A sea surface ship target rapid detection method based on remote sensing image

Abstract

The application relates to a kind of sea surface ship target fast detection methods based on remote sensing image, including through inorganic person shoots sea surface image of different height position, and image information is introduced into image processing system;Image processing system obtains image information after through two ways to sea surface ship image of different height shooting, through edge detection mode, ship and other objects are distinguished;Adopt the minimum circumscribed rectangle to frame the ship connected body, to read its length and width.Specifically, draw rectangle, carry out rectangular mark sequence, set rectangle length and width, display rectangle number.The sea surface ship target fast detection method based on remote sensing image of the application is based on remote sensing image, adopts threshold method, edge detection method, mathematical morphology and other methods, can guarantee under the premise of certain identification accuracy, quickly and effectively identify the target different from background in a short time, so as to realize the fast detection of water surface ship.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of earth observation, and particularly relates to a sea surface ship target rapid detection method based on remote sensing images. BACKGROUND

[0002] In recent years, with the rapid development of earth observation technology, a large number of high spatial resolution optical remote sensing imaging satellites have emerged, and sub-meter resolution panchromatic images can be obtained. Unmanned aerial vehicles and other aerial images can achieve near-ground high-definition target acquisition. Space and aerial remote sensing provides an extremely rich data source for sea target detection and identification, and the application range of remote sensing images in military and civilian fields is becoming increasingly widespread.

[0003] As an important target for sea monitoring and wartime attack, the detection and identification of ships can monitor the distribution of ships in key sea areas, analyze the combat strength of the enemy, master sea combat intelligence, and carry out precision guidance. In addition, it can also meet the needs of actual work such as sea traffic supervision, sea search and rescue, and anti-smuggling. However, due to the long shooting distance, cloud cover, haze shielding, uneven illumination, low contrast between ship and sea surface brightness and color, sea clutter, and island interference, it is easy to cause false alarms and missed detections in detection. How to accurately, quickly and stably detect and extract sea surface ship targets to gain more reaction and processing time has become an urgent problem to be solved.

[0004] In the early days, due to the limitations of hardware technologies such as sensors, the spatial resolution of visible light remote sensing imaging was low, and ship targets were often distributed in the form of spots. The spot-shaped ship targets occupy a small number of pixels in the entire image, generally tens of pixels, and contain less ship target detail information. The detection result of the spot target is limited to judging the presence or absence of the target and positioning. With the improvement of image spatial resolution, the ship targets in visible light remote sensing images have more characteristics, and rich image information is more convenient for confirming targets and analyzing their behavior. It also plays a crucial role in evaluating sea conditions and target trends and formulating effective monitoring arrangements.

[0005] Visible light remote sensing images can more realistically reflect the shape, color and texture characteristics of ship targets, providing rich information when detecting targets, but also causing many background interference problems and increasing the workload. Current ship detection methods for ship targets can be summarized as methods based on sea surface gray scale distribution, methods based on machine learning, methods based on deep learning, methods based on target model expression, methods based on sparse representation and methods based on visual saliency. Among them, deep learning and machine learning require more prior knowledge, making it difficult to quickly identify unknown water vessels, and their requirements for software and hardware are high, which cannot be universally applicable.

[0006] Therefore, it is urgent to develop a method for high-precision, rapid target recognition and rapid detection of water surface ships. SUMMARY

[0007] The purpose of the present application is to provide a method for rapid detection of sea surface ship targets based on remote sensing images, using threshold method, edge detection method, mathematical morphology and other methods, to solve the problem of high-precision, rapid and effective identification of targets different from the background, and to realize the rapid detection of sea surface ship targets.

[0008] The purpose of the present application is achieved by the following technical solutions:

[0009] A sea surface ship target rapid detection method based on remote sensing images, comprising the following steps:

[0010] A. Take sea surface images at different height positions by an unmanned aerial vehicle, and import the image information into an image processing system;

[0011] B. After the image processing system obtains the image information, it distinguishes the ships from other objects through edge detection in two ways.

[0012] C. The minimum circumscribed rectangle is used to frame the connected body of the ship to read its length and width. Specifically, a rectangle is drawn, the rectangle is labeled, the length and width of the rectangle are set, and the number of rectangles is displayed.

[0013] Further, step B, the first way comprises the following steps:

[0014] B1. After the image processing system obtains the image information, it reduces the resolution to remove noise and smoothness;

[0015] B2. The edges in the image are extracted by the canny operator, which is the best edge detection operator at present, and the broken parts of the ship are connected to form a connected body through closing operation;

[0016] B3. Get the information of all connected bodies in the graph, distinguish the ship from other objects according to the roundness, establish the minimum circumscribed convex polygon for the connected body, set the filling degree threshold to identify and remove noise.

[0017] Further, step B1, the program running time is set to 4.628s, 3.943s, 3.254s, 6.604s, 6.132s, 5.716s, the resolution reduction coefficient is set to 0.35 when the unmanned aerial vehicle is shooting at a height of 50 meters or less; the resolution reduction coefficient is set to 0.45 when the unmanned aerial vehicle is shooting at a height of 100 meters or less; the resolution reduction coefficient is set to 0.5 when the unmanned aerial vehicle is shooting at a height of 150 meters or less; the resolution reduction coefficient is set to 0.6 when the unmanned aerial vehicle is shooting at a height of 200 meters or less; the resolution reduction coefficient is set to 0.7 when the unmanned aerial vehicle is shooting at a height of 200 meters or more.

[0018] Further, in step B2, the edge detection is followed by a closed operation structure element of 50*50.

[0019] Further, in step B3, the roundness=4*S*Pi / C^2, and the roundness threshold is 0.85.

[0020] Further, in step B3, the filling degree=connected body area / minimum circumscribed convex polygon area, and the filling degree threshold is set to 0.5.

[0021] Further, in step B, the second mode comprises the following steps:

[0022] B1, after the image processing system acquires the image information, a two-color reflection model is used to remove the reflection area in the sea surface image;

[0023] B2, the edges in the image are extracted by a canny operator, which is the optimal edge detection operator at present, the broken parts of the ship are connected by a closed operation to form a connected body, and then the ship connected body is subjected to aperture filling;

[0024] B3, different thresholds are set for images taken at different heights, and connected bodies smaller than a certain area are removed, and the difference in area is used for denoising.

[0025] Further, in step B2, the edge detection is followed by a closed operation structure element of 50*50.

[0026] Further, in step B3, the values of the small connected bodies removed in images taken at different heights are different, when the shooting height is 50m, the size of the small connected bodies removed is 8500px, when the shooting height is 100m, the size of the small connected bodies removed is 8500px, when the shooting height is 150m, the size of the small connected bodies removed is 6000px, when the shooting height is 200m, the size of the small connected bodies removed is 6000px, when the shooting height is 250m, the size of the small connected bodies removed is 1000px, when the shooting height is 300m, the size of the small connected bodies removed is 800px, when the shooting height is 350m, the size of the small connected bodies removed is 300px, when the shooting height is 400m, the size of the small connected bodies removed is 450px, when the shooting height is 450m, the size of the small connected bodies removed is 350px, when the shooting height is 500m, the size of the small connected bodies removed is 100px.

[0027] Compared with the prior art, the present application has the following beneficial effects:

[0028] The application discloses a sea surface ship target rapid detection method based on remote sensing images. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows, and it should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the premise of the drawings.

[0030] Figure 1 The flow chart of the sea surface ship target rapid detection method based on remote sensing images;

[0031] Figures 2-6 Step B adopts the first way experimental result;

[0032] Figures 7-12 Step B adopts the second way experimental result. DETAILED DESCRIPTION

[0033] The present application will be further described below in combination with the embodiments:

[0034] The present application will be further described below in combination with the embodiments:

[0035] It should be noted that: similar labels and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second" and the like are only used for distinguishing description, and cannot be understood as indicating or implying relative importance.

[0036] As shown in the drawings, the sea surface ship target rapid detection method based on remote sensing images comprises the following steps: Figure 1

[0037] Step one, the sea surface images in different height positions are shot by inorganic people, and the image information is artificially introduced into the image processing system;

[0038] Step two, after the image processing system acquires the image information, the sea surface ship images shot in different heights are distinguished from other objects through the edge detection method.​

[0039] Step three, the minimum circumscribed rectangle is used to frame the ship connected body to read its length and width. Specifically, a rectangle is drawn, the rectangle is marked, the rectangle length and width are set, and the number of rectangles is displayed.

[0040] In step two, the following two methods are included:

[0041] The first method is to combine the methods of reducing resolution, canny operator, roundness and filling degree.

[0042] 1. After the image processing system obtains the image information, the sea surface ship images taken at different heights are denoised and smoothed by reducing the resolution. Specifically, different parameters are set at different heights, the resolution is reduced, and the image is converted to a gray image.

[0043] Generally, the filter processing does not change the total number of image elements, so the information of the original image is retained after processing. The method of reducing the resolution more thoroughly reduces the retention of information on the original image, thereby reducing the noise and the picture size, and achieving the effect of rapid identification. The best resolution reduction multiple of the images taken at different heights is different.

[0044] After multiple tests, the image shooting height of the application is set to 50m, 100m, 150m, 200m, 250m, 300m; the resolution reduction coefficient, i.e. the ratio of the resolution after reduction to the resolution before reduction, is set to 0.35, 0.45, 0.5, 0.6, 0.7; T is the program running time, which is set to 4.628s, 3.943s, 3.254s, 6.604s, 6.132s, 5.716s. Specifically, for 50 meters or less, the resolution reduction coefficient is set to 0.35; for 100 meters or less, the resolution reduction coefficient is 0.45; for 150 meters or less, the resolution reduction coefficient is 0.5; for 200 meters or less, the resolution reduction coefficient is 0.6; for 200 meters or more, the resolution reduction coefficient is 0.7.

[0045] 2. The edges in the image are extracted by the canny operator. The canny operator is the most optimal edge detection operator at present, which realizes edge extraction by filtering and denoising processing, difference calculation amplitude and direction, non-maximum suppression, and double-threshold edge determination, and connects the broken parts of the ship into a connected body by closing operation. After edge detection, the closing operation structure element is 50*50.

[0046] 3. The information of all connected bodies in the graph is obtained, the ship and other objects are distinguished according to the roundness, the minimum circumscribed convex polygon of the connected body is established, and the filling degree threshold is set to identify and remove noise.

[0047] Specifically, the roundness of the sea surface buoy, circular island and the like is high, while the roundness of the ship is low, so the ship and other objects can be distinguished according to the roundness. In the present application, the roundness = 4*S*Pi / C^2, and through multiple tests, the roundness of most ships is lower than 0.85, so the roundness threshold is set to 0.85.

[0048] Since the normally recognized ship is a convex polygon, and the noise shape is irregular, most of them are irregular polygons. The minimum circumscribed convex polygon is established for the connected body, and the better fitting is the ship, and the poor fitting is the noise. The filling degree = connected body area / minimum circumscribed convex polygon area, and through multiple tests, the threshold is set to 0.5.

[0049] Using the method, all ships are detected, the edge is well preserved, the noise is completely removed, and the time is short

[0050] It is found that the higher the flight height, the larger the image coverage, and the more diverse the types of ships (the greater the size difference), and the size of small ships and part of the noise is similar in the image, which is easy to be removed or retained at the same time. Therefore, the detection accuracy is also reduced.

[0051] As shown in Figures 2-6 , step B uses the first way of experimental results, and the image shooting height is 50m, 100m, 150m, 200m, 250m, 300m in turn, and the number of recognized ships is 2, 3, 9, 12, 13, 13 in turn. The aspect ratio of the ship is the ratio of the length to the width (in the experiment, the aspect ratio of the minimum circumscribed rectangle is used instead), which can be used to reflect the characteristic information of the ship.

[0052] The second way is to remove large reflective light, canny operator and remove small connected body.

[0053] 1. After the image processing system obtains the image information, the double-color reflection model is used to remove the reflective light area in the sea surface image. Due to the low shooting height of the unmanned aerial vehicle and the shooting angle, there is a large area of reflective light in part of the image. If it is not removed, the canny operator will identify it, so the double-color reflection model is used to remove it. The principle is as follows:

[0054] 1) I(x) = I D (x) + I S (x), that is, the pixel point brightness value I(x) = diffuse reflection component I D (x) + specular reflection component I S (x).

[0055] 2) The ratio of the maximum RGB channel value of a certain pixel point to the RGB channel range value of the point is defined as the brightness value Q(x).

[0056] The research found that the brightness value of a completely diffuse reflection pixel point is irrelevant to the surface geometry of the object.

[0057] I(x)max = max(I(x)r, I(x)g, I(x)b);

[0058] I(x)min = min(I(x)r, I(x)g, I(x)b);

[0059] Q(x) = I(x)max / (I(x)max-I(x)min);

[0060] 3) If the sea surface point is a completely diffuse reflection pixel point, its Q(x) is similar, and at the same time, if a sea surface point is a non-completely diffuse reflection pixel point, that is, there is a specular reflection component in addition to the diffuse reflection component, its Q(x) will be larger. If we set a threshold T p , we can distinguish them.

[0061] This threshold should be set to different values according to different ground objects, but since we are dealing with specific objects such as the ocean, ships, and islands, we can try to set it to an empirical value φ.

[0062] 4) I S (x) = max(I(x)max-φ×(I(x)max-I(x)min), 0);

[0063] The specular reflection amount is obtained, so the diffuse reflection component I D (x) of a pixel point is I S (x) = I(x)-I (x). If each pixel point is processed in this way, all the reflection amounts of the image can be removed.

[0064] However, since there are many high-light pixels that reach saturation in the sea surface reflection area, that is, there is at least one color component in the RGB channel that exceeds the sensing range of the light sensing element (i.e., its value exceeds 255, which can only be displayed as 255), so the image obtained by this processing is only a seemingly diffuse reflection component, which is not accurate.

[0065] 5) However, the specular reflection component of each point of the image obtained by step 4) is "accurate" for those unsaturated points of light, that is, the lower limit is guaranteed, although the upper limit is not accurate. Therefore, set an empirical threshold T z , when the specular reflection component of a certain point is greater than this threshold, it is considered to belong to the high-light area. The pixel point value less than T z is set to 0, and greater than it is set to 255, which obtains a mask image I S chuThis involves representing the location of the pixels in the highlight area, and then using dilation to expand the area. This is to include data from the transition region between the highlight and non-reflective areas, facilitating subsequent highlight value replacement processing. The mask image is named I. S dilate This detects the highlight areas. However, some other, less prominent areas (usually pixels with larger pixel values) may also be detected, potentially affecting subsequent processing. This is because most of the pixels to be detected have values ​​between 150 and 200, and if the threshold T... z If the settings are not reasonable, the hull or some noise points may be detected and their original values ​​may be changed after the next pixel value processing operation.

[0066] 6) After detection, highlight areas are removed. The principle is to use the pixel values ​​of the transition area obtained from the previous dilation, perform weighted processing, and then replace the pixel values ​​of the surrounding highlight areas. This removes the highlights, and because it's a weighted smoothing process, the change isn't abrupt, thus benefiting subsequent edge processing. The steps are as follows:

[0067] ① Obtain the grayscale image I of the original image. gray Filter out I S chu (x')>=T z And I gray (x')>=T h Points x' are the objects of the operation. h This is a settable image grayscale threshold used to exclude points with high reflective brightness but which are not highlights.

[0068] ② Find the point I within a neighborhood of size M×M, with x' as the center. gray (x') and each pixel I gray The difference between (m') and I sum .

[0069]

[0070] Among them, I gray (m') are all pairs of integers in an M×M neighborhood that satisfy I dilate (m') <T z T h Points that meet the conditions; if there are no points within this range that satisfy the conditions, i.e., I sum If M = 0, then M = M + 2, and we return to the previous step to start over. The initial value of M is 20.

[0071] ③ Calculate the pixel points I in the M×M neighborhood. gray The weight β(m′) of (m').

[0072]

[0073] ④ Calculate the I of each pixel in the M×M neighborhood. gray The weighted sum of points (m') and the weighted sum of points I for reflective pixels. gray Replace (x').

[0074]

[0075] ⑤ With the updated pixel I gray Using (x') as the center, find pixels with a specular reflection component greater than 0 within an 8-neighborhood, and use I... gray The updated value of (x') replaces these pixel values.

[0076] ⑥ Switch to the next reflective pixel x′, restore the search neighborhood range to its initial size, and return to step ②, until all pixels I have been traversed. S dilate All non-zero points within (x'), at this time the graph I gray It is an image of the sea surface after removing reflective areas.

[0077] 2. Edges are extracted from the image using the Canny operator. The Canny operator is currently the best edge detection operator. It extracts edges from the image by connecting and mapping the broken parts of the ship through a closing operation. The closing operation structuring element after edge detection is 50*50.

[0078] 3. For images taken at different heights, remove connected components smaller than a certain area. Use the porosity-filling function in Matlab to fill the pores in the hull, and use the area difference to reduce noise.

[0079] The edges extracted by the Canny operator are not all ship edges; there is also a lot of noise. The noise area created by waves and small reflections on the sea surface is smaller than that of a typical ship, so noise reduction is achieved by addressing this area difference. The value for small connected component removal varies depending on the image height. The small connected component size is determined by removing connected components with a pixel count smaller than a given value. After multiple tests, specific reference parameters are shown in Table 1.

[0080] Table 1

[0081]

[0082]

[0083] like Figures 7-12 As shown, the experimental results of step B using the second method were as follows: the image shooting heights were 50m (with islands), 50m, 200m, 250m, and 400m, and the number of identified ships were 8, 2, 2, 2, 13, and 13, respectively.

[0084] As Figure 7 shown, the small connected body size is 700, the running time is 18.497s, the large area reflection is removed, and the island is basically removed.

[0085] As Figure 8 shown, the shooting height is 50m, the small connected body size is 8500, the running time is 188.098s, and the large area reflection is basically removed.

[0086] As Figure 9 shown, the shooting height is 50m, the small connected body size is 8500, and the running time is 16.527s.

[0087] As Figure 10 shown, the shooting height is 200m, the small connected body size is 6000, and the running time is 21.464s.

[0088] As Figure 11 shown, the shooting height is 250m, the small connected body size is 1000, and the running time is 18.400s.

[0089] As Figure 12 shown, the shooting height is 400m, the small connected body size is 450, and the running time is 17.154s.

[0090] Note that the above are only preferred embodiments of the present application and the technical principles applied. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and those skilled in the art can make various obvious changes, re-adjustments and substitutions without departing from the scope of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A method for detecting a ship target on a sea surface based on a remote sensing image, characterized in that, It comprises the following steps: A. Taking images of the sea surface at different heights by an unmanned aerial vehicle and importing the image information into an image processing system; B. The image processing system distinguishes ships from other objects through edge detection after obtaining the image information; B1. The image processing system reduces the resolution to remove noise and smooth the image after obtaining the image information. The program running time is set to 4.628s, 3.943s, 3.254s, 6.604s, 6.132s, and 5.716s. The resolution reduction coefficient is set to 0.35 when the unmanned aerial vehicle is below 50 meters, 0.45 when it is below 100 meters, 0.5 when it is below 150 meters, 0.6 when it is below 200 meters, and 0.7 when it is above 200 meters; B2. The edges in the image are extracted by the Canny operator, which is the best edge detection operator at present. The broken parts of the ship are connected to form a connected body through closing operation. The edge detection closing operation structure element is 50*50; B3. The information of all connected bodies in the image is obtained. The ship and other objects are distinguished according to the circularity. The minimum circumscribed convex polygon of the connected body is established, and the noise is identified and removed by setting the filling degree threshold. The circularity = 4*S*Pi / C^2, and the circularity threshold is 0.85; C. The ship connected body is framed by the minimum circumscribed rectangle to read its length and width. Specifically, the rectangle is drawn, the rectangle is labeled, the length and width of the rectangle are set, and the number of rectangles is displayed; The filling degree = connected body area / minimum circumscribed convex polygon area, and the filling degree threshold is set to 0.

5.

2. The method according to claim 1, characterized in that: Step B, the second method comprises the following steps: B1. The image processing system removes the reflective area in the sea surface image by using the two-color reflection model after obtaining the image information; B2. The edges in the image are extracted by the Canny operator, which is the best edge detection operator at present. The broken parts of the ship are connected to form a connected body through closing operation, and then the ship connected body is filled with holes; B3. Different thresholds are set for images taken at different heights to remove connected bodies smaller than a certain area, and noise is removed by the difference in area.

3. The method according to claim 2, characterized in that: Step B2, the edge detection closing operation structure element is 50*50.

4. The method for rapid detection of ship targets on the sea surface based on remote sensing imagery according to claim 2, characterized in that: Step B3, the value of the image of the small connected component removed is different at different heights, when the shooting height is 50 m, the size of the small connected component removed is 8500 px, when the shooting height is 100 m, the size of the small connected component removed is 8500 px, when the shooting height is 150 m, the size of the small connected component removed is 6000 px, when the shooting height is 200 m, the size of the small connected component removed is 6000 px, when the shooting height is 250 m, the size of the small connected component removed is 1000 px, when the shooting height is 300 m, the size of the small connected component removed is 800 px, when the shooting height is 350 m, the size of the small connected component removed is 300 px, when the shooting height is 400 m, the size of the small connected component removed is 450 px, when the shooting height is 450 m, the size of the small connected component removed is 350 px, when the shooting height is 500 m, the size of the small connected component removed is 100 px.

Images