A Fast Detection Method for Large Ship Targets Based on GF-4 Satellite

Abstract

The present invention discloses a method for rapid detection of large ship targets based on GF-4 satellites. The method includes: performing land-sea area segmentation on PMS image data to extract land-sea mask data; wherein the PMS image data includes a near-infrared band and a red band; extracting feature points from the near-infrared band based on the land-sea mask data; obtaining slice data based on the feature points, and classifying the histogram of oriented gradients features corresponding to the tangent point data to determine large ship targets. The present invention can improve the detection accuracy and efficiency of large ship targets through rough detection and fine detection.

Description

Technical Field

[0001] The present invention belongs to the field of information processing, and particularly relates to a method for quickly detecting large ship targets based on GF-4 satellites. Background Art

[0002] In order to detect large ship targets, in the prior art, image data is usually repeatedly read and written for different satellites and different payloads for detection. This method cannot detect large ships in real time and dynamically, and the process of repeatedly reading and writing images will seriously affect the detection efficiency, resulting in too long detection time. Summary of the Invention

[0003] In order to solve the above problems existing in the prior art, the present invention provides a method for quickly detecting large ship targets based on GF-4 satellites. The technical problems to be solved by the present invention are realized through the following technical solutions:

[0004] A method for quickly detecting large ship targets based on GF-4 satellites, the method comprising: Step 1: performing land-sea area segmentation on PMS image data to extract land-sea mask data; wherein, the PMS image data includes a near-infrared band and a red band; Step 2: extracting feature points from the near-infrared band based on the land-sea mask data; Step 3: obtaining slice data based on the feature points, and classifying the histogram of oriented gradients features corresponding to the slice data to determine large ship targets.

[0005] In an embodiment of the present invention, the Step 1 includes: Step 1-1: performing linear stretching on the near-infrared band Nir and the red band Red according to preset stretching parameters; Step 1-2: calculating the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the linearly stretched near-infrared band, and calculating the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the linearly stretched near-infrared band; Step 1-3: calculating the vegetation index NDVI based on the normalized original pixel values, expressed as: NDVI = (Nir – Red) / (Nir + Red); Step 1-4: performing land-sea area segmentation on the PMS image data based on the vegetation index and a preset segmentation threshold to extract land-sea mask data.

[0006] In an embodiment of the present invention, the Step 2 includes: Step 2-1: determining the high-brightness points in the near-infrared band; Step 2-2: determining the pixel neighborhood range corresponding to the high-brightness points with the high-brightness points as the center; Step 2-3: extracting feature points from the pixel neighborhood range corresponding to the high-brightness points in the near-infrared band based on the land-sea mask through a preset algorithm, wherein the preset algorithm corresponds to a preset pixel value difference threshold and a preset tolerance value.

[0007] Advantages of the present invention:

[0008] The present invention can take PMS image data as input, extract feature points from the near-infrared band based on the segmented land-sea mask data, that is, rough detection, and then classify the histogram of oriented gradients features of each feature point, that is, fine detection, to achieve the detection of large ship targets. Therefore, the present invention can perform two-level detection through rough detection and fine detection, improving the detection accuracy and efficiency of large ship targets.

[0009] The present invention will be further described in detail below with reference to the drawings and embodiments. Description of the Drawings

[0010] Figure 1 is a schematic flowchart of a method for quickly detecting large ship targets based on GF-4 satellite provided by an embodiment of the present invention;

[0011] Figure 2 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Detailed Embodiments

[0012] The present invention will be further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0013] In recent years, with the development of remote sensing satellite technology and the improvement of application requirements, the spatial resolution and temporal resolution of remote sensing satellites have been greatly improved. With the launch of China's first geosynchronous orbit remote sensing satellite (GF-4) in 2015, China has the ability to observe the surrounding areas of China in a pointed manner. The GF-4 satellite is equipped with a visible light payload of 50m and a medium-wave infrared payload of 400m, and has the ability to detect large ships under good meteorological conditions such as sea conditions, and also has the ability to continuously observe the southeastern waters of China. The present invention is based on GF-4 data to realize the function of quickly detecting large ships and has the ability of continuous observation. This technology has certain application value in the military, maritime and other fields.

[0014] Embodiment 1

[0015] Please refer to Figure 1 , Figure 1 which is a schematic diagram of a method for quickly detecting large ship targets based on GF-4 satellite provided by an embodiment of the present invention. The method includes:

[0016] Step 1: Perform land-sea area segmentation on the PMS image data to extract land-sea mask data; wherein, the PMS image data includes a near-infrared band and a red band.

[0017] The CF-4 refers to the Gaofen-4 satellite, and the PMS refers to the visible and near-infrared multispectral image. The input PMS image data of the present invention is GF-4-PMS data, with a total of five bands. This step uses the near-infrared band (Nir) and the red band (Red).

[0018] Optionally, step 1 includes:

[0019] Step 1-1: Linearly stretch the near-infrared band Nir and the red band Red according to preset stretching parameters.

[0020] The preset stretching parameters are set by those skilled in the art according to business needs, and the present invention does not limit this. For example, the preset stretching parameter is 2%. The present invention can first stretch the two bands respectively in the way of 2% linear stretching, then calculate the maximum and minimum normalized original pixel values of each band respectively, then calculate the vegetation index NDVI, and set a threshold for segmentation.

[0021] Step 1-2: Calculate the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the linearly stretched near-infrared band, and calculate the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the linearly stretched red band.

[0022] Step 1-3: Based on the normalized original pixel values, calculate the vegetation index NDVI, expressed as:

[0023] NDVI = (Nir – Red) / (Nir + Red).

[0024] Step 1-4: Based on the vegetation index and a preset segmentation threshold, perform sea-land area segmentation on the PMS image data to extract sea-land mask data.

[0025] The present invention performs relative radiometric calibration through the normalization processing of the red and near-infrared bands, obtains the vegetation index by using the calculation formula of the normalized vegetation index, sets the segmentation threshold of the normalized vegetation index to obtain the sea-land mask data of the panoramic image data. This step adopts the idea of parallel processing, which can effectively suppress the land interference in the detection process.

[0026] Step 2: Extract feature points from the near-infrared band based on the sea-land mask data.

[0027] What step 2 realizes is the rough extraction of large ship targets.

[0028] Optionally, step 2 includes:

[0029] Step 2-1: Determine the high-brightness points in the near-infrared band.

[0030] Step 2-2: Taking the highlight as the center, determine the pixel neighborhood range corresponding to the highlight.

[0031] Step 2-3: Based on the land-sea mask, extract feature points from the pixel neighborhood range corresponding to the highlight in the near-infrared band through a preset algorithm, where the preset algorithm corresponds to a preset pixel value difference threshold and a preset tolerance value.

[0032] Among them, the preset pixel value difference threshold is used to extract feature points. When the difference is greater than the preset pixel value threshold, the current point is considered a feature point.

[0033] Taking a 10*10 neighborhood as an example, a total of 38 points need to be compared with the center point. The preset tolerance value refers to the number of pixel points that are not satisfied with the preset pixel value threshold allowed in the 38 points in the neighborhood.

[0034] The preset algorithm is selected by those skilled in the art according to business needs, and the present invention does not limit this. For example, the preset algorithm is the ORB (Oriented FAST and Rotated BRIEF) parallel algorithm. The present invention can quickly extract feature points through parallel calculation using the ORB algorithm in the near-infrared band.

[0035] The feature points refer to the highlights in a relatively small area of the image.

[0036] Based on the land-sea mask data, the near-infrared band is processed pixel by pixel through the ORB parallel feature point fast detection algorithm, and this method can quickly extract the highlighted areas of suspected large ship targets in the near-infrared band.

[0037] Step 3: Based on the feature points, obtain slice data and classify the histogram of oriented gradients features corresponding to the slice data to determine large ship targets.

[0038] Optionally, the step 3 includes:

[0039] Step 3-1: Obtain slice data with the feature point as the center, and the slice data corresponds to preset neighborhood parameters.

[0040] For example, based on the feature points obtained by rough detection, slice data within its 32*32 neighborhood is obtained with this feature point as the center, and this data is stored in the memory. The 32*32 is the preset neighborhood parameter.

[0041] Step 3-2: Calculate the histogram of oriented gradients features of the slice data.

[0042] The histogram of oriented gradients feature is simply called the hog feature.

[0043] Step 3-3: Classify the Histogram of Oriented Gradients (HOG) features based on a Support Vector Machine (SVM) to divide the HOG features into interference target HOG features and large ship target HOG features.

[0044] Optionally, the SVM needs to be trained before classification. The training data is the target data in the near-infrared band of GF-4 collected, with a data size of 32*32. The trained SVM can accurately classify the HOG features of interference targets such as island reef broken clouds and the HOG features of large ship targets.

[0045] The present invention can use PMS image data as input, perform land-sea segmentation by calculating the Normalized Difference Vegetation Index (NDVI), and then divide it into two steps: rough detection and fine detection. The rough detection target: Use the ORB parallel algorithm to detect the high-brightness feature points of the image as suspicious ship targets; the fine detection target: Extract the HOG features of the suspicious targets, use the SVM classifier to screen the ship targets, and delete interference targets such as island reef broken clouds. Through fine detection, among the suspicious ship targets (feature points), further screen out the interference targets to improve the detection accuracy of large ships.

[0046] Step 3-4: Determine the large ship target HOG features as large ship targets.

[0047] Based on the grayscale histogram features of the slice data corresponding to the large ship target, classify it with an SVM classifier. If the target is classified as a ship, it is considered that the point position is the detected large ship target.

[0048] It should be noted that the Hog feature can well describe the directional density distribution of the gradient or edge of a large ship.

[0049] The present invention solves the problem of repeatedly reading and writing image data for radiation correction for different satellites and different payloads in the existing preprocessing, improves the efficiency of preprocessing standard product production, can meet the real-time and rapid radiation correction processing of different satellites and different payloads, and performs radiation correction on images in real time.

[0050] In summary, the present invention can use PMS image data as input, extract feature points from the near-infrared band based on the land-sea mask data after segmentation processing, that is, rough detection, and then classify the HOG features of each feature point, that is, fine detection, to achieve the detection of large ship targets. Therefore, the present invention can perform two-level detection through rough detection and fine detection to improve the detection accuracy and efficiency of large ship targets.

[0051] Embodiment 2

[0052] The present invention provides a large ship target rapid detection device based on GF-4 satellite, and the device includes:

[0053] A first extraction module, configured to perform land-sea area segmentation on PMS image data to extract land-sea mask data; wherein, the PMS image data includes a near-infrared band and a red band.

[0054] A second extraction module, configured to extract feature points from the near-infrared band based on the land-sea mask data.

[0055] A target determination module, configured to obtain slice data based on the feature points, and classify the histogram of oriented gradients features corresponding to the tangent point data to determine a large ship target.

[0056] Optionally, the first extraction module includes:

[0057] A linear stretching sub-module, configured to perform linear stretching on the near-infrared band Nir and the red band Red according to preset stretching parameters;

[0058] A sub-module for calculating normalized original pixel values, configured to calculate the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the linearly stretched near-infrared band, and calculate the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the linearly stretched near-infrared band;

[0059] A sub-module for calculating a vegetation index, configured to calculate a vegetation index NDVI based on the normalized original pixel values, expressed as:

[0060] NDVI = (Nir – Red) / (Nir + Red);

[0061] A sub-module for extracting mask data, configured to perform land-sea area segmentation on the PMS image data based on the vegetation index and a preset segmentation threshold to extract land-sea mask data.

[0062] Optionally, the second extraction module includes:

[0063] A sub-module for determining high-brightness points, configured to determine high-brightness points in the near-infrared band;

[0064] A sub-module for determining a pixel neighborhood range, configured to determine a pixel neighborhood range corresponding to the high-brightness point with the high-brightness point as the center;

[0065] A sub-module for extracting feature points, configured to extract feature points from the pixel neighborhood range corresponding to the high-brightness points in the near-infrared band based on the land-sea mask through a preset algorithm, wherein the preset algorithm corresponds to a preset pixel value difference threshold and a preset error tolerance value.

[0066] Optionally, the determining target module includes:

[0067] A slicing data acquisition sub-module, configured to acquire slicing data centered on the feature point, where the slicing data corresponds to preset neighborhood parameters;

[0068] A feature calculation sub-module, configured to calculate the histogram of oriented gradients feature of the slicing data;

[0069] A classification sub-module, configured to classify the histogram of oriented gradients feature based on a support vector machine, so as to classify the histogram of oriented gradients feature into an interference target histogram of oriented gradients feature and a large ship target histogram of oriented gradients feature;

[0070] A target determination sub-module, configured to determine the large ship target histogram of oriented gradients feature as a large ship target.

[0071] In summary, the present invention can use PMS image data as input, based on the sea-land mask data after segmentation processing, extract feature points from the near-infrared band, that is, rough detection, and then classify the histogram of oriented gradients feature of each feature point, that is, fine detection, to achieve the detection of large ship targets. Therefore, the present invention can perform two-level detection through rough detection and fine detection, improving the detection accuracy and efficiency of large ship targets.

[0072] Embodiment 3

[0073] The embodiment of the present invention further provides an electronic device, as Figure 2 shown, including a processor 21, a communication interface 22, a memory 23, and a communication bus 24, where the processor 21, the communication interface 22, and the memory 23 complete communication with each other through the communication bus 24,

[0074] The memory 23 is used to store a computer program;

[0075] The processor 21 is configured to implement the following steps when executing the program stored on the memory 23:

[0076] Step 1: Perform sea-land area segmentation on the PMS image data to extract sea-land mask data; where the PMS image data includes a near-infrared band and a red band;

[0077] Step 2: Based on the sea-land mask data, extract feature points from the near-infrared band;

[0078] Step 3: Based on the feature points, acquire slicing data, and classify the histogram of oriented gradients feature corresponding to the tangent point data to determine a large ship target.

[0079] The communication bus mentioned in the above-mentioned electronic device can be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience in representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0080] The communication interface is used for communication between the above-mentioned electronic device and other devices.

[0081] The memory can include a Random Access Memory (RAM), and can also include a Non-Volatile Memory (NVM), such as at least one disk memory. Optionally, the memory can also be at least one storage device located far from the aforementioned processor.

[0082] The above-mentioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0083] The method provided by the embodiments of the present invention can be applied to an electronic device. Specifically, the electronic device can be: a desktop computer, a portable computer, a smart mobile terminal, a server, etc. There is no limitation here, and any electronic device that can implement the present invention belongs to the protection scope of the present invention.

[0084] For the device / electronic device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for the relevant parts, refer to the partial description of the method embodiments.

[0085] It should be noted that the device and the electronic device in the embodiments of the present invention are respectively the device and the electronic device applying the above-mentioned method for quickly detecting large ship targets based on GF-4 satellites. All embodiments of the above-mentioned method for quickly detecting large ship targets based on GF-4 satellites are applicable to the device and the electronic device, and can achieve the same or similar beneficial effects.

[0086] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0087] The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for rapid detection of large ship targets based on GF-4 satellite, characterized in that, The method includes: Step 1: Perform land-sea area segmentation on the PMS image data to extract land-sea mask data; wherein, the PMS image data includes a near-infrared band and a red band; Step 2: Extract feature points from the near-infrared band based on the land-sea mask data; Step 3: Based on the feature points, obtain slice data and classify the histogram of oriented gradients (HOG) features corresponding to the slice data to determine large ship targets. Among them, Step 1 includes: Step 1-1: Perform linear stretching on the near-infrared band Nir and the red band Red according to preset stretching parameters; Step 1-2: Calculate the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the linearly stretched near-infrared band, and calculate the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the linearly stretched near-infrared band; Step 1-3: Calculate the vegetation index NDVI based on the normalized original pixel values, expressed as: NDVI = (Nir – Red) / (Nir + Red); Step 1-4: Perform land-sea area segmentation on the PMS image data based on the vegetation index and a preset segmentation threshold to extract land-sea mask data; Among them, Step 2 includes: Step 2-1: Determine the highlight points in the near-infrared band; Step 2-2: Determine the pixel neighborhood range corresponding to the highlight points with the highlight points as the center; Step 2-3: Based on the land-sea mask, extract feature points from the pixel neighborhood range corresponding to the highlight points in the near-infrared band through a preset algorithm, wherein the preset algorithm corresponds to a preset pixel value difference threshold and a preset tolerance value.

2. The method according to claim 1, wherein Step 3 includes: Step 3-1: Obtain slice data with the feature points as the center, and the slice data corresponds to preset neighborhood parameters; Step 3-2: Calculate the histogram of oriented gradients (HOG) features of the slice data; Step 3-3: Classify the histogram of oriented gradients (HOG) features based on a support vector machine to divide the histogram of oriented gradients (HOG) features into interference target histogram of oriented gradients (HOG) features and large ship target histogram of oriented gradients (HOG) features; Step 3-4: Determine the large ship target histogram of oriented gradients (HOG) features as large ship targets.

3. A rapid detection device for large ship targets based on GF-4 satellite, characterized in that, The device includes: A first extraction module for performing land-sea area segmentation on the PMS image data to extract land-sea mask data; wherein, the PMS image data includes a near-infrared band and a red band; A second extraction module for extracting feature points from the near-infrared band based on the land-sea mask data; A target determination module for obtaining slice data based on the feature points and classifying the histogram of oriented gradients (HOG) features corresponding to the slice data to determine large ship targets; Among them, the first extraction module includes: A linear stretching sub-module for performing linear stretching on the near-infrared band Nir and the red band Red according to preset stretching parameters; The normalized original pixel value calculation sub-module is used to calculate the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the near-infrared band after linear stretching, and to calculate the maximum normalized original pixel value and the minimum normalized original pixel value corresponding to the near-infrared band after linear stretching; The vegetation index calculation sub-module is used to calculate the vegetation index NDVI based on the normalized original pixel value, expressed as: NDVI = (Nir–Red) / (Nir+Red); The mask data extraction sub-module is used to perform land-sea area segmentation on the PMS image data based on the vegetation index and a preset segmentation threshold to extract land-sea mask data; Among them, the second extraction module includes: The high-brightness point determination sub-module is used to determine the high-brightness points in the near-infrared band; the pixel neighborhood range determination sub-module is used to determine the pixel neighborhood range corresponding to the high-brightness points with the high-brightness points as the center; The feature point extraction sub-module is used to extract feature points from the pixel neighborhood range corresponding to the high-brightness points in the near-infrared band based on the land-sea mask through a preset algorithm, where the preset algorithm corresponds to a preset pixel value difference threshold and a preset error tolerance value.

4. The device according to claim 3, characterized in that, The target determination module includes: The slice data acquisition sub-module is used to acquire slice data centered on the feature points, and the slice data corresponds to preset neighborhood parameters; The feature calculation sub-module is used to calculate the histogram of oriented gradients features of the slice data; The classification sub-module is used to classify the histogram of oriented gradients features based on a support vector machine to classify the histogram of oriented gradients features into interference target histogram of oriented gradients features and large ship target histogram of oriented gradients features; The target determination sub-module is used to determine the large ship target histogram of oriented gradients features as a large ship target.

5. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus; The memory is used to store computer programs; The processor is used to implement the method steps described in claim 1 or 2 when executing the programs stored on the memory.

Images