CN111127506B - A Comprehensive Detection Method of Maritime Moving Targets Based on Sequence Images - Google Patents

Abstract

The invention relates to a comprehensive detection method for moving targets on the sea based on sequential images, which belongs to the field of image processing and application of remote sensing satellites; step 1, photographing the sea surface; and judging whether there is a cloud layer in a single frame image; step 2, setting the cloud layer Gradient detection function Δ(δ); when Δ(δ) is 1, enter step 3; when Δ(δ) is 0, enter step 4; step 3, perform cloud removal, hole filling and target enhancement on the image in sequence Processing; enter step five; step four, take multi-frame continuous image shooting, use clustering method to determine the target white point, and perform cloud removal and target enhancement processing on the rest of the interfering target points; step five, when the target white point and trail exist at the same time When the detection of the target ship and the determination of the direction of motion of the target ship are completed, the speed of the target ship is calculated; the present invention judges the direction of motion and the state of motion of the target through multi-frame detection, and solves the problem of most of the sea clouds on the false alarm of target detection. influence and increase the probability of detection.

Description

A Comprehensive Detection Method of Maritime Moving Targets Based on Sequence Images

technical field

The invention belongs to the field of image processing and application of remote sensing satellites, and relates to a comprehensive detection method for moving targets at sea based on sequence images.

Background technique

CN201810513672.0 A moving ship detection and tracking method based on satellite sequence images. The method proposed by the invention does not propose a solution to cloud interference, but in the optical image of the sea surface, various types of cloud interference are common, and there are few chances of no clouds, so the applicability of the method is restricted.

"Journal of Electronics and Information Technology", Volume 37, Issue 8, August 2015, Rapid detection method for ships moving on sea surface by geostationary remote sensing satellites. This paper proposes a moving ship detection algorithm using sequence images, but this method does not mention the method of removing false alarms under cloudy conditions.

CN201310256096.3 Ship detection method in optical remote sensing image with cloud interference. This method takes geometric features as the main judgment basis, and is more suitable for medium-resolution (about 10m) images, but for high-orbit optical satellite images with 50m resolution, the method will be restricted due to the loss of target geometric characteristics.

"Computer Engineering and Science", Volume 32, Issue 12, 2010, a sea ship target detection method that is resistant to cloud fragmentation. The core of this method is to detect ship wakes through radon transformation to eliminate false alarms of broken clouds. However, for low-resolution high-orbit optical satellites, it is difficult to reflect ship wakes, and this method will be limited.

"Computer Engineering and Application", Volume 43, Issue 14, 2007, a novel real-time detection method for moving targets at sea. A method for target detection using imaging sequences in the visible light range is proposed, using the time difference of deformation The method realizes fast anti-interference target detection. Although this method has good real-time performance, it is not effective in processing cloud and fog interference images containing large reflective areas.

"Telecommunications Technology", Vol. 48, No. 1, 2008, A ship detection method based on wavelet direction filtering in remote sensing images with clouds. A method of fusion of image wavelet decomposition and directional filtering for detection direction is proposed to eliminate cloud interference, and finally realize target detection in remote sensing images with cloud interference, but it is not suitable for broken cloud false alarms remove.

"Journal of Northeast Normal University: Natural Science Edition", Volume 41, No. 2, June 2009, "Research on Weak Target Detection Algorithm Based on OTSU Segmentation in Cloud Background". It is proposed to use the maximum inter-class variance OTSU segmentation algorithm to remove the cloud interference in the background, which is easy to calculate. However, this method is not suitable for the elimination of broken cloud false alarms.

Contents of the invention

The technical problem solved by the present invention is: to overcome the deficiencies of the prior art, to propose a comprehensive detection method for moving targets at sea based on sequence images, to judge the direction and state of motion of the target through multi-frame detection, and to solve the problem of most sea clouds affecting the target. Detect the impact of false alarms and increase the probability of detection.

The technical solution of the present invention is:

A method for comprehensive detection of moving targets on the sea based on sequential images, comprising the following steps:

Step 1, the sea surface is photographed by a high-orbit optical remote sensing satellite; and whether there is a cloud layer in a single frame image is judged; when there is a cloud layer, enter step 2; when there is no cloud layer, enter step 5;

Step 2, setting the cloud layer gradient detection function Δ (δ); according to the value of the cloud layer gradient detection function Δ (δ), the cloud layer situation is judged; when the cloud layer gradient detection function Δ (δ) is 1, enter step three; when the cloud layer When the gradient detection function Δ(δ) is 0, go to step 4;

Step 3: Perform cloud removal, hole filling, and target enhancement processing on the image in turn; realize cloud layer weakening; enter step 5;

Step 4: Carry out multi-frame continuous image shooting, use the clustering method to judge the movement direction of all target points according to the multi-frame images; determine the target white point, and perform cloud removal and target enhancement processing on the rest of the interfering target points; enter step 5 ;

Step 5. Determine whether there are target white spots and wakes; when the target white spots and wakes exist at the same time, complete the detection of the target ship and the determination of the direction of movement of the target ship, and calculate the speed of the target ship; otherwise, return to step 1.

In above-mentioned a kind of comprehensive detection method of moving target on the sea based on sequence image, in described step 2, cloud layer gradient detection function Δ (δ) is:

Δ(δ)=Δ(X+)·Δ(X-)·Δ(Y+)·Δ(Y-)

In the formula, X is the x-direction coordinate of the detected pixel in the image;

Y is the y-direction coordinate of the detected pixel in the image;

Δ(X+) and Δ(X-) are the image gray gradient values of the pixel at two adjacent positions in the x direction;

Δ(Y+) and Δ(Y-) are the image gray gradient values of the pixel at two adjacent positions in the y direction;

The calculation method of the cloud layer gradient detection function Δ(δ) is:

Set the threshold a according to experience; when the cloud layer gradient detection function Δ(δ) is greater than a, set Δ(δ)=1; otherwise Δ(δ)=0.

In the above-mentioned a kind of comprehensive detection method for moving targets on the sea based on sequence images, in said step 2, when the cloud layer gradient detection function Δ (δ) is 1, the cloud layer is a continuous distribution, and there is no cloud layer to target white spots and trails Interference term; when the cloud layer gradient detection function Δ(δ) is 0, the cloud layer is distributed as broken clouds, and the broken clouds are the interference target points of the target white point and wake.

In the above-mentioned a kind of comprehensive detection method for moving targets on the sea based on sequential images, in the step 4, the specific method for determining the target white point is:

According to multiple frames of images, judge the direction of movement of all target points; when one of the target points is inconsistent with the direction of movement of other target points, the target point is the target white point; other target points are interference target points formed by broken clouds.

In the above-mentioned a kind of comprehensive detection method for moving targets on the sea based on sequential images, in the step five; the method for judging the wake is:

S1. When there is a target white point and a sharp wave-shaped wake, and the tip of the wake points to the target white point; the target white point is the target ship; the pointing direction is the direction of movement of the target ship;

S2. When there is a target white point and a linear wake, and one end of the linear wake points to the target white point, the target white point is the target ship; the direction along the line-shaped wake pointing to the target white point is the target ship’s movement direction .

In the aforementioned comprehensive detection method for moving targets at sea based on sequential images, in S1, when the wake is in the shape of a sharp wave, the included angle of the sharp wave is 32°-39°, and the trajectory is determined to be valid.

In the aforementioned method for comprehensive detection of moving targets at sea based on sequential images, in S2, when the wake is linear and the length of the wake is more than three times the length of the target white spot, the track is judged to be valid.

In the above-mentioned comprehensive detection method for moving targets at sea based on sequential images, the calculation method of the speed of the target ship is: according to the continuous multi-frame images, performing correlation calculation to obtain the speed of the target ship.

The beneficial effect of the present invention compared with prior art is:

(1) The present invention adopts a brand-new processing flow to meet the needs of ship target detection in high-orbit optical satellite sequence images. The sequence images of high-orbit optical satellites are not video images, and the frame rate is as low as 3 to 5 minutes per frame. In order to quickly obtain the moving state of the target, generally only 5 to 10 frames are continuously obtained, which cannot provide a large number of samples. Traditional video image moving target detection methods Cannot be used effectively. The high time resolution characteristics of high-orbit optical satellites help to realize continuous tracking and monitoring of moving targets on the sea surface. Through the effective processing of sequence images in the same area, the detection of moving targets on the sea surface can be realized;

(2) The present invention improves the detection rate of ship targets and reduces the false alarm rate caused by interference such as broken clouds. Due to the distance from the earth and the poor resolution of high-orbit satellites, the state of moving targets on the sea surface is similar to that of broken clouds and reefs, and the false alarm rate of ordinary detection algorithms is high. This comprehensive processing flow combining static and dynamic effectively reduces the occurrence of such false alarms;

(3) The present invention integrates multiple detection methods such as single-frame image ship wake detection, sequence image target detection under reduced cloud interference, sequence image moving target detection, and the like. The wake detection can effectively confirm the moving ship target, the sequence image gradient change detection + clustering method can effectively reduce the influence of clouds and broken clouds, and the sequence image moving target detection can estimate and predict the state of the moving target;

(4) The present invention uses the method to process high-orbit satellite sequence images, which can be used for ship target detection in important areas, and provide important reference data for industries such as national defense construction, shipping management, and fishery management.

Description of drawings

Fig. 1 is a flow chart of comprehensive detection of moving targets at sea in the present invention;

Fig. 2 is the schematic diagram of the moving ship target form of the linear wake of the present invention;

Fig. 3 is a schematic diagram of the moving ship target form of the Kelvin wake of the present invention.

Detailed ways

The present invention will be further elaborated below in conjunction with embodiment.

The invention provides a comprehensive detection method for moving targets on the sea oriented to high-orbit optical satellite sequence images. This method takes into account the data characteristics of high-orbit optical satellite sequence images, combines relatively mature static image ship detection means and image processing technology, and adopts a static-dynamic combination method to improve the detection probability of moving targets on the sea surface. When , the cloud area is classified and processed, and the false alarm rate is reduced. The method is aimed at the sequence images that have been preprocessed, and the preprocessing content includes: geometric correction of each frame image, radiation correction and pixel-level registration between sequence images, sea and land segmentation, and sea island segmentation. Among them, in addition to correcting system errors, radiometric correction also needs to consider the dynamic range adjustment problem in the ocean background; pixel-level registration mainly considers eliminating imaging position errors caused by orbital perturbation and attitude jitter during the imaging process.

As shown in Figure 1, the comprehensive detection method for maritime moving targets based on sequence images mainly includes the following steps:

Step 1. Shoot the sea surface through high-orbit optical remote sensing satellites; and judge whether there are clouds in a single frame image; when there are clouds, enter step 2; when there are no clouds, enter step 5; because of high-orbit optical remote sensing satellites, other types The trail of the target is not obvious, but through trail detection, the moving direction of the moving target can be directly locked with only a single frame image, and then searched along the moving direction of the target in the sequence of images, the detection of the moving target and the estimation of the moving state can be quickly completed.

Step 2. Sea surface moving target detection without cloud layer interference. For areas covered by clouds, it is impossible to directly detect them. It is necessary to classify the clouds first, and then process them separately according to the characteristics of the clouds and targets. The specific process is as follows: firstly, classify the cloud cover characteristics. The area with continuous gradient change and no target bulge in the cloud layer is the continuous thick cloud cover and no suspected target area. The area is directly treated with cloud removal and hole filling without further processing. Target detection operation; for areas with continuous gradient changes but with target bulges as continuous thick clouds and suspected target areas, target enhancement and cloud layer suppression are used for this area; for multiple fragmented cloud areas, multi-frame images are used, and clustering is used. method, to identify the overall movement direction of the broken cloud, and select the target point inconsistent with the overall movement direction of the cloud as the suspected target; for the area of a single piece of broken cloud (or an independent suspected moving ship target), use multiple frames of images to detect its shape , Whether the gray level changes, to rule out the suspected possibility. Set the cloud layer gradient detection function Δ (δ); according to the value of the cloud layer gradient detection function Δ (δ), the cloud situation is judged; when the cloud layer gradient detection function Δ (δ) is 1, enter step three; when the cloud layer gradient detection function When Δ(δ) is 0, go to step 4; the cloud layer gradient detection function Δ(δ) is:

Δ(δ)=Δ(X+)·Δ(X-)·Δ(Y+)·Δ(Y-)

In the formula, X is the x-direction coordinate of the detected pixel in the image;

Y is the y-direction coordinate of the detected pixel in the image;

Δ(X+) and Δ(X-) are the image gray gradient values of the pixel at two adjacent positions in the x direction;

Δ(Y+) and Δ(Y-) are the image gray gradient values of the pixel at two adjacent positions in the y direction;

The calculation method of the cloud layer gradient detection function Δ(δ) is:

The detection threshold of Δ(δ) can be determined by calculation or machine learning. If the calculation method is adopted, a certain numerical calculation method is used to determine the threshold value. For example, the threshold value is defined as the fourth power of the mean value of the gray gradient matrix of the image in the selected area; Gentle training samples, learning support vector machine model: In the test phase, use the model and the gradient features of the current sample to determine the category as a sharp/smooth change, and the change is sharply higher than the threshold; the gentle change is lower than the threshold. Set the threshold a; when the cloud layer gradient detection function Δ(δ) is greater than a, set Δ(δ)=1; otherwise Δ(δ)=0. When the cloud layer gradient detection function Δ(δ) is 1, the cloud layer is a continuous distribution, and there is no interference item to the target white point and trail in the cloud layer; when the cloud layer gradient detection function Δ(δ) is 0, the cloud layer is a fragmented cloud distribution, The broken cloud is the interference target point of the target white point and trail.

Step 3: Perform cloud removal, hole filling, and target enhancement processing on the image in turn; realize cloud layer weakening; enter step 5;

Step 4. Carry out multi-frame continuous image shooting, and use clustering method to judge the direction of motion of all target points according to the multi-frame images; determine the target white point, and the specific method for determining the target white point is: according to the multi-frame images, judge all The direction of motion of the target point is judged; when the direction of motion of one of the target points is inconsistent with other target points; then the target point is the target white point; If the motion direction between image frames is stable, the motion trajectory has no bending angle greater than 90°), and the brightness of the target does not change significantly in the image, then it can be confirmed that the suspected target is a moving ship target, and the track is marked. Other target points are interference target points formed by broken clouds. Perform cloud removal and target enhancement processing on the remaining interference target points; enter step five;

Step 5. Judging whether there are target white spots and trails; the method of judging the trails is:

S1. When there is a target white point and a sharp wave-shaped wake, and the tip of the wake points to the target white point; the target white point is the target ship; the direction the point points to is the direction of movement of the target ship; when the wake is a sharp wave In the shape, the included angle of the sharp wave is 32°-39°, and the trajectory is judged to be valid, as shown in Figure 3.

S2. When there is a target white point and a linear wake, and one end of the linear wake points to the target white point, the target white point is the target ship; the direction along the line-shaped wake pointing to the target white point is the target ship’s movement direction . When the trail is in a linear shape and the length of the trail is more than three times the length of the target white point, the trail is judged to be valid, as shown in Figure 2. When the target white spot and wake exist at the same time, complete the detection of the target ship and the determination of the direction of movement of the target ship, and calculate the speed of the target ship; the calculation method of the target ship speed is: according to the continuous multi-frame image to carry out the correlation calculation , to obtain the speed of the target ship. Otherwise return to step one.

Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention, and any person skilled in the art can use the methods disclosed above and technical content to analyze the present invention without departing from the spirit and scope of the present invention. Possible changes and modifications are made in the technical solution. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention, which do not depart from the content of the technical solution of the present invention, all belong to the technical solution of the present invention. protected range.

Claims (7)

1. A comprehensive detection method of an offshore moving target based on a sequence image is characterized in that: the method comprises the following steps:

shooting the sea surface through a high-orbit optical remote sensing satellite; judging whether a cloud layer exists in the single-frame image or not; entering a second step when a cloud layer exists; entering a fifth step when the cloud layer does not exist;

step two, setting a cloud layer gradient detection function delta (delta); judging the cloud layer condition according to the value of a cloud layer gradient detection function delta (delta); when the cloud layer gradient detection function delta (delta) is 1, entering a step III; when the cloud layer gradient detection function delta (delta) is 0, entering a step four;

the cloud gradient detection function delta (delta) is:

Δ(δ)＝Δ(X+)·Δ(X-)·Δ(Y+)·Δ(Y-)

wherein X is the X-direction coordinate of the detected pixel in the image;

y is the Y-direction coordinate of the detected pixel in the image;

delta (X+) and delta (X-) are the image gray scale gradient values of the pixel at two adjacent positions in the X direction;

delta (Y+) and delta (Y-) are the image gray scale gradient values of the pixel at two adjacent positions in the Y direction;

the cloud layer gradient detection function delta (delta) is calculated by the following steps:

setting a threshold value a according to experience; let Δ (δ) =1 when the cloud gradient detection function Δ (δ) is greater than a; otherwise Δ (δ) =0;

sequentially carrying out cloud removal, hole filling and target enhancement treatment on the image; realizing the weakening of cloud layer; step five, entering a step five;

fourthly, shooting a plurality of frames of continuous images, and judging the motion directions of all target points according to the plurality of frames of images by adopting a clustering method; determining a target white point, and carrying out cloud removal and target enhancement treatment on the rest interference target points; step five, entering a step five;

step five, judging whether a target white point and a trail exist; when the target white point and the wake are simultaneously present, the detection of the target ship and the determination of the movement direction of the target ship are completed, and the speed of the target ship is calculated; otherwise, returning to the step one.

2. The method for comprehensively detecting the marine moving target based on the sequence image according to claim 1, wherein the method comprises the following steps: in the second step, when the cloud layer gradient detection function delta (delta) is 1, the cloud layer is continuously distributed, and the cloud layer does not have interference items on target white points and trails; when the cloud layer gradient detection function delta (delta) is 0, cloud layers are distributed in a cloud crushing mode, and the cloud crushing mode is an interference target point of target white points and trails.

3. The method for comprehensively detecting the marine moving target based on the sequence image according to claim 2, wherein the method comprises the following steps: in the fourth step, the specific method for determining the target white point is as follows:

judging the motion directions of all target points according to the multi-frame images; when the motion direction of one target point is inconsistent with that of the other target points; the target point is a target white point; the other target points are interference target points formed by the broken cloud.

4. The method for comprehensively detecting the marine moving target based on the sequence image according to claim 3, wherein the method comprises the following steps of: in the fifth step; the judgment method for the trail comprises the following steps:

s1, when a target white point and a wake in a spike shape exist and a wake tip points to the target white point; the target white point is the target ship; the pointed direction is the movement direction of the target ship;

s2, when a target white point and a linear-shaped tail trace exist and one end of the linear tail trace points to the target white point, the target white point is the target ship; the direction pointing to the target white point along the linear wake is the movement direction of the target ship.

5. The method for comprehensively detecting the marine moving target based on the sequence image according to claim 4, wherein the method comprises the following steps: in the step S1, when the wake is in a sharp wave shape, the included angle of the sharp wave is 32-39 degrees, and the track is judged to be effective.

6. The method for comprehensively detecting the marine moving target based on the sequence image according to claim 5, wherein the method comprises the following steps: in the step S2, when the trail is in a linear shape, the trail length is 3 times or more the target white point length, and the trail is judged to be valid.

7. The method for comprehensively detecting the marine moving target based on the sequence image according to claim 6, wherein the method comprises the following steps: the calculation method of the target ship speed comprises the following steps: and (5) performing shape-advancing association calculation according to the continuous multi-frame images to obtain the speed of the target ship.

Images