CN108133468B - Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection - Google Patents

Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection Download PDF

Info

Publication number
CN108133468B
CN108133468B CN201711422484.9A CN201711422484A CN108133468B CN 108133468 B CN108133468 B CN 108133468B CN 201711422484 A CN201711422484 A CN 201711422484A CN 108133468 B CN108133468 B CN 108133468B
Authority
CN
China
Prior art keywords
target
ship
pixel
image
apt
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
Application number
CN201711422484.9A
Other languages
Chinese (zh)
Other versions
CN108133468A (en
Inventor
任侃
孟思岐
陈钱
顾国华
钱惟贤
路东明
王敏
王佳佳
朱宇遥
过玲钰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN201711422484.9A priority Critical patent/CN108133468B/en
Publication of CN108133468A publication Critical patent/CN108133468A/en
Application granted granted Critical
Publication of CN108133468B publication Critical patent/CN108133468B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration by the use of histogram techniques
    • 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
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • 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
    • G06T2207/10044—Radar image

Abstract

The invention provides a constant false alarm rate ship detection method for self-adaptive parameter enhancement and trail auxiliary detection, which comprises the following steps of: step 1, mapping an input SAR image to an APT domain; step 2, CFAR detection of the ship target in the APT domain; and 3, eliminating false alarms of the ship target.

Description

Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection
Technical Field
The invention relates to an image enhancement and target detection technology, in particular to a constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection.
Background
Synthetic Aperture Radar (SAR) images have been widely used for marine traffic monitoring and fishing vessel detection. In all traditional and modern tools, such as land-based, ship-based and airborne detection modes, visible light and infrared remote sensing wave bands and the like, the SAR has various advantages of all-weather, long distance, wide range and the like in the aspect of data collection. Due to its special imaging principle, SAR can provide high resolution remote sensing images. Thus, it facilitates the detection of ships and trails, and additional parameters can be extracted from the wake for estimating the ship's state of motion, such as speed and heading.
Ships often appear as bright objects in SAR images because they are strong reflectors of radar pulses transmitted by satellites. In the image with higher resolution, the structure of the ship can be distinguished, and the ship can be detected and even the type of the ship can be identified. And the more image space information, the better the ship detection effect. In SAR images, the most obvious feature of the trail is the linear feature. Vessel trails often appear as light or dark lines that may remain several kilometers behind the vessel. However, due to the influence of sea clutter and speckle noise, the target characteristics are not obvious in the SAR image, and the detection of ships and trails is greatly influenced. Many algorithms have achieved good results in the detection of objects on ships. Constant False Alarm Rate (CFAR) detection is one of the most common ship detection methods, which is based on a constant false alarm rate and an adaptive threshold. The common CFAR algorithms are mainly: two parameters CFAR (2p-CFAR) and cell average CFAR (CA-CFAR), etc. They use only the intensity features of the target for detection, which is a waste of target spatial structure information in high resolution SAR images. In addition, there is still a lack of effective constraints on the characteristics of targets in ship target detection. A target screening criterion is defined according to actual target characteristics, and the accuracy of detection is ensured by the whole system method by combining ship trail detection.
Disclosure of Invention
The invention provides a constant false alarm rate ship detection method for self-adaptive parameter enhancement and trail auxiliary detection, which comprises the following steps of:
step 1, mapping an input SAR image to an APT domain;
step 2, CFAR detection of the ship target in the APT domain;
and 3, eliminating false alarms of the ship target.
By adopting the method, the specific process of the step 1 is as follows:
step 1.1, quantizing 256-level gray scale features in the SAR image into 16 levels;
step 1.2, obtaining the similarity between two pixel positions (i, j) and (m, n) in the image, wherein the similarity is represented by an intensity difference id
id＝|level[f(i,j)]-level[f(m,n)]| (1)
Wherein f (i, j) and f (m, n) are gray values of (i, j) and (m, n), respectively, level represents a quantized level of a pixel, and | represents an absolute value;
step 1.3, determining a threshold value L, wherein the threshold value L is larger than the size of the sea clutter and smaller than half of the minimum size of the ship width;
step 1.4, dynamically distributing the weight t;
step 1.5, mapping the input SAR image to APT domain by the formula (2)
Figure BDA0001523254950000021
Wherein f isAPT(i, j) is the APT threshold, f (i, j) is the value of the grayscale image at (i, j), t is the adaptive weight of the different pixels, R (i ', j') is the euclidean distance between the pixel (i, j) and the pixel (i ', j'), id is the intensity similarity, α is a constant controlling the decay rate.
By adopting the method, the specific process of the step 1.4 is as follows:
step 1.4.1, set discriminant function
Figure BDA0001523254950000022
If true, then the target (the target comprises a ship and a trail), and go to step 1.4.2; if the target is the flase, carrying out no treatment;
f (i, j) is the grayscale image of the SAR image, μ is the image mean, σ is the image standard deviation,
Figure BDA0001523254950000023
is a design parameter;
step 1.4.2, setting a sliding window with the size of L at the position (i, j), and calculating the weight of a sub-window where the adjacent pixel (i, j) is located by using a formula (3), wherein the size of the sub-window is L; if the target is the flase, carrying out no treatment;
step 1.4.3, obtaining new value f of weightingenh(i,j)
Figure BDA0001523254950000031
Wherein the content of the first and second substances,
Figure BDA0001523254950000032
sumweightthe weight sum is obtained by calculating step 1.4.1 in the main window and calculating 1.4.2 in the sub-window; sumthresholdIs a segmentation threshold of the weight; Ω (L) represents the space occupied by a sliding window of L size, k represents the pixels within a sliding window of L size, k is ∈ [1,8 ]]。
By adopting the method, the target detection is realized by the following formula in the step 2:
Figure BDA0001523254950000033
Figure BDA0001523254950000034
wherein x isAPTIs a test pixel value, σbIs the standard deviation of the background and the standard deviation,
Figure BDA0001523254950000035
t is a detector design parameter; erf denotes the error function.
By adopting the method, the nuclear density estimation, the length-width ratio and the target area characteristics are selected in the step 3 to eliminate the false alarm of the ship target, if the characteristics are larger than the preset threshold value, the ship target is determined, wherein the ship target is determined
Obtaining a nuclear density estimate using equation (7)
Figure BDA0001523254950000036
Wherein x is1,x2,…,xjIs a pixel of a potential target;
the target area is the number of pixels of the potential target.
Compared with the prior art, the invention has the following remarkable advantages: (1) the method combines the strength and the spatial information to detect the ship by an APT-CFAR method, and in an APT domain, the characteristics of a target (a ship or a trail) are more prominent, and the model has stronger robustness; (2) under the condition of high target contrast, the ship target detection is more accurate by combining the self characteristics of the target and a trail auxiliary detection method.
The invention is further described below with reference to the accompanying drawings.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a SAR image histogram.
Fig. 3 is a process of calculating dynamic weights of neighboring pixels.
Fig. 4 is an APT process example.
Fig. 5 is a comparison of simulated image effects before and after transformation.
FIG. 6 is a comparison of the effects of using Terras SAR-X data.
FIG. 7 is a comparison of the ship target and wake in the original image and the ship after-mask wake in the APT domain.
FIG. 8 is a detection result of the enhanced two-value image of the trail and the trail.
Fig. 9 is a comparison of detection performance curves.
Detailed Description
With reference to fig. 1, the constant false alarm rate ship detection algorithm based on adaptive parameter enhancement and trail auxiliary detection provided by the invention is based on the APT threshold value. First, the image is algorithmically mapped to a new transform domain. Secondly, screening out candidate target pixels and highlighting the target ship through a CFAR detector and defining target characteristics. And replacing ship pixels with uniform ocean background, and processing the enhanced image by using a Niblack algorithm to obtain a wake binary image. Finally, the Normalized Hough Transform (NHT) is used to detect the trail in the binary image to verify the presence of the ship.
The specific implementation steps for realizing the constant false alarm rate ship detection algorithm based on the self-adaptive parameter enhancement and the trail auxiliary detection are as follows:
(1) and (4) adaptive parameter transformation. The correlation of the pixel gray level in the SAR image with the distribution distance is inseparable. The closer the distance between pixels, the stronger the gray scale similarity. And converting the intensity value of the original image space into a new transformation threshold value by combining the gray characteristic and the position space relation of the image pixel. The detailed steps are as follows:
1) difference in image gray level
The intensity values of neighboring pixels in the SAR image have very high similarity. We quantize 256 levels of grayscale features in the SAR image to 16 levels. For example, intensity values of 0-15 are on the order of 1, intensity values of 16-31 are on the order of 2, and intensity values of 240-255 are on the order of 16. The positions of the two pixels in the image are (i, j) and (m, n), respectively, and their gray values correspond to f (i, j) and f (m, n), respectively. The similarity of two pixels is defined by the difference of their respective levels. The intensity difference (id) is defined as follows.
id＝|level[f(i,j)]-level[f(m,n)]|
(1) We quantify the intensity difference between pixels by absolute value.
(2) Effective range
The effect of one pixel on another decreases with increasing distance. Clearly, there is a range L such that such effects are negligible. The range should not be too large to introduce excessive computer load, nor too small to limit the spatial relationship information we want to utilize. L depends on the ship target and the strength of the sea clutter. Experiments have shown that it should be larger than the size of the sea clutter and smaller than half the smallest dimension of the ship's width, so that the content contained by the sliding window can correctly reflect their spatial differences. Although L typically varies over a wide range, we prefer the smallest L for two reasons: one is that a smaller L can reduce more time consumption; and secondly, the space extraction influence on the ship target edge is small by using the small L.
3) Dynamic allocation of weights
To make it easier to compute the distribution of pixel intensities, we assume that the background clutter fits into a gaussian distribution. The proportion of background areas in the image is large. The vessel and wake regions are not only small in the image, but also darker (wake) or lighter (vessel or wake) than the background. Experiments have shown that the vast majority of the target pixels are clearly distributed at both ends of the histogram. According to this rule, dynamic distribution weights for different intensity values are given. Fig. 2 presents a histogram of a SAR image containing a ship and a trail.
From the above analysis, we incorporate the "3 σ" criterion of the Gaussian distribution. In case of avoiding missing detection, the discriminant function is:
Figure BDA0001523254950000051
where f (i, j) is the raw intensity value of the SAR image, μ is the image mean, σ is the image standard deviation,
Figure BDA0001523254950000052
is a design parameter. The empirical value of the intensity distribution is 2.5-3 based on a large amount of experimental data. When the intensity values satisfy equation (2), the pixel may be considered a ship or ship wake, however speckle noise may not be completely excluded.
In consideration of the gaussian distribution, when the pixel intensity value satisfies formula (2), the weight assignment of the neighboring pixels is calculated by setting a sliding window of L size at that position. Fig. 3 shows a process of calculating pixel weights. The main window is the process of converting the intensity image into APT domain, and the sub-window is the process of calculating the neighboring pixel weights. In the sub-window, the number of pixels satisfying the target condition is calculated. The weighting formula is as follows:
Figure BDA0001523254950000061
where n is the number of pixels that satisfy the condition in the sub-window, and L is the size of the window, and t is the weight of the neighboring pixels. However, the center pixel is not calculated in the main window. To suppress the effects of background clutter and speckle noise, the sum of the neighboring pixel weights is calculated in the main window.
Figure BDA0001523254950000062
If the weight satisfies the following condition, the formula is as follows:
Figure BDA0001523254950000063
in the above equation, f (i, j) is the original intensity value, fenh(i, j) is the new value of the weight, trestrainIs a suppression parameter, tkIs a weighting parameter of the pixel intensity, sumweightIs the sum of the weights of the pixels of the main window, sumthresholdIs the segmentation threshold of the weights.
By this method it can be concluded that: the more adjacent pixels satisfying the condition, the greater the probability of determining as the target pixel, and the greater the weight of the pixel. Conversely, when the number of pixels satisfying the condition in the sub-window is small, the pixels may be speckle noise, and the weight of the pixels is smaller. By this adaptive weight selection, image noise is relatively suppressed and contrast is enhanced.
4) Transition to APT Domain
According to the distribution of the pixel intensity values, the pixels are dynamically weighted by using the intensity similarity between adjacent pixels, and the APT domain value is calculated by combining the spatial distance. The more adjacent the pixel intensity value is similar to the adjacent pixel, the more obvious the superposition effect is. The mapping formula is as follows:
Figure BDA0001523254950000064
where f (i, j) is the value of the intensity image at (i, j), fAPT(i, j) is the APT threshold, t is the adaptive weight of the different pixels, R (i ', j') is the euclidean distance between pixel (i, j) and pixel (i ', j'), id is the intensity similarity. α is a constant that controls the decay rate, together with the intensity similarity id, controls the decay effect. It should be greater than 1 to ensure the amount of attenuation. If alpha is too small, pixels of different intensities will result in almost the same enhancement. If a is too large, the contribution of similar but not exactly equal intensity pixels will be very limited. This conclusion can be deduced from equation (6). Based on a large amount of experimental data, when the alpha value is 1.5-1.8, the SAR image is adjusted properly.
Fig. 4 shows an adaptive parameter transformation process. And taking a 3 × 3 image block on the SAR image. First, the image pixel intensity is magnified as shown in the upper portion of FIG. 4. The pixel level difference is calculated by different pixel intensity values. In the statistical block, the difference between the neighborhood pixel and the center pixel is represented as id, as shown in equation (1). Second, as shown in the middle part of fig. 4, a pixel weight t around the center pixel is calculated in each statistical block according to the intensity distribution. The total weight coefficient is calculated by the binding strength level id. Thirdly, by combining the above two steps, the central pixel intensity value and the euclidean distance between adjacent pixels are combined to calculate a new transform domain value. It can be seen that adjacent pixels with the same intensity value contribute differently to the central pixel. Furthermore, the influence of the same range of pixels on the center also varies with the respective intensities. This process is repeated for all image pixels. We normalized these new AP values to 0-255. Finally, we can obtain an adaptive parametric transform domain image that combines intensity and spatial structure information.
In the whole process we need to consider the following in particular: first, a region of interest (ROI) contains bright or dark pixel intensity values. The intensity values of these pixels are greatly enhanced in the transform domain. Second, background and speckle noise are slightly enhanced because the original intensity values are given less weight and the intensity similarity between these pixels is weak.
(2) CFAR detection in APT domain. CFAR is one of the adaptive thresholding algorithms, mainly used to search for pixel values that are usually bright compared to the surrounding sea area. It requires a high contrast of the target with respect to the background and enables target detection by comparing each pixel intensity with a certain threshold.
In the APT field, the characteristics of the target area are greatly enhanced based on the intensity and spatial information of the pixels. In contrast, the superposition effect of the ocean background clutter and the speckle noise area is not obvious. Generally, the image probability density distribution (PDF) of the APT field is close to a gaussian distribution. In addition, the enhanced image is closer to a Gaussian distribution model.
Based on the gaussian distribution model of the background clutter, the threshold can be calculated according to the false alarm Probability (PFA) formula:
Figure BDA0001523254950000071
fpdf(x) Is the probability density function (pdf) and the formula is as follows:
Figure BDA0001523254950000072
wherein x isAPTIs a test pixel value, mubIs the background mean value, σbIs the background standard deviation and t is the detector design parameter. t and PfaSatisfies the formula (9):
Figure BDA0001523254950000081
the error function (erf) is defined as:
Figure BDA0001523254950000082
according to the above formula, the threshold values are as follows:
Figure BDA0001523254950000083
finally, we achieve target detection by the following formula:
Figure BDA0001523254950000085
(3) target false alarm rejection. On the premise of ensuring higher detection rate, the CFAR detector is adopted to process the SAR image, and the problem of false alarm is inevitably caused. In combination with the characteristics of the target, we used the following characteristics as screening criteria.
1) Nuclear density estimation (KDE)
The density of nuclei reflects the degree of aggregation of the target, providing internal structural features of the target. Since false alarms typically occur in the form of discrete points or thin line segments and the ship target is displayed as a collection of pixels, it is clear that the target (ship or ship wake) has a higher KDE value. The estimate of the nuclear density is given by:
Figure BDA0001523254950000084
wherein x is1,x2,…,xjAre the pixels of the potential target.
2) Aspect Ratio (AR)
Aspect ratio refers to the ratio of the length to the width of a candidate object. The AR is independent of the size of the ship body and is the characteristic of the ship on the sea. It can be used to identify ship targets and highlights at sea.
3) Target area (TR)
The target area is the number of pixels of the potential target. The number of pixels of the vessel controls the target in size. Based on the pixel pitch, we can set a minimum threshold for the target size.
And if the characteristic is greater than a preset threshold value, the ship target is determined.
(4) Trail detection and result fusion
The characteristics of the marine vessel wake are very apparent compared to the vessel itself, usually representing a line segment of up to several kilometers behind the vessel. In fact, the azimuth offset appears in the SAR image due to the distance component of the moving ship, which appears on the graph as a deviation of the ship from the actual course. However, the wake indicates the true position of the ship. Therefore, ship trail detection can be convenient for ship detection and calculation of ship motion parameters.
The ship target and wake relationship is indivisible. The presence of the trail enables verification of the authenticity of the ship target. When the back side of the ship target can detect the trail, the target is considered as a ship; when only a ship or wake is detected, we still consider that there is a target. However, when the sea state is very complicated, the SAR image is affected by speckle noise and sea clutter, and in some cases, manual work is still required for target detection.
The effect of the present invention can be further illustrated by the following simulation results:
(1) naval vessel target enhancement and detection
1) Simulation data:
to verify the effect of the APT field, we tested the performance of the proposed algorithm using simulated images. Fig. 5(1.a) and (2.a) simulate the characteristics of the bright and dark regions, respectively. They are created by attaching a block of pixels with the same intensity value to a sea clutter SAR image without any targets. The simulated sea clutter background comes from TerrasAR-X, and the resolution is 3 meters. The image block size pointed to by the arrow is 5 x 10.
Fig. 5(1.a) shows that although the region itself has a higher intensity, it is still disturbed by sea clutter and speckle noise, as is clear from fig. 5(1. c). In view of these conditions, we use the proposed algorithm to convert the image to the new APT domain. Fig. 5(1.b) shows APT results. We can see that bright places become stronger, the background is suppressed, the image contrast becomes larger, so the object is easier to detect. In contrast, as shown in fig. 5(2.a), the pixel block has a darker intensity value. Fig. 5(2.c) shows the pixel blocks completely submerged in the background. The dark targets are also greatly enhanced according to the algorithm presented herein, taking into account the sea clutter model and target intensity.
2) Terras SAR-X data
The terrasaar-X image is from singapore sea area, as shown in fig. 6. The resolution is 3 meters azimuth. The polarization mode and channel are unipolar and HH, respectively. As shown in fig. 6, the first row displays the original intensity image, the enhanced image and the corresponding detected image. The second row shows a poor sea state image and corresponding enhancement and detection results. As can be seen from fig. 6, the target clutter contrast is low, and individual targets are less obvious. After the algorithm, the background clutter is suppressed, and the target is more prominent. In the enhanced image, the algorithm can accurately detect the target with less interference. In general, APT-CFAR can effectively improve image contrast and more accurately detect ship targets according to screening criteria.
(2) Trail enhancement and detection
In SAR images, the ship trail is either dark or bright, and is generally low in contrast and therefore not perceptible. Fig. 7(1.a) shows a dark trail accompanying the ship in the original image. Fig. 7(1.b) shows an enhanced trail image in the APT domain after the ship is masked. Fig. 7(2.a) and (2.b) are corresponding three-dimensional images, respectively. It can be seen that the trail is greatly enhanced and the target-clutter contrast becomes large. Furthermore, we know that the combination of space and intensity becomes more compact based on the proposed algorithm in the APT domain. Finally, we can see that the enhancement of the trail is very significant.
Fig. 8 shows the detection results of the binary image and the trail. For the enhanced image, we use the local threshold Niblack algorithm for image segmentation. To further enable detection results and parameter extraction, we labeled the trail using a normalized Hough transform. For a linear trail, it can effectively mark trail points. However, for non-linear trails, it may be necessary to employ multiple extrema in the integral domain to obtain a complete trajectory.
(3) Comparison with other algorithms
To evaluate the performance of the algorithm, we compared it to the standard CFAR and the bilateral CFAR. Standard CFAR uses intensity as a criterion for detection. Whereas bilateral CFARs use fused distributions for detection, but use intensity and spatial information in a different way. To evaluate the performance of the algorithm in detail, we obtained a Receiver Operating Characteristic (ROC) curve using the image from TerraSAR-X. Meanwhile, we define the target detection probability as:
Figure BDA0001523254950000101
wherein N istdIs the number of targets detected, Ntotal-targetIs the number of all targets. The estimation of the false alarm probability is defined as:
Figure BDA0001523254950000102
wherein N isfdIs the number of false alarm targets, Nground-trueIs the number of real targets.
The ROC curve is shown in fig. 9. It is clear that APTCFAR and bilateral CFAR are better than standard CFAR, just as the area under the curve (AUC) of the latter is smaller than the others. Since it only takes into account the intensity characteristics of the image. The trends for APT-CFAR and bilateral CFAR are similar. However, when less than 10-4At times, the APT-CFAR value is higher than the double CFAR. This means that the detection accuracy of the former is higher under more complicated sea conditions and lower contrast. Moreover, as the overall detection performance of the APT-CFAR is improved, the effect is more prominent, mainly because the process is based on dynamic enhancement of pixel weights, so that the target contrast is improved.

Claims (3)

1.A constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection is characterized by comprising the following steps:
step 1, mapping an input SAR image to an APT domain;
step 2, CFAR detection of the ship target in the APT domain;
step 3, removing false alarms of the ship target;
the specific process of the step 1 is as follows:
step 1.1, quantizing 256-level gray scale features in the SAR image into 16 levels;
step 1.2, obtaining the similarity between two pixel positions (i, j) and (m, n) in the image, wherein the similarity is represented by an intensity difference id
Figure DEST_PATH_IMAGE001
（1）
Wherein f (i, j) and f (m, n) are gray values of (i, j) and (m, n), respectively, and level represents the quantized pixel where the pixel is located
Level, |, denotes an absolute value;
step 1.3, determining a threshold value L, wherein the threshold value L is larger than the size of the sea clutter and smaller than half of the minimum size of the ship width; step 1.4, dynamically distributing the weight t;
step 1.5, mapping the input SAR image to APT domain by the formula (2)
Figure DEST_PATH_IMAGE002
（2）
Wherein f isAPT(i, j) is an APT threshold, f (i, j) is a value of the gray image at (i, j), t is an adaptive weight value of different pixels, R (i ', j') is a euclidean distance between the pixel (i, j) and the pixel (i ', j'), id is an intensity similarity, and α is a constant controlling the attenuation speed;
the specific process of step 1.4 is as follows:
step 1.4.1, set discriminant function
Figure DEST_PATH_IMAGE003
（3）
If true, the ship and the trail are determined, and the step 1.4.2 is carried out; if the target is the flase, carrying out no treatment; f (i, j) is the grayscale image of the SAR image, μ is the image mean, and σ is the image standard deviation
Figure DEST_PATH_IMAGE004
Is a design parameter;
step 1.4.2, setting a sliding window with the size of L at the position (i, j), and calculating the weight of a sub-window where the adjacent pixel (i, j) is located by using a formula (3), wherein the size of the sub-window is L; if the target is the flase, carrying out no treatment;
step 1.4.3, obtaining new value f of weightingenh(i,j)
Figure DEST_PATH_IMAGE005
（4）
Wherein
Figure DEST_PATH_IMAGE006
，trestrainIs a suppression parameter, tkIs a weighting parameter of the pixel intensity, sumweightThe weight sum is obtained by calculating step 1.4.1 in the main window and calculating 1.4.2 in the sub-window; sumthresholdIs a segmentation threshold of the weight; Ω (L) represents the space occupied by a sliding window of L size, k represents the pixels within a sliding window of L size, k is ∈ [1,8 ]]。
2. The method of claim 1, wherein the target detection is achieved in step 2 by:
Figure DEST_PATH_IMAGE007
（5）
Figure DEST_PATH_IMAGE008
（6）
wherein x isAPTIs a test pixel value, σbIs the standard deviation of the background
Figure DEST_PATH_IMAGE009
T is a detector design parameter; erf denotes the error function.
3. The method of claim 1, wherein the kernel density estimation, aspect ratio, and target area characteristics are selected in step 3 to exclude false alarm of the ship target, and if the characteristics are greater than a preset threshold, the ship target is selected, wherein the ship target is selected
Obtaining a nuclear density estimate using equation (7)
Figure DEST_PATH_IMAGE010
（7）
Wherein x is1 ,x2 ,… ,xjIs a pixel of a potential target; the target area is the number of pixels of the potential target.
CN201711422484.9A 2017-12-25 2017-12-25 Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection Active CN108133468B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711422484.9A CN108133468B (en) 2017-12-25 2017-12-25 Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711422484.9A CN108133468B (en) 2017-12-25 2017-12-25 Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection

Publications (2)

Publication Number Publication Date
CN108133468A CN108133468A (en) 2018-06-08
CN108133468B true CN108133468B (en) 2021-09-03

Family

ID=62392464

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711422484.9A Active CN108133468B (en) 2017-12-25 2017-12-25 Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection

Country Status (1)

Country Link
CN (1) CN108133468B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110889843B (en) * 2019-11-29 2023-04-18 西安电子科技大学 SAR image ship target detection method based on maximum stable extremal region
CN111027511B (en) * 2019-12-23 2022-04-29 西安电子科技大学 Remote sensing image ship detection method based on region of interest block extraction
CN111369507B (en) * 2020-02-27 2023-08-11 杭州电子科技大学 Trail detection method based on normalized gray scale Hough transform and local CFAR
CN113256720B (en) * 2021-06-03 2021-09-24 浙江大学 Method for simultaneously detecting SAR image ship and trail thereof
CN113391274B (en) * 2021-06-15 2022-08-05 电子科技大学 Low-altitude target optimal search method for airborne phased array radar

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1920594A (en) * 2006-05-16 2007-02-28 中国科学院研究生院 Wavelet reinforcement based compound aperture radar image method for detecting ship object
US20070139251A1 (en) * 2005-12-15 2007-06-21 Raytheon Company Target recognition system and method with unknown target rejection
CN106646469A (en) * 2016-12-21 2017-05-10 中国科学院遥感与数字地球研究所 SAR (Synthetic Aperture Radar) ship detection optimization method based on variation coefficient method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070139251A1 (en) * 2005-12-15 2007-06-21 Raytheon Company Target recognition system and method with unknown target rejection
CN1920594A (en) * 2006-05-16 2007-02-28 中国科学院研究生院 Wavelet reinforcement based compound aperture radar image method for detecting ship object
CN106646469A (en) * 2016-12-21 2017-05-10 中国科学院遥感与数字地球研究所 SAR (Synthetic Aperture Radar) ship detection optimization method based on variation coefficient method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Inshore Ship Detection with High-Resolution SAR Data Using Salience Map and Kernel Density;Wei Liu 等;《Eighth International Conference on Digital Image Processing》;20160829;正文第2-6页 *
Wei Liu 等.Inshore Ship Detection with High-Resolution SAR Data Using Salience Map and Kernel Density.《Eighth International Conference on Digital Image Processing》.2016,正文第2-6页. *
红外面阵旋转搜索系统下的恒虚警检测方法;费小亮 等;《红外与激光工程》;20160228;第45卷(第2期);第0204003-1 - 0204003-7页 *

Also Published As

Publication number Publication date
CN108133468A (en) 2018-06-08

Similar Documents

Publication Publication Date Title
CN108133468B (en) Constant false alarm rate ship detection method based on adaptive parameter enhancement and trail auxiliary detection
Li et al. An improved superpixel-level CFAR detection method for ship targets in high-resolution SAR images
Xia et al. Infrared small target detection based on multiscale local contrast measure using local energy factor
Tao et al. A segmentation-based CFAR detection algorithm using truncated statistics
Li et al. Automatic detection of ship targets based on wavelet transform for HF surface wavelet radar
Williams Fast unsupervised seafloor characterization in sonar imagery using lacunarity
CN106646469B (en) SAR ship detection optimization method based on VC Method
CN109117802B (en) Ship detection method for large-scene high-resolution remote sensing image
CN103971127B (en) Forward-looking radar imaging sea-surface target key point detection and recognition method
CN112766221B (en) Ship direction and position multitasking-based SAR image ship target detection method
CN111476159A (en) Method and device for training and detecting detection model based on double-angle regression
Jarabo-Amores et al. Spatial-range mean-shift filtering and segmentation applied to SAR images
CN110889843B (en) SAR image ship target detection method based on maximum stable extremal region
CN110765912B (en) SAR image ship target detection method based on statistical constraint and Mask R-CNN
CN107274410B (en) Self-adaptive artificial target constant false alarm rate detection method
CN107784655A (en) A kind of visual attention model SAR naval vessels detection algorithm of adaptive threshold
CN114764801A (en) Weak and small ship target fusion detection method and device based on multi-vision significant features
Yang et al. Evaluation and mitigation of rain effect on wave direction and period estimation from X-band marine radar images
CN106803078A (en) A kind of SAR image Ship Target dividing method
Andriyanov et al. Pattern recognition on radar images using augmentation
CN112101250B (en) Method for detecting offshore ship target based on context semantic perception
Meng et al. A novel ship CFAR detection algorithm based on adaptive parameter enhancement and wake-aided detection in SAR images
CN107369163A (en) A kind of quick SAR image object detection method based on best entropy Double Thresholding Segmentation
CN105551029A (en) Multi-spectral remote sensing image-based ship detection method
CN115564683A (en) Ship detection-oriented panchromatic remote sensing image self-adaptive enhancement method

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