VIIRSDNB noctilucent fishing boat adaptive identification method based on improved gradient method

Abstract

The invention discloses a VIIRSDNB noctilucent fishing boat adaptive identification method based on an improved gradient method, and belongs to the technical field of fishing boat business monitoring. Comprising the following steps: obtaining VIIRS / DNB noctilucent remote sensing image data, and performing preprocessing; carrying out gradient detection on the preprocessed noctilucent remote sensing image data by adopting an improved gradient method to obtain a night lamplight fishing boat target; the improved gradient method is used for distinguishing strong and weak lamplight fishing boats, fusing the strong and weak lamplight fishing boats, eliminating halo pixels generated by lamplight of one or more lamplight trapping operation fishing boats through local neighborhood calculation, and finally obtaining an accurate night lamplight fishing boat target. According to the method, the strong light fishing boats and the weak light fishing boats are distinguished, the strong light fishing boats and the weak light fishing boats are extracted through the local threshold value and the global threshold value respectively, the neighborhood analysis algorithm is combined to eliminate the halo influence generated when lamplight diffuses to the periphery, finally, the accurate night lamplight fishing boat target is obtained, and the method has the advantages of being high in distinguishing capacity and high in precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fishing boat business monitoring, and particularly relates to a VIIRS_DNB night light fishing boat self-adaptive identification method based on an improved gradient method. BACKGROUND

[0002] Remote sensing technology can observe a large area of sea surface at the same time. Therefore, its role in marine monitoring is becoming more and more important. However, traditional optical remote sensing relies on the detection of sunlight reflection and cannot provide effective night data. As a new direction and front hot spot of remote sensing development, low-light remote sensing can detect weak light sources on the ground at night or dusk, making up for the limitations of traditional optical remote sensing and providing night radiation products.

[0003] Night light remote sensing data is obtained by means of a specific low-light sensor carried on a satellite or an aerial platform, based on the detection of night light radiation on the earth's surface, and can be used to observe the distribution of night light on the earth. Based on this data, different types of light sources such as population distribution, urban development, industrial production, agricultural irrigation and fishing activities can be reflected. According to the statistics of the Food and Agriculture Organization (FAO) of the United Nations, the northwest Pacific is the region with the largest fishing potential among the 15 fishing areas in the world. The northwest Pacific fishery resources are mainly composed of midwater fish, among which squid (Ommastrephes bartramii), tuna (Thunnus) and mackerel are the main commercial fishing species. These fish are mainly caught at night by using nets such as purse seine, trawl and gillnet with the help of light. The light fishing boat refers to a fishing boat that uses light to catch marine fish by taking advantage of the phototaxis of fish. The unique operation mode of the light fishing boat provides the possibility for using low-light remote sensing images to detect night remote sensing images to obtain fishing boat operation information.

[0004] The typical night light remote sensing data sources are the Day / Night Band (DNB) imagery data (VIIRS / DNB) and the Night Light Satellite data (DMSP-OLS) provided by the National Oceanic and Atmospheric Administration (NOAA). Since the 1970s, the Operational Linescan System (OLS) carried on the Defense Meteorological Satellite Program (DMSP) platform has had the ability to image under low light conditions at night. Since then, the night low light imaging data collected by the meteorological satellite sensor has been used to study the detection of light from fishing vessels. However, the DMSP / OLS has the disadvantages of coarse spatial resolution and lack of on-orbit calibration. Compared with the DMSP / OLS data, the VIIRS / DNB carried by the Suomi National Polar-orbiting Partnership (S-NPP) satellite provides global 742-meter spatial resolution data, has higher spatial and temporal resolution, lower detection limit, wider dynamic range, and solves the problem of lack of on-orbit calibration. DNB shows black in the background area where no light is detected, which is suitable for using threshold method to eliminate background noise. Compared with DMSP, VIIRS can detect more light fishing vessel features and has sufficient resolution to identify the ship light source.

[0005] Many domestic and foreign scholars have carried out algorithm research on automatic identification of night light fishing vessels based on night light remote sensing data. The algorithms are all designed based on the characteristics of the large difference between the light emitting pixels and the background in the night light image. The mainstream algorithms are the spike detection algorithm proposed by Elvidge and the gradient method proposed by colozoline. Based on these two algorithms, there are different degrees of improvement in threshold division and other aspects. Commonly used methods include: empirical threshold, maximum entropy threshold segmentation, etc. In addition, in recent years, deep learning technology has made significant progress in target detection tasks. This method extracts the features of night light fishing vessels, establishes a feature data set, and uses a large training data set to automatically learn the features, thereby improving the recognition ability of small target complex light spot patterns.

[0006] However, the current technology still faces multiple challenges: first, night light remote sensing data is affected by natural factors such as atmospheric scattering and cloud cover, which affects the recognition accuracy; second, the background light source pollution caused by the expansion of coastal or island cities and shipping activities, etc. human factors interfere with the extraction of fishing vessel light; third, the deep learning model is highly dependent on the size and quality of the data, and the required computing resources threshold of the model, resulting in high algorithm training cost and deployment difficulty. Therefore, it is urgent to develop a new night light fishing vessel automatic identification algorithm. SUMMARY

[0007] In view of defects of the prior art, the present application proposes a method for extracting strong light fishing vessels and weak light fishing vessels by using local threshold and global threshold based on VIIRS / DNB night light remote sensing data provided by the National Oceanic and Atmospheric Administration (NOAA), and has the characteristics of strong resolution and high precision in combination with a neighborhood analysis algorithm.

[0008] In order to achieve the above purpose, the present application proposes a VIIRS_DNB night light fishing vessel self-adaptive identification method based on an improved gradient method, comprising the following steps:

[0009] (1) obtaining VIIRS / DNB night light remote sensing image data;

[0010] (2) preprocessing the VIIRS / DNB night light remote sensing image data;

[0011] (3) performing gradient detection on the preprocessed night light remote sensing image data by using an improved gradient method to obtain a night light fishing vessel target; the improved gradient method is used to distinguish two types of light fishing vessels, i.e., strong light fishing vessels and weak light fishing vessels, and to perform fusion, and then to realize elimination of halos of pixels generated by single or multiple light fishing vessels through local neighborhood calculation, so as to finally obtain an accurate night light fishing vessel target;

[0012] Further, the step (3) is specifically as follows:

[0013] (3.1) performing radiation amplification on the preprocessed night light remote sensing image data, and using logarithmic transformation to perform radiation enhancement to obtain a radiation enhanced image;

[0014] (3.2) performing horizontal band lightning detection on the radiation enhanced image data, discarding pixels marked as lightning, i.e., setting the pixel value as NaN, to obtain a modified radiation enhanced image;

[0015] (3.3) performing gray value transformation on the modified radiation enhanced image data to make it adapt to a gray scale [0-255] to obtain a gray image;

[0016] (3.4) performing gradient calculation on the gray image data, taking the maximum value as the gradient value of the pixel by subtracting the center pixel from eight surrounding pixels;

[0017] Comparing the gradient value of each pixel with a set gradient threshold value, removing pixels smaller than the set gradient threshold value, i.e., setting the pixel value as NaN, to obtain a gradient image;

[0018] (3.5) calculating the gradient threshold value:

[0019] (3.5.1) Calculate the gradient mean value gradient_mean of the acquired gradient image, and extract the gray image data corresponding to all gradient values exceeding the gradient_mean as the candidate threshold vector;

[0020] (3.5.2) According to the candidate threshold vector, calculate the local threshold, which is calculated locally by using the first sliding window to obtain the local threshold of each window;

[0021]

[0022] Where: i represents the i-th sliding window, min i and max i represent the minimum and maximum values of the candidate threshold vector in the i-th sliding window; mean i is the average value of the candidate threshold vector in the i-th sliding window, threshold local，i represents the local threshold of the i-th sliding window;

[0023] (3.5.3) According to the candidate threshold vector, calculate the global threshold;

[0024]

[0025] Where: min and max are the minimum and maximum values of all the candidate threshold vectors, median is the median of all the candidate threshold vectors; threshold global is the global threshold;

[0026] (3.6) For the gray image data in each first sliding window in the gray image, keep the points higher than the local threshold, and set the other points to NaN; then for the entire gray image, keep the points higher than the global threshold, and set the other points to NaN, to obtain the modified gray image;

[0027] (3.7) The modified gray image at this time is the gray image of the strong light fishing boat. Set the valid value points in the gray image of the strong light fishing boat to NaN, and obtain the image as a new modified radiation enhancement image. Repeat steps (3.3)-(3.6) to obtain the modified gray image of the weak light fishing boat;

[0028] (3.8) Add the gray image of the strong light fishing boat and the gray image of the weak light fishing boat, and perform local neighborhood calculation to obtain the night light fishing boat target;

[0029] (3.8.1) Set the second sliding window for the added image, calculate the average value of the standard deviation of all valid data in the window, and mark it as threshold std ;

[0030] (3.8.2) Calculate the mean value of each window, if two-thirds of the data in the window is greater than the mean value of the window or the mean value of the window is greater than a set mean value threshold, then the data in the window is directly retained;

[0031] (3.8.3) Calculate the standard deviation of the remaining data in the window and compare it with threshold std , if less than threshold std , then discard the data in the window, i.e. set all the data in the window to NaN; if greater than or equal to threshold std , then discard the data in the window that is less than the mean value of the window;

[0032] (3.8.4) For non-zero data points, compare each point with the data in the four directions above, below, left and right, retain the large value and remove the small value and set it to NaN;

[0033] (3.9) All pixels that are not set to NaN and are non-zero are the target fishing boat.

[0034] Further, the horizontal band lightning detection is: for the image data after the radiation enhancement processing, calculate the radiation difference between 16 rows of pixels, and the region with a difference greater than 0.1 Log10DNB and 24 or more consecutive pixels in the horizontal direction is marked as lightning.

[0035] Further, the gray value transformation on the corrected radiation enhancement image data is:

[0036]

[0037] intercept = 1 - (slope x min)

[0038] image_grey(i) = image(i) x slope + intercept

[0039] Where: max is the maximum value of the radiation enhancement image data after removing the lightning; min is the minimum value of the radiation enhancement image data after removing the lightning; image(i) represents the i-th pixel value in the radiation enhancement image data after removing the lightning; image_grey(i) represents the i-th pixel value in the gray image data.

[0040] Further, the maximum gradient is:

[0041]

[0042] Where: Max represents the maximum value of the difference between the center pixel (x, y) and the neighborhood eight pixels, and image (x, y) represents the gray image data of the pixel (x, y).

[0043] Further, the first sliding window is a 3*3 window, the step is 2, and the overlapping area is 1.

[0044] The second sliding window is a 3*3 window, the step is 1, and the overlapping area is 2.

[0045] Further, the preprocessing comprises:

[0046] (1.1) The impurity light of the night light remote sensing image data is removed, that is, the removed impurity light points are set to NaN, and the impurity light includes sunlight, moonlight, clouds, land light and sea fire points;

[0047] (1.2) The latitude and longitude of the night light remote sensing image data after removing the impurity light are geometrically corrected and the research area is cut.

[0048] Further, the impurity light removal processing is specifically:

[0049] Sunlight processing, according to the SolarZenithAngle attribute of VIIRS / DNB night light remote sensing image data, data less than 101° is considered as daytime and removed;

[0050] Moonlight processing, VIIRS / DNB night light remote sensing image data MoonPhaseAngle attribute, data in the range of 180°±20° is considered as full moon and removed;

[0051] Cloud processing is to use VCM data and VIIRS / DNB data for matching, and the VIIRS / DNB data matching area corresponding to the value 2 and 3 in the VCM data is removed;

[0052] Land light processing, using Global-land mask data, setting the continent and island reef land area in the data to 10km buffer zone as near sea, and removing the VIIRS / DNB data corresponding to the area;

[0053] Sea fire point processing, using VNF data, setting a 1km buffer zone of VNF data, and removing the VIIRS / DNB data corresponding to the area.

[0054] The beneficial effects of the present application are:

[0055] The present application is based on night light remote sensing VIIRS / DNB data, distinguishes strong light fishing boats and weak light fishing boats, respectively adopts local threshold value and global threshold value to extract strong light fishing boats and weak light fishing boats, and combines neighborhood analysis algorithm to eliminate the halo influence generated by light diffusion to the surrounding, and finally obtains accurate night light fishing boat targets, and has the characteristics of strong resolution and high precision. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 It is a VIIRS_DNB night light fishing boat self-adaptive recognition method flowchart based on the improved gradient method of the embodiment of the present application.

[0057] Figure 2 It is a VIIRS / DNB image after logarithmic transformation processing of the embodiment of the present application.

[0058] Figure 3 It is a gray scale image after the gray scale value of the embodiment of the present application is unified to [0-255].

[0059] Figure 4 It is a gradient image after gradient calculation of the embodiment of the present application.

[0060] Figure 5 It is a strong light gray scale image extracted after local threshold value segmentation of the embodiment of the present application.

[0061] Figure 6 It is a strong light gray scale image extracted after global threshold value segmentation of the embodiment of the present application.

[0062] Figure 7 It is a weak light gray scale image extracted after local threshold value segmentation of the embodiment of the present application.

[0063] Figure 8 It is a weak light gray scale image extracted after global threshold value segmentation of the embodiment of the present application.

[0064] Figure 9 It is a light fishing boat recognition result graph obtained after the strong light recognition result and the weak light recognition result of the embodiment of the present application are combined.

[0065] Figure 10 It is a more accurate light fishing boat recognition result graph obtained after neighborhood calculation and neighborhood analysis optimization of the embodiment of the present application.

[0066] Figure 11 It is a gradient method light fishing boat recognition result graph of the embodiment of the present application.

[0067] Figure 12 It is a comparison graph of the gradient method and the EOG data product of the embodiment of the present application. DETAILED DESCRIPTION

[0068] In order to better illustrate and understand the technical solutions and advantages of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and embodiments, but the content of the present invention is not limited to the following embodiments.

[0069] The following description uses the Sea of ​​Japan area ranging from 124.5°E to 128°E and 32°N to 34.5°N on April 1, 2020 as an example.

[0070] like Figure 1 As shown, the embodiment of the present invention provides a VIIRS_DNB luminous fishing boat adaptive recognition method based on the improved gradient method, comprising the following steps:

[0071] S101. Acquire VIIRS / DNB night light remote sensing image data.

[0072] The remote sensing image data used in the embodiments of the present invention all comes from the National Oceanic and Atmospheric Administration (NOAA). The Visible Infrared Imaging Radiometer Suite / Day-Night Band (VIIRS / DNB) data is acquired from NOAA's Suomi National Polar-orbiting Partnership (Suomi NPP) satellite and subsequent satellites.

[0073] VIIRS is an advanced visible infrared imaging radiometer suite, a multi-band remote sensing instrument that captures visible, infrared, and mid-infrared radiation from the Earth's surface. With a high spatial resolution of 742 meters and a wide range of observation bands, it provides global data on climate, weather, oceans, and land, monitoring various meteorological and environmental parameters such as clouds, aerosols, temperature, and humidity. The DNB band is a specialized band on VIIRS, primarily used to observe the Earth's surface at night. Its high sensitivity allows it to detect very faint light sources, such as city lights, fires, and bioluminescence from marine plankton. It can also detect nighttime clouds and climate phenomena, contributing to scientific research and policymaking. DNB appears black in background areas without light detection, rather than "salt and pepper" noise, making it suitable for thresholding to remove background noise. Compared to DMSP, VIIRS is more capable of detecting fishing vessel features in low-light imagery and has sufficient resolution to identify the source of vessel lights.

[0074] S102: Preprocessing the VIIRS / DNB night light remote sensing image data.

[0075] (1) The impurity light of the night light remote sensing image data is removed, that is, the removed impurity light points are set to NaN, and the impurity light includes sunlight, moonlight, clouds, land light and sea fire points.

[0076] The sunlight treatment is that according to the SolarZenithAngle attribute of the VIIRS / DNB night light remote sensing image data, the data less than 101° is considered as daytime and removed.

[0077] The moonlight treatment is that the MoonPhaseAngle attribute of the VIIRS / DNB night light remote sensing image data is used, and the data in the range of 180°±20° is considered as full moon and removed.

[0078] The cloud treatment is that the VCM data is matched with the VIIRS / DNB data, and the VIIRS / DNB data matching area corresponding to the values 2 and 3 in the VCM data is removed.

[0079] The land light treatment is that the Global-land mask data is used, and the land and island areas in the data are set as a 10km buffer zone as near sea, and the VIIRS / DNB data corresponding to the area is removed.

[0080] The sea fire point treatment is that the VNF data is used, and a 1km buffer zone is set for the VNF data, and the VIIRS / DNB data corresponding to the area is removed.

[0081] (2) The longitude and latitude of the night light remote sensing image data after removing the impurity light are geometrically corrected and the research area is cut.

[0082] The longitude and latitude of the night light remote sensing image data are geometrically corrected, which is used to eliminate the geometric distortion in the night light remote sensing image, so that the geographical coordinates in the image are consistent with the position in the real world.

[0083] The cutting of the research sea area is to cut the night light remote sensing image data by using the land mask and the longitude and latitude range of the research area, to obtain the image data of the research sea area.

[0084] S103, the improved gradient method is used to detect the gradient of the preprocessed night light remote sensing image data, and the night light fishing boat target is obtained; the improved gradient method is used to distinguish strong and weak two types of light fishing boats, and then the light halo pixels generated by single or multiple light fishing boat light traps are eliminated through local neighborhood calculation, and finally the accurate night light fishing boat target is obtained.

[0085] The identification of the night fishing boat is to distinguish the light-emitting pixels corresponding to the light emitted by the boat and the non-light-emitting pixels corresponding to the dark sea, wherein the gray level of the target object pixel is significantly different from the gray level of the background pixel. Therefore, threshold segmentation is a simple and effective tool for separating the target object from the background. Since the propagation of light has the characteristics of Gaussian distribution, the light will spread around the center pixel and lose intensity, and the light radiation values of different types of fishing boats have strong and weak differences. Therefore, the improved gradient algorithm of the embodiment of the present application is to obtain local and global thresholds for strong and weak light fishing boats respectively, and the obtained thresholds can accurately distinguish the boat light from the background dark sea, and then the halo pixels generated by the light of the single or multiple light fishing boats are removed through local neighborhood calculation, and finally the accurate night light fishing boat target is obtained.

[0086] The specific process is as follows:

[0087] (1) The pre-processed night remote sensing image data is subjected to radiation amplification (rad*10^9), and logarithmic transformation is used for radiation enhancement (log10(rad)) to obtain a radiation enhanced image. As shown in the following formula: Figure 2

[0088] (2) The radiation enhanced image data is subjected to horizontal band lightning detection, and the pixels marked as lightning are discarded, that is, the pixel value is set to NaN, to obtain a modified radiation enhanced image.

[0089] The horizontal band lightning detection is that, for the image data subjected to radiation enhancement processing, the radiation difference between rows of 16 consecutive rows of pixels is calculated, and the region with a difference greater than 0.1 Log10 DNB and 24 or more consecutive pixels in the horizontal direction is marked as lightning.

[0090] (3) The modified radiation enhanced image data is subjected to gray value transformation to adapt to the gray scale [0-255] to obtain a gray image. As shown in the following formula: Figure 3

[0091]

[0092] intercept=1-(slope×min)

[0093] image_grey(i)=image(i)×slope+intercept

[0094] Wherein: max is the maximum value in the modified radiation enhanced image data; min is the minimum value in the modified radiation enhanced image data; image(i) represents the i-th pixel value in the modified radiation enhanced image data; image_grey(i) represents the i-th pixel value in the gray image data. ​​

[0095] The gray value transformation can enhance the light-dark contrast, making the image features clearer, as shown in Figure 3 .

[0096] (4) The gradient calculation is performed on the gray image data, which aims to highlight the features of the light fishing boat and facilitate the identification of the target light fishing boat. Since the light spreads in all directions in the night and presents a Gaussian distribution, the difference between the center pixel and the surrounding eight pixels is taken when calculating the gradient image (image_gradients), and the maximum value is taken as the gradient value of the pixel.

[0097]

[0098] wherein: represents the maximum value of the difference between the center pixel (x, y) and the eight neighboring pixels, and image(x, y) represents the gray image data of the pixel (x, y).

[0099] The gradient value of each pixel is compared with the set gradient threshold value, and the pixels less than the set gradient threshold value are removed, i.e. the pixel value is set to NaN, to obtain the gradient image. As shown in Figure 4 .

[0100] (5) Calculate the gradient threshold value:

[0101] (5.1) Calculate the gradient image mean value gradient_mean for the obtained gradient image, and extract the gray image data corresponding to all gradient values greater than gradient_mean as the candidate threshold vector.

[0102] (5.2) According to the candidate threshold vector, calculate the local threshold value, and use the first sliding window to calculate the local threshold value of each window. The first sliding window is a 3*3 window, the step length is 2, and the overlapping area is 1.

[0103]

[0104] wherein: i represents the i-th sliding window, min i and max i represent the minimum and maximum values of the candidate threshold vector in the i-th sliding window; mean i is the average value of the candidate threshold vector in the i-th sliding window, and threshold local，i represents the local threshold value of the i-th sliding window.

[0105] (5.3) According to the candidate threshold vector, calculate the global threshold value.

[0106]

[0107] wherein: min and max are the minimum and maximum values of all the candidate threshold vectors, median is the median of all the candidate threshold vectors; threshold global is the global threshold.

[0108] (6) For the gray image data in each sliding window, keep the points higher than the local threshold, and set the other points as NaN. Then, for the whole gray image, keep the points higher than the global threshold, and set the other points as NaN, to obtain the modified gray image.

[0109] (7) The modified gray image at this time is the gray image of the strong light fishing boat. Set the corresponding points in the gray image of the valid value points in the gray image of the strong light fishing boat as NaN, and obtain the image as a new modified radiation enhancement image. Repeat steps (3)-(6) to obtain the modified gray image, which is the gray image of the weak light fishing boat. As shown in Figures 5-8 .

[0110] (8) Add the gray image of the strong light fishing boat and the gray image of the weak light fishing boat, and perform local neighborhood calculation to obtain the night light fishing boat target.

[0111] (8.1) For the added image, as shown in Figure 9 , set a second sliding window, calculate the average of the standard deviation of all valid data windows, and record it as threshold std . The second sliding window is a 3*3 window, the step is 1, and the overlapping area is 2.

[0112] (8.2) Calculate the mean value of the data in each window. If two-thirds of the data in the window is greater than the mean value of the window or the mean value of the window is greater than the set mean value threshold, the window data is directly retained.

[0113] (8.3) Calculate the standard deviation of the data in the remaining window and compare it with threshold std . If it is less than threshold std , discard the window data, i.e. set all data in the window as NaN; if it is greater than or equal to threshold std , discard the data in the window that is less than the mean value of the window.

[0114] (8.4) For valid data points, compare each point with the data in the upper, lower, left and right directions, retain the large value and remove the small value, i.e. set the small value as NaN, as shown in Figure 10 .

[0115] (9) All pixels that are not set as NaN and are non-zero are the target light fishing boat, as shown in Figure 11 .

[0116] The present invention compares the improved gradient method with the EOG data product, and the results are as follows: Figure 12 As shown in the figure. This figure takes the sea area of ​​124.5°E~128°E, 32°N~34.5°N in the Sea of ​​Japan on April 1, 2020 as an example. The red cross marks the position of the fishing boat with lights extracted by the improved gradient method, and the yellow dot is the fishing boat position information provided by the Earth Observation Group (VIIRS Boat Detection) in the corresponding sea area on that day. It can be seen from the figure that the fishing boat positions extracted by the improved gradient method are highly matched with the fishing boat positions provided by EOG. It can be seen from the figure that in areas with relatively dim lighting, the improved gradient method can extract relatively more fishing boat targets, and in areas with weaker lighting, EOG can identify relatively more fishing boats. The fishing boat targets extracted by the two methods can complement each other.

[0117] The above specific description further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A VIIRS_DNB luminous fishing boat adaptive recognition method based on an improved gradient method, characterized in that: The steps include: (1) Obtain VIIRS / DNB night light remote sensing image data; (2) preprocessing the VIIRS / DNB night light remote sensing image data; (3) An improved gradient method is used to perform gradient detection on the pre-processed night light remote sensing image data to obtain the night light fishing vessel target; the improved gradient method is used to distinguish between two types of light fishing vessels, strong and weak, and fuse them, and then eliminate the halo pixels generated by the lights of single or multiple light trapping fishing vessels through local neighborhood calculation, and finally obtain the accurate night light fishing vessel target.

2. The VIIRS_DNB luminous fishing vessel adaptive recognition method based on the improved gradient method according to claim 1 is characterized in that: The step (3) is specifically as follows: (3.1) performing radiation amplification on the pre-processed night light remote sensing image data and performing radiation enhancement using a logarithmic transformation to obtain a radiation enhanced image; (3.2) Perform horizontal strip lightning detection on the radiation-enhanced image data, discard pixels marked as lightning, and set the pixel value to NaN to obtain a corrected radiation-enhanced image; (3.3) performing grayscale value transformation on the corrected radiation enhanced image data to adapt it to the grayscale level [0-255] to obtain a grayscale image; (3.4) Calculate the gradient of the grayscale image data by taking the difference between the central pixel and the eight surrounding pixels, and taking the maximum value as the gradient value of the pixel; Compare the gradient value of each pixel with the set gradient threshold, remove the pixels whose gradient value is less than the set gradient threshold, and set the pixel value to NaN to obtain the gradient image; (3.5) Calculate the gradient threshold: (3.5.1) Calculate the average gradient value gradient_mean of the obtained gradient image, and extract the grayscale image data corresponding to all gradient values ​​exceeding gradient_mean as the candidate threshold vector; (3.5.2) Calculating a local threshold based on the selected threshold vector, wherein the local threshold is locally calculated using the first sliding window to obtain a local threshold for each window; Where: i represents the i-th sliding window, min i and max i represents the minimum and maximum values ​​of the threshold vector to be selected in the i-th sliding window; mean i is the average value of the threshold vector to be selected in the i-th sliding window, thresold local，i represents the local threshold of the i-th sliding window; (3.5.3) Calculate the global threshold based on the threshold vector to be selected; Wherein: min and max are the minimum and maximum values ​​of all the threshold vectors to be selected, median is the median of all the threshold vectors to be selected; threshold global is the global threshold; (3.6) For the grayscale image data within each first sliding window in the grayscale image, retain the points above the local threshold and set the other points to NaN; then, for the entire grayscale image, retain the points above the global threshold and set the other points to NaN, to obtain a corrected grayscale image; (3.7) The corrected grayscale image at this point is the grayscale image of the fishing boat with strong lights. The corresponding points in the grayscale image of the fishing boat with strong lights are set to NaN. The resulting image is used as the new corrected radiation-enhanced image. Repeat steps (3.3)-(3.6) to obtain the corrected grayscale image of the fishing boat with weak lights. (3.8) The grayscale image of the fishing boat with strong lights and the grayscale image of the fishing boat with weak lights are added together and then local neighborhood calculation is performed to obtain the target of the fishing boat with night lights. (3.8.1) Set a second sliding window for the added image and calculate the average standard deviation of the valid data in all windows, which is recorded as threshold std ; (3.8.2) Calculate the mean of the data in each window separately. If two-thirds of the data in the window is greater than the mean of the window or the mean of the window is greater than the set mean threshold, then directly retain the data in the window; (3.8.3) Calculate the standard deviation of the data in the remaining window respectively and compare it with the threshold std For comparison, if it is less than threshold std , then discard the window data, that is, set all data in the window to NaN; if it is greater than or equal to threshold std , then discard the data in the window that is smaller than the mean of the window and set the discarded value to NaN; (3.8.4) For non-zero data points, compare each point with the data in the four directions above, below, left, and right. Keep the larger value and remove the smaller value, that is, set the smaller value to NaN. (3.9) All pixels that are not set to NaN values ​​and are non-zero are the target fishing vessels with lights.

3. The VIIRS_DNB luminous fishing vessel adaptive recognition method based on the improved gradient method according to claim 2 is characterized in that: The horizontal strip lightning detection is as follows: for the image data after radiation enhancement processing, the radiation difference between 16 rows of pixels is calculated and separated. The area where the difference is greater than 0.1Log10 DNB and lasts for 24 or more consecutive pixels in the horizontal direction is marked as lightning.

4. The VIIRS_DNB luminous fishing vessel adaptive recognition method based on the improved gradient method according to claim 2 is characterized in that: The grayscale value transformation of the modified radiation enhanced image data is: intercept = 1-(slope × min) image_grey(i)=image(i)×slope+intercept Wherein: max is the maximum value in the corrected radiation-enhanced image data; min is the minimum value in the corrected radiation-enhanced image data; image(i) represents the i-th pixel value in the corrected radiation-enhanced image data; image_grey(i) represents the i-th pixel value in the grayscale image data.

5. The VIIRS_DNB luminous fishing vessel adaptive recognition method based on the improved gradient method according to claim 2 is characterized in that: The maximum gradient is: in: It represents the maximum value of the difference between the central pixel (x, y) and the eight neighboring pixels, and image(x, y) represents the grayscale image data of the pixel (x, y).

6. The VIIRS-DNB luminous fishing vessel adaptive recognition method based on the improved gradient method according to claim 2 is characterized by: The first sliding window is a 3*3 window with a step size of 2 and an overlapping area of ​​1; The second sliding window is a 3*3 window with a step size of 1 and an overlapping area of ​​2.

7. The VIIRS_DNB luminous fishing boat adaptive recognition method based on the improved gradient method according to claim 1 is characterized in that: The pretreatment includes: (1.1) performing impurity light removal processing on the night light remote sensing image data, i.e., setting the removed impurity light points to NaN, wherein the impurity lights include: sunlight, moonlight, clouds, land lights, and sea fire points; (1.2) Geometric correction is performed on the longitude and latitude of the night light remote sensing image data after removing the impurity lights, and the research area is clipped.

8. The VIIRS_DNB luminous fishing boat adaptive recognition method based on the improved gradient method according to claim 7 is characterized in that: The impurity light removal process is specifically as follows: daylight processing, according to the SolarZenithAngle attribute of VIIRS / DNB night light remote sensing image data, the data with an angle less than 101° is considered as daytime and removed; Moonlight processing: The MoonPhaseAngle attribute of VIIRS / DNB night light remote sensing image data is used to treat data within the range of 180°±20° as a full moon and remove it; Cloud processing uses VCM data to match VIIRS / DNB data, and removes the VIIRS / DNB data matching areas corresponding to VCM data median values ​​of 2 and 3; For land lighting processing, we used the global land mask data and set a 10km buffer zone around the land areas such as the continents and islands in the data, treating them as offshore areas. The VIIRS / DNB data corresponding to this area were removed. For offshore fire point processing, VNF data is used, a 1km buffer is set for the VNF data, and the VIIRS / DNB data corresponding to this area are removed.

Images