Map described in the present embodiment and can be regarded as the matrix of m × n, in matrix the value of each element 0,
Between 1.By above-mentioned formula (6) it is found that joint Abnormal Map is merged to obtain by free air anomaly figure, time anomaly figure, trajectory predictions figure.
Joint Abnormal Map has investigated spatial singularity, time singularity, motion continuity in high spectrum image.Therefore compared to single
The target detection of Abnormal Map, joint abnormality detection can preferably detect the target in complex background, and can reduce void
The alert probability occurred.
Optionally, the above method may also include：
105, prediction locus is post-processed；For example, anti-interference process and adaptive thresholding etc..
It is anti-interference：In the target detection of high spectrum image, the signal-to-noise ratio of target to be detected and interference is different.It is dry
Disturb the decades of times that the value in joint Abnormal Map STP (x, y, t) is often target.In order to enhance the detection result of target, to STP
(x, y, t) enhances into row index, as following formula indicates：
STP (x, y, t)=(STP (x, y, t))r (7)
In formula, r is an empirical, and is set as in this experiment
Adaptive threshold：Since finally obtained joint Abnormal Map is shown between blur motion target and complex background
Inherent difference.In order to maximize signal-to-noise ratio (SNR), the present embodiment is provided with an adaptive threshold, as follows:
Th=μSTP+k·σSTP (8)
In formula, μSTPIt is the mean value of joint Abnormal Map STP (x, y, t), σSTPTo combine the variance of Abnormal Map STP (x, y, t),
K is empirical, and 10 are set as in this experiment.
In the present embodiment using adaptive threshold will combine Abnormal Map in each point handle, obtain treated connection
Close Abnormal Map.
For example, each point is a probability value in joint Abnormal Map, by adaptive threshold by each point in figure
It is classified as target or background.It illustrates, it is assumed that adaptive threshold is set as 0.5, and it is 0.3 to combine the point value in Abnormal Map, small
In threshold value, it is set as 0；Another point value is 0.8, is more than threshold value, is set as 1.Adaptive threshold is changed into binary map by Abnormal Map is combined,
The pixel that numerical value is 0 in figure, it is background in original image to represent the point；The pixel that numerical value is 1 in figure, represents the point and exists
It is target in original image.
Further, during specific implementation, above-mentioned steps 101 to step 105 are directed in high spectrum image
Each frame traverses each frame in image sequence.
In addition, in a preferred embodiment, can also save and calculate the time, each pass in high spectrum image is selected
Key frame executes above-mentioned step 101 to step 105, the selection of key frame can be used the technology of identification key frame in the prior art into
Row can also be to select at interval of N frames as one frame of selection as key frame, for example, N=3,4,5 etc..
In following experimental examples, inventor is to choose an image cube as key frame, only every five frames
In order to reduce the calculation amount in experiment.Even if not choosing key frame, each frame is all used to calculate, and effect is also the same.
Experimental example
In order to evaluate the performance of weak moving target detection method, by the high spectrum image sequence group under complicated cloud background
At a test data set.As shown in Figure 4 and Figure 5, Fig. 4 shows a wave band (wave in hyperspectral image data sequence
A length of 0.68 μm) effect, Fig. 5 shows tripleplane's effect of the wave band hypograph, due to Weak target size very
It is small, it is difficult to distinguish interested target from background；And cloud background and noise clutter can lead to low signal-to-noise ratio.
Using the method for the embodiment of the present invention, according to fixed interval l=5, selection needs to examine in order from original series
The key frame of survey.Moving target in key frame is labeled, to carry out qualitative assessment.PCA principal components in this experiment
The size of convolution kernel ζ in dimension k and trajectory predictions figure is respectively set to 10 and 3.
Above-mentioned Fig. 6 shows effect of this method in committed step.Under above-mentioned parameter configuration, to the data constructed
All images concentrated all are tested, and visualization result is as shown in Figure 5.It is worth noting that, what the present embodiment was proposed
Space-time joint method can more reduce false alarm rate than his method.Experimental result can prove the feasibility of EO-1 hyperion Dim targets detection
And performance, as shown in Figure 6
The embodiment of the present invention solves the problems, such as weak moving target detection in the prior art.The experimental result of the present invention
Show that the space-time joint exception method of proposition obtains good EO-1 hyperion weak moving target detection effect.The above method expands
Exhibition is applied in such as target identification, target following scene.
Another aspect according to the ... of the embodiment of the present invention, the embodiment of the present invention also provide a kind of electronic equipment, the electronic equipment
Including memory, processor, bus and store the computer program that can be run on a memory and on a processor, the place
Manage the method and step realized when device executes described program such as above-mentioned any embodiment.The electronic equipment of the present embodiment can be mobile
Terminal, fixed terminal etc..
Further, the present embodiment also provides a kind of computer storage media, is stored thereon with computer program, the journey
The method and step such as above-mentioned any embodiment is realized when sequence is executed by processor.
Finally it should be noted that：Above-described embodiments are merely to illustrate the technical scheme, rather than to it
Limitation；Although the present invention is described in detail referring to the foregoing embodiments, it will be understood by those of ordinary skill in the art that：
It can still modify to the technical solution recorded in previous embodiment, or to which part or all technical features into
Row equivalent replacement；And these modifications or substitutions, it does not separate the essence of the corresponding technical solution various embodiments of the present invention technical side
The range of case.
Claims (11)
1. a kind of space-time joint method for detecting abnormality based on high spectrum image, which is characterized in that including：
101, it is directed to high spectrum image sequence, obtains the free air anomaly figure of each frame image in the high spectrum image sequence；
102, the time anomaly figure of each frame image in the high spectrum image sequence is obtained；
103, according to the object detection results of previous frame image, the trajectory predictions figure of present frame target detection is obtained；Previous frame is
The frame adjacent with present frame, the object detection results are the free air anomaly figure, time anomaly figure and rail by previous frame image
What mark prognostic chart generated；
104, according to the free air anomaly figure, time anomaly figure and trajectory predictions figure of each frame, the high spectrum image is obtained
The joint Abnormal Map of target in sequence.
2. according to the method described in claim 1, it is characterized in that, before the step 101, the method further includes：
100, dimension-reduction treatment is carried out to the high spectrum image sequence, obtains the high spectrum image sequence after dimensionality reduction；
Correspondingly, the step 101 is specially：
Obtain the free air anomaly figure of each frame in the high spectrum image sequence after dimensionality reduction；
The step 102 is specially：
Obtain the time anomaly figure of each frame in the high spectrum image sequence after dimensionality reduction.
3. method according to claim 1 or 2, which is characterized in that the step 101 includes：
Using formula S (x, y, t)=(Vt-μt)T·(Φt)-1·(Vt-μt) obtain free air anomaly figure S (x, y, t)；
Wherein, Vt∈R1×kRefer to the feature vector of the pixel to be detected in PC (x, y, k, t), μt∈R1×kRefer to PC's (x, y, k, t)
Mean value, ΦtRefer to the auto-covariance of PC (x, y, k, t)；PC (x, y, k, t) be dimensionality reduction after high spectrum image sequence or original bloom
Compose image sequence.
4. method according to any one of claims 1 to 3, which is characterized in that the step 102 includes：
According to formula T (x, y, t)=(Vt-μt)T·(Φt+1)-1·(Vt-μt) obtain time anomaly figure T (x, y, t)；
Wherein, Vt∈R1×kRefer to the feature vector of the pixel to be detected in PC (x, y, k, t), μt∈R1×kRefer to PC's (x, y, k, t)
Mean value, Φt+1Refer to the auto-covariance of PC (x, y, k, t+1)；
PC (x, y, k, t) be dimensionality reduction after high spectrum image sequence or original high spectrum image sequence present frame, PC (x, y, k,
T+1) be dimensionality reduction after high spectrum image sequence or original high spectrum image sequence in present frame next frame.
5. method according to any one of claims 1 to 4, which is characterized in that the step 103 includes：
According to formulaObtain the trajectory predictions figure P (x, y, t) of the target detection of present frame；
STP (x, y, t-1) identifies the joint Abnormal Map of previous frame；
STP (x, y, t)=N (ST (x, y, t)+C) N (P (x, y, t)+C), N () refer to normalization operation, and C is that an experience is normal
Number, ST (x, y, t)=N (S (x, y, t)) N (T (x, y, t)).
6. method according to any one of claims 1 to 5, which is characterized in that the step 104 includes：
It is obtained in the high spectrum image sequence according to formula S TP (x, y, t)=N (ST (x, y, t)+C) N (P (x, y, t)+C)
The joint Abnormal Map STP (x, y, t) of target；
Wherein, ST (x, y, t)=N (S (x, y, t)) N (T (x, y, t)), P (x, y, t) are trajectory predictions figure, T (x, y, t) is
Time anomaly figure, S (x, y, t) are free air anomaly figure.
7. method according to any one of claims 1 to 6, which is characterized in that the method further includes：
105, the joint Abnormal Map of target in the high spectrum image sequence of acquisition is post-processed；
The post-processing includes：The processing of anti-interference process and adaptive threshold；
Or；
The anti-interference process includes：Using formula S TP (x, y, t)=(STP (x, y, t))rTo joint Abnormal Map STP (x, y, t)
Enhance into row index, r is an empirical；
Using formula Th=μSTP+k·σSTPAdaptive threshold Th is obtained, each in Abnormal Map will be combined using adaptive threshold
Point handled, obtain treated joint Abnormal Map；
Wherein, μSTPIt is the mean value of joint Abnormal Map STP (x, y, t), σSTPFor the variance of joint Abnormal Map STP (x, y, t), k is
Empirical.
8. method according to any one of claims 1 to 7, which is characterized in that
The spectral region of the high spectrum image sequence includes following whole wave bands or subband：
0.39~0.7 μm of visible light, 0.76~2.5 μm of short infrared, 3~5 μm of MID INFRARED and long-wave infrared 8~12
μm。
9. according to the method described in claim 2, it is characterized in that, the step 100 includes：
Dimension-reduction treatment is carried out to the high spectrum image sequence using principal component analysis mode.
10. a kind of electronic equipment, which is characterized in that on a memory and can be including memory, processor, bus and storage
The computer program run on processor, the processor are realized when executing described program such as claim 1-9 any one
Step.
11. a kind of computer storage media, is stored thereon with computer program, it is characterised in that：Described program is held by processor
It is realized such as the step of claim 1-9 any one when row.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810493266.2A CN108765460B (en) | 2018-05-22 | 2018-05-22 | Hyperspectral image-based space-time joint anomaly detection method and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810493266.2A CN108765460B (en) | 2018-05-22 | 2018-05-22 | Hyperspectral image-based space-time joint anomaly detection method and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108765460A true CN108765460A (en) | 2018-11-06 |
| CN108765460B CN108765460B (en) | 2021-03-30 |
Family
ID=64008377
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810493266.2A Active CN108765460B (en) | 2018-05-22 | 2018-05-22 | Hyperspectral image-based space-time joint anomaly detection method and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108765460B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109697431A (en) * | 2018-12-29 | 2019-04-30 | 哈尔滨工业大学 | A kind of detection method of small target based on high spectrum image |
| CN109949278A (en) * | 2019-03-06 | 2019-06-28 | 西安电子科技大学 | Hyperspectral abnormity detection method based on confrontation autoencoder network |
| CN112990106A (en) * | 2021-04-19 | 2021-06-18 | 中国人民解放军国防科技大学 | Underwater object detection method and device, computer equipment and storage medium |
| CN113553914A (en) * | 2021-06-30 | 2021-10-26 | 核工业北京地质研究院 | Abnormal target detection method for CASI (computer aided design) hyperspectral data |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103426167A (en) * | 2013-07-09 | 2013-12-04 | 哈尔滨工程大学 | Hyperspectral real-time detection method based on recursive analysis |
| CN103606157A (en) * | 2013-11-28 | 2014-02-26 | 中国科学院光电研究院 | Method and device for processing hyperspectral image |
| CN104616299A (en) * | 2015-01-30 | 2015-05-13 | 南京邮电大学 | Method for detecting weak and small target based on space-time partial differential equation |
| CN105825200A (en) * | 2016-03-31 | 2016-08-03 | 西北工业大学 | High-spectrum abnormal object detection method based on background dictionary learning and structure sparse expression |
| CN106101732A (en) * | 2016-07-05 | 2016-11-09 | 重庆邮电大学 | The vector quantization scheme of Fast Compression EO-1 hyperion signal |
| CN106600602A (en) * | 2016-12-30 | 2017-04-26 | 哈尔滨工业大学 | Clustered adaptive window based hyperspectral image abnormality detection method |
| CN107704835A (en) * | 2017-10-16 | 2018-02-16 | 北京市遥感信息研究所 | A kind of method using spectral remote sensing image recognition sea Artificial facilities |
-
2018
- 2018-05-22 CN CN201810493266.2A patent/CN108765460B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103426167A (en) * | 2013-07-09 | 2013-12-04 | 哈尔滨工程大学 | Hyperspectral real-time detection method based on recursive analysis |
| CN103606157A (en) * | 2013-11-28 | 2014-02-26 | 中国科学院光电研究院 | Method and device for processing hyperspectral image |
| CN104616299A (en) * | 2015-01-30 | 2015-05-13 | 南京邮电大学 | Method for detecting weak and small target based on space-time partial differential equation |
| CN105825200A (en) * | 2016-03-31 | 2016-08-03 | 西北工业大学 | High-spectrum abnormal object detection method based on background dictionary learning and structure sparse expression |
| CN106101732A (en) * | 2016-07-05 | 2016-11-09 | 重庆邮电大学 | The vector quantization scheme of Fast Compression EO-1 hyperion signal |
| CN106600602A (en) * | 2016-12-30 | 2017-04-26 | 哈尔滨工业大学 | Clustered adaptive window based hyperspectral image abnormality detection method |
| CN107704835A (en) * | 2017-10-16 | 2018-02-16 | 北京市遥感信息研究所 | A kind of method using spectral remote sensing image recognition sea Artificial facilities |
Non-Patent Citations (9)
| Title |
|---|
| CHEIN-I CHANG 等: "Anomaly detection and classification for hyperspectral imagery", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
| I. S. REED 等: "Adaptive multiple-band CFAR detection of an optical pattern with unknown spectral distribution", 《IEEE TRANSACTIONS ON ACOUSTICS SPEECH AND SIGNAL PROCESSING》 * |
| IHAB MAKKI: "A survey of landmine detection using hyperspectral imaging", 《ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING》 * |
| M. D. FARRELL: "On the impact of PCA dimension reduction for hyperspectral detection of difficult targets", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
| YANSHENG LI: "A novel spatio-temporal saliency approach for robust dim moving target detection from airborne infrared image sequences", 《INFORMATION SCIENCES》 * |
| 史振威: "RX 及其变种在高光谱图像中的异常检测", 《红外与激光工程》 * |
| 成宝芝: "联合空间预处理与谱聚类的协同稀疏高光谱异常检测", 《光学学报》 * |
| 赵春晖 等: "基于逐行处理的高光谱实时异常目标检测", 《光学学报》 * |
| 鞠建华: "《国土资源遥感应用》", 31 December 2011, 地质出版社 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109697431A (en) * | 2018-12-29 | 2019-04-30 | 哈尔滨工业大学 | A kind of detection method of small target based on high spectrum image |
| CN109697431B (en) * | 2018-12-29 | 2021-11-23 | 哈尔滨工业大学 | Hyperspectral image-based weak and small target detection method |
| CN109949278A (en) * | 2019-03-06 | 2019-06-28 | 西安电子科技大学 | Hyperspectral abnormity detection method based on confrontation autoencoder network |
| CN109949278B (en) * | 2019-03-06 | 2021-10-29 | 西安电子科技大学 | Hyperspectral anomaly detection method based on antagonistic self-coding network |
| CN112990106A (en) * | 2021-04-19 | 2021-06-18 | 中国人民解放军国防科技大学 | Underwater object detection method and device, computer equipment and storage medium |
| CN113553914A (en) * | 2021-06-30 | 2021-10-26 | 核工业北京地质研究院 | Abnormal target detection method for CASI (computer aided design) hyperspectral data |
| CN113553914B (en) * | 2021-06-30 | 2024-03-19 | 核工业北京地质研究院 | CASI hyperspectral data abnormal target detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108765460B (en) | 2021-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kwon et al. | Kernel RX-algorithm: A nonlinear anomaly detector for hyperspectral imagery | |
| Li et al. | Single image deraining using scale-aware multi-stage recurrent network | |
| CN108765460A (en) | Space-time joint method for detecting abnormality based on high spectrum image and electronic equipment | |
| CN104408482B (en) | A kind of High Resolution SAR Images object detection method | |
| US10614736B2 (en) | Foreground and background detection method | |
| US10706558B2 (en) | Foreground and background detection method | |
| Johnson et al. | AutoGAD: An improved ICA-based hyperspectral anomaly detection algorithm | |
| Maggio et al. | Learning scene context for multiple object tracking | |
| CN110400294B (en) | Infrared target detection system and detection method | |
| Hou et al. | Spatial–spectral weighted and regularized tensor sparse correlation filter for object tracking in hyperspectral videos | |
| Medouakh et al. | Improved object tracking via joint color-LPQ texture histogram based mean shift algorithm | |
| Shu et al. | Small moving vehicle detection via local enhancement fusion for satellite video | |
| Xu et al. | COCO-Net: A dual-supervised network with unified ROI-loss for low-resolution ship detection from optical satellite image sequences | |
| Liang et al. | Robust infrared small target detection using Hough line suppression and rank-hierarchy in complex backgrounds | |
| Li et al. | A cascade method for infrared dim target detection | |
| Alam et al. | Hyperspectral target detection using Gaussian filter and post-processing | |
| US20190251695A1 (en) | Foreground and background detection method | |
| Wang et al. | A background self-learning framework for unstructured target detectors | |
| CN115984325A (en) | Target tracking method for target volume searching space-time regularization | |
| Prakash et al. | Tracking of moving object using energy of biorthogonal wavelet transform | |
| CN114429593A (en) | Infrared small target detection method based on rapid guided filtering and application thereof | |
| Patil et al. | Detection and tracking of moving object: A survey | |
| Chen et al. | Color–depth multi-task learning for object detection in haze | |
| Yu et al. | Visual Saliency via Multiscale Analysis in Frequency Domain and Its Applications to Ship Detection in Optical Satellite Images | |
| Pulpito et al. | Infrared saliency enhancement techniques for extended naval target detection in open sea scenario |
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 |