Remote sensing image ship target detection and self-adaptive cutting method based on dual sea-land segmentation and screening

Abstract

The invention provides a remote sensing image ship target detection and self-adaptive cutting method based on dual sea-land segmentation and screening. The method comprises the following steps: firstly, carrying out gridding dicing on a large-format remote sensing image according to a preset size and an overlapping ratio; thirdly, analyzing sea-land attributes of four corner points of each block by adopting a land-sea segmentation detection algorithm, screening and retaining slices containing a sea area, and removing pure land slices to reduce subsequent calculation load; then, the reserved ocean slices are subjected to enhancement treatment; then, an improved deep learning target detection algorithm is adopted to carry out ship target detection on the enhanced slices; thirdly, a land-sea segmentation detection algorithm is adopted to analyze detected ship target center point sea-land attributes, and false alarm targets located in a land area are screened and removed to improve the detection precision; then, self-adaptive image cutting is carried out by taking the detected ship target as the center, and a self-defined cutting size is supported; and finally, coastline information is extracted, and the distance between the ship target and the coastline is calculated.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing image processing and computer vision technology, and in particular to a method for ship target detection and adaptive cropping of remote sensing images based on dual land-sea segmentation filtering. Background Technology

[0002] With the rapid development of satellite remote sensing technology and the increasing demand for marine monitoring, ship target detection based on remote sensing imagery has become a core technological requirement in fields such as marine monitoring, maritime management, port safety, and fisheries supervision. However, existing ship detection technologies face many challenges when processing large-format remote sensing imagery.

[0003] First, large-format remote sensing imagery data is massive, making full-map detection computationally extremely complex and demanding on computational resources. While traditional gridded processing can reduce the amount of data processed per cycle, it often involves detecting and calculating tiles containing large land areas, resulting in a large amount of invalid computation and severely impacting processing efficiency.

[0004] Secondly, existing ship detection algorithms generally suffer from a high false alarm rate, especially in near-shore areas, where land-based targets such as buildings and vehicles are easily misidentified as ships, leading to insufficient reliability of detection results. Although some methods attempt to remove false alarms through post-processing, an effective land-sea area differentiation mechanism is lacking.

[0005] Furthermore, after detecting a ship target, how to effectively crop the target area is also a technical challenge. Fixed-size cropping methods cannot adapt to the characteristics of ships of different sizes, while manually adjusting the cropping parameters reduces the automation level of the system.

[0006] Furthermore, for marine monitoring applications, the distance between ship targets and the coastline is an important analytical indicator, but existing systems often lack integrated distance calculation functions, requiring the use of additional GIS software for analysis, which increases the complexity of the workflow.

[0007] Therefore, there is an urgent need to develop a highly efficient, accurate, and automated method for detecting and processing ship targets in large-format remote sensing images. This method should effectively reduce invalid calculations, lower the false alarm rate, achieve adaptive cropping, and integrate distance analysis functions to meet practical application needs. Summary of the Invention

[0008] To address the problems existing in the prior art, this invention proposes for the first time a remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation screening. This method solves the technical problems of high computational resource consumption, high false alarm rate, and inflexible cropping methods in the prior art, and achieves efficient detection and accurate positioning of ship targets in large-format remote sensing images.

[0009] This invention provides a method for ship target detection and adaptive cropping in remote sensing images based on dual land-sea segmentation filtering, comprising the following steps:

[0010] Step S1: Parameter Configuration

[0011] Key parameters, including the size of large-format remote sensing image slices, can be set via a JSON configuration file. Slicing overlap rate parameter and target type selection ;

[0012] Step S2: Image Preprocessing

[0013] Read the input large-format remote sensing image file, obtain the spatial resolution, georeferenced information and pixel size of the remote sensing image, select appropriate block size parameters according to the resolution, perform grid segmentation using a sliding window method, calculate the pixel coordinate range and geographic coordinate range of each block, and establish coordinate mapping relationship;

[0014] Step S3: First-level land-sea segmentation and filtering

[0015] For each image block, extract the geographic coordinates of the four corner points, calculate the index position in the global land-sea database based on latitude and longitude, read the land-sea attribute identifiers corresponding to the corner points, discard the block if all four corner points are land, otherwise retain it for subsequent processing, count the filtering results, and record the number of retained ocean blocks;

[0016] Step S4: Image Enhancement Processing

[0017] For each retained ocean slice, linear gray-scale stretching, optimal linear stretching, and square root stretching algorithms are applied to generate enhancement candidate results. The information entropy, standard deviation, average gradient, and histogram uniformity index of each candidate result are calculated. The optimal enhancement result is selected by weighted scoring.

[0018] Step S5: Ship Target Detection

[0019] The resolution-adaptive dual-model ONNX inference framework is used to detect ship targets in the enhanced ocean slices. Confidence thresholding and non-maximum suppression algorithms are applied to remove low-quality and duplicate detections, and the coordinates of the detection results are transformed back to the original image coordinate system.

[0020] Step S6: Second-level land-sea segmentation and filtering

[0021] For each detected ship target, extract its center point pixel coordinates and convert them into geographic coordinates. Query the attribute identifier of the center point in global land and sea data, retain only targets located in the ocean area, remove false alarms on land, and statistically analyze the change in the number of targets before and after filtering.

[0022] Step S7: Adaptive cropping process

[0023] Centered on the ship target that has passed the double screening verification, the system automatically calculates the optimal cropping range based on the ship target's bounding box size and geometric features, combined with the user-preset cropping size constraint parameters, and dynamically generates standardized local image fragments containing the complete target and appropriate contextual information.

[0024] Step S8: Calculate the coastline distance

[0025] The UTM projection zone and EPSG encoding of the ship target are determined based on its geographical coordinates. The coastline vector data of the corresponding area are read, and the coastline and target point are projected onto the UTM coordinate system. The shortest distance from the target point to the coastline is calculated using a geometric algorithm, and the unit is converted and the result is output.

[0026] Optionally, step S2 specifically includes:

[0027] Obtain remote sensing image block size parameters and the overlap rate parameter of the slice ;

[0028] The sliding window's step size is calculated based on the block size and block overlap ratio:

[0029]

[0030] in, This indicates a floor function that ensures the step size is an integer pixel value. , indicating the overlap ratio of adjacent windows;

[0031] For the original image size is For remote sensing imagery, calculate the number of grid rows and columns that can be completely divided:

[0032]

[0033]

[0034] For grid index Calculate the pixel coordinate range of the segmented region:

[0035]

[0036]

[0037]

[0038]

[0039] in, ;

[0040] Using the affine transformation parameters of the image pixel coordinates Convert to geographic projection coordinates :

[0041]

[0042]

[0043] Then, the projected coordinates are converted to WGS84 latitude and longitude coordinates through coordinate system transformation. :

[0044]

[0045] in, This represents a transformation function from a specific projected coordinate system to the WGS84 geographic coordinate system;

[0046] To ensure complete coverage of the image edge areas, three edge cases are processed:

[0047] Right edge processing conditions:

[0048]

[0049] Bottom edge processing conditions:

[0050]

[0051] Processing the bottom right corner area: Both of the above conditions must be met simultaneously;

[0052] in, ;

[0053] For each segmented slice, update its geographic transformation parameters:

[0054]

[0055]

[0056] Optionally, step S3 specifically includes:

[0057] Load the global land and sea database with a resolution of 0.01°, containing land and sea attribute information of 648,054,001 grid points;

[0058] For any latitude and longitude coordinates Calculate its index position in the global land and sea database:

[0059]

[0060] Among them, the longitude range is The latitude range is ;

[0061] Read the value at the corresponding position in the global land and sea database based on the index position. :

[0062]

[0063] For the coordinates of the four corner points of each preprocessed slice Count the number of ocean-attribute corner points:

[0064]

[0065] in, For indicator functions;

[0066] The filtering criteria are:

[0067]

[0068] Optionally, step S4 specifically includes:

[0069] For input images Perform linear grayscale stretching:

[0070]

[0071] in, ;

[0072] For input images Perform optimal linear stretching based on percentiles:

[0073]

[0074]

[0075]

[0076]

[0077]

[0078] For input images Perform square root stretching:

[0079]

[0080] Calculate information entropy:

[0081]

[0082] in, grayscale value The probability distribution;

[0083] Calculate the standard deviation (contrast index):

[0084]

[0085] Calculate the average gradient (sharpness index):

[0086]

[0087] Histogram uniformity:

[0088]

[0089] Design a comprehensive evaluation function to assess image quality:

[0090]

[0091] Optionally, step S5 specifically includes:

[0092] Based on image spatial resolution Select detection model:

[0093]

[0094] For input image size Adaptive detection strategy:

[0095]

[0096] For large images, a sliding window block detection method with a 30% overlap rate is used:

[0097]

[0098]

[0099] in, Input dimensions for the model;

[0100] For each detected target, calculate the key geometric properties:

[0101] Calculate the center pixel coordinates:

[0102]

[0103]

[0104] in, These are the coordinates of the four vertices of the rotated bounding box;

[0105] Calculate the target length:

[0106]

[0107]

[0108]

[0109]

[0110] Convert the pixel coordinates of the target center point to geographic coordinates:

[0111]

[0112]

[0113]

[0114] Optionally, step S6 specifically includes:

[0115] For the center point coordinates of each detected target Calculate its land and sea database index:

[0116]

[0117] Step 5.2: Target type matching and filtering. Based on configuration parameters. Target filtering:

[0118]

[0119] in, .

[0120] Optionally, step S7 specifically includes:

[0121] Determine the cutting dimensions based on the width and height of the target object:

[0122]

[0123] in, and These are the width and height of the target, respectively. To amplify the magnification factor, ensure that the cropped area contains sufficient background information surrounding the target;

[0124] Calculate the boundary of the clipping region with the target center point as the center:

[0125]

[0126]

[0127]

[0128]

[0129] When the cut area is smaller than the specified size, a zero-fill strategy is used:

[0130]

[0131]

[0132] like or Then create The zero matrix is ​​obtained, and the actual cropped data is copied to the corresponding position;

[0133] Update the geographic transformation parameters of the cropped tiles:

[0134]

[0135]

[0136] Optionally, step S8 specifically includes:

[0137] Based on the latitude and longitude coordinates of the target point, calculate its UTM projection zone number and determine the corresponding EPSG code:

[0138]

[0139]

[0140] Project the target point and coastline data into the same UTM coordinate system:

[0141]

[0142]

[0143] in, Coastline data in the WGS84 coordinate system. For the corresponding UTM coordinates;

[0144] Calculate the shortest Euclidean distance from the target point to the coastline:

[0145]

[0146] in, These are the coordinates of discrete points on the coastline.

[0147] Convert distance from meters to kilometers:

[0148] Attached Figure Description

[0149] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0150] Figure 1 A schematic diagram of a remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering provided in an embodiment of the present invention;

[0151] Figure 2 This is an example image of the image preprocessing gridding segmentation result in an embodiment of the present invention;

[0152] Figure 3 This is a schematic diagram of the first-level land-sea segmentation and screening principle in an embodiment of the present invention;

[0153] Figure 4 These are comparison images of image enhancement algorithms in embodiments of the present invention.

[0154] Figure 5 This is an example diagram of the ship target detection and identification results in an embodiment of the present invention;

[0155] Figure 6 This is a schematic diagram of the second-level land-sea segmentation and screening principle in an embodiment of the present invention;

[0156] Figure 7 This is an example diagram of the adaptive cropping result in an embodiment of the present invention;

[0157] Figure 8 This is a diagram of the UTM projection band selection mechanism in an embodiment of the present invention. Detailed Implementation

[0158] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0159] like Figure 1 As shown, this embodiment of the invention provides a method for ship target detection and adaptive cropping in remote sensing images based on dual land-sea segmentation filtering, including the following steps:

[0160] S1. Parameter Configuration is responsible for providing the core parameter configurations for system operation. These parameters are stored in JSON format, and the specific parameter definitions are as follows:

[0161] : Cutting size.

[0162] : Block overlap rate.

[0163] Target type, defaults to "Marine target".

[0164] S2. Image preprocessing is responsible for dividing the input large-format remote sensing image into grids. The specific implementation process is as follows:

[0165] Obtain remote sensing image block size parameters and the overlap rate parameter of the slice ;

[0166] The sliding window's step size is calculated based on the block size and block overlap ratio:

[0167]

[0168] in, This indicates a floor function that ensures the step size is an integer pixel value. , indicating the overlap ratio of adjacent windows;

[0169] For the original image size is For remote sensing imagery, calculate the number of grid rows and columns that can be completely divided:

[0170]

[0171]

[0172] For grid index Calculate the pixel coordinate range of the segmented region:

[0173]

[0174]

[0175]

[0176]

[0177] in, ;

[0178] Using the affine transformation parameters of the image pixel coordinates Convert to geographic projection coordinates :

[0179]

[0180]

[0181] Then, the projected coordinates are converted to WGS84 latitude and longitude coordinates through coordinate system transformation. :

[0182]

[0183] in, This represents a transformation function from a specific projected coordinate system to the WGS84 geographic coordinate system;

[0184] To ensure complete coverage of the image edge areas, three edge cases are processed:

[0185] Right edge processing conditions:

[0186]

[0187] Bottom edge processing conditions:

[0188]

[0189] Processing the bottom right corner area: Both of the above conditions must be met simultaneously;

[0190] in, ;

[0191] For each segmented slice, update its geographic transformation parameters:

[0192]

[0193]

[0194] like Figure 2 As shown, the process of gridding large-format remote sensing imagery is illustrated.

[0195] S3. First-level land-sea segmentation and filtering: Rapidly filter land-sea attributes from the preprocessed image slices, such as... Figure 3 As shown, it specifically includes:

[0196] Load the global land and sea database with a resolution of 0.01°, containing land and sea attribute information of 648,054,001 grid points;

[0197] For any latitude and longitude coordinates Calculate its index position in the global land and sea database:

[0198]

[0199] Among them, the longitude range is The latitude range is ;

[0200] Read the value at the corresponding position in the global land and sea database based on the index position. :

[0201]

[0202] For the coordinates of the four corner points of each preprocessed slice Count the number of ocean-attribute corner points:

[0203]

[0204] in, For indicator functions;

[0205] The filtering criteria are:

[0206]

[0207] S4. Image Enhancement Processing: Adaptive enhancement processing is performed on the image slices that have passed the first-level screening, specifically including:

[0208] For input images Perform linear grayscale stretching:

[0209]

[0210] in, ;

[0211] For input images Perform optimal linear stretching based on percentiles:

[0212]

[0213]

[0214]

[0215]

[0216]

[0217] To enhance contrast in low grayscale areas, the input image Perform square root stretching:

[0218]

[0219] Calculate information entropy:

[0220]

[0221] in, grayscale value The probability distribution;

[0222] Calculate the standard deviation (contrast index):

[0223]

[0224] Calculate the average gradient (sharpness index):

[0225]

[0226] Histogram uniformity:

[0227]

[0228] Design a comprehensive evaluation function to assess image quality:

[0229]

[0230] like Figure 4 As shown, the comparison of different image enhancement algorithms is illustrated. By calculating a comprehensive evaluation function, the system can automatically select the enhancement algorithm most suitable for the current image features, ensuring optimal results for subsequent object detection.

[0231] Step S5: Ship target detection achieves high-precision ship target detection based on deep learning, specifically including:

[0232] Based on image spatial resolution Select detection model:

[0233]

[0234] For input image size Adaptive detection strategy:

[0235]

[0236] For large images, a sliding window block detection method with a 30% overlap rate is used:

[0237]

[0238]

[0239] in, Input dimensions for the model;

[0240] For each detected target, calculate the key geometric properties:

[0241] Calculate the center pixel coordinates:

[0242]

[0243]

[0244] in, These are the coordinates of the four vertices of the rotated bounding box;

[0245] Calculate the target length:

[0246]

[0247]

[0248]

[0249]

[0250] Convert the pixel coordinates of the target center point to geographic coordinates:

[0251]

[0252]

[0253]

[0254] like Figure 5 As shown, the ship target detection results are displayed.

[0255] Step S6: Second-level land-sea segmentation and filtering

[0256] like Figure 6 The diagram illustrates the working principle of the second-level land-sea segmentation and filtering module. Unlike the first-level filtering, the second-level filtering performs precise land-sea attribute determination on specific detected targets. It queries a global land-sea database using the geographic coordinates of the target's center point to achieve more accurate target classification and false alarm removal. Specifically, this includes:

[0257] For the center point coordinates of each detected target Calculate its land and sea database index:

[0258]

[0259] Step 5.2: Target type matching and filtering. Based on configuration parameters. Target filtering:

[0260]

[0261] in, .

[0262] Step S7: Adaptive cropping process intelligently crops the ship targets that have passed the double screening, specifically including:

[0263] Determine the cutting dimensions based on the width and height of the target object:

[0264]

[0265] in, and These are the width and height of the target, respectively. To amplify the magnification factor, ensure that the cropped area contains sufficient background information surrounding the target;

[0266] Calculate the boundary of the clipping region with the target center point as the center:

[0267]

[0268]

[0269]

[0270]

[0271] When the cut area is smaller than the specified size, a zero-fill strategy is used:

[0272]

[0273]

[0274] like or Then create The zero matrix is ​​obtained, and the actual cropped data is copied to the corresponding position;

[0275] Update the geographic transformation parameters of the cropped tiles:

[0276]

[0277]

[0278] like Figure 7 The image shows an example of the results of adaptive cropping. The cropping process uses the target center as a reference to ensure that the target is displayed in the center of the cropped result, while maintaining sufficient background information for subsequent analysis. The cropping size is adaptively adjusted according to the target size.

[0279] Step S8: Coastline Distance Calculation This step calculates the distance from the ship target to the nearest coastline, specifically including:

[0280] like Figure 8The diagram illustrates the UTM projection zone selection mechanism. Based on the latitude and longitude coordinates of the target point, the UTM projection zone number is calculated, and the corresponding EPSG code is determined.

[0281]

[0282]

[0283] Project the target point and coastline data into the same UTM coordinate system:

[0284]

[0285]

[0286] in, Coastline data in the WGS84 coordinate system. For the corresponding UTM coordinates;

[0287] Calculate the shortest Euclidean distance from the target point to the coastline:

[0288]

[0289] in, These are the coordinates of discrete points on the coastline.

[0290] Convert distance from meters to kilometers:

[0291]

[0292] Although the present invention has been disclosed above with reference to embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering, comprising the following steps: Step S1: Parameter Configuration Key parameters, including the size of large-format remote sensing image slices, can be set via a JSON configuration file. Slicing overlap rate parameter and target type selection ; Step S2: Image Preprocessing Read the input large-format remote sensing image file, obtain the spatial resolution, georeferenced information and pixel size of the remote sensing image, select appropriate block size parameters according to the resolution, perform grid segmentation using a sliding window method, calculate the pixel coordinate range and geographic coordinate range of each block, and establish coordinate mapping relationship; Step S3: First-level land-sea segmentation and filtering For each image block, extract the geographic coordinates of the four corner points, calculate the index position in the global land-sea database based on latitude and longitude, read the land-sea attribute identifiers corresponding to the corner points, discard the block if all four corner points are land, otherwise retain it for subsequent processing, count the filtering results, and record the number of retained ocean blocks; Step S4: Image Enhancement Processing For each retained ocean slice, linear gray-scale stretching, optimal linear stretching, and square root stretching algorithms are applied to generate enhancement candidate results. The information entropy, standard deviation, average gradient, and histogram uniformity index of each candidate result are calculated. The optimal enhancement result is selected by weighted scoring. Step S5: Ship Target Detection The resolution-adaptive dual-model ONNX inference framework is used to detect ship targets in the enhanced ocean slices. Confidence thresholding and non-maximum suppression algorithms are applied to remove low-quality and duplicate detections, and the coordinates of the detection results are transformed back to the original image coordinate system. Step S6: Second-level land-sea segmentation and filtering For each detected ship target, extract its center point pixel coordinates and convert them into geographic coordinates. Query the attribute identifier of the center point in global land and sea data, retain only targets located in the ocean area, remove false alarms on land, and statistically analyze the change in the number of targets before and after filtering. Step S7: Adaptive cropping process Centered on the ship target that has passed the double screening verification, the system automatically calculates the optimal cropping range based on the ship target's bounding box size and geometric features, combined with the user-preset cropping size constraint parameters, and dynamically generates standardized local image fragments containing the complete target and appropriate contextual information. Step S8: Calculate the coastline distance The UTM projection zone and EPSG encoding of the ship target are determined based on its geographical coordinates. The coastline vector data of the corresponding area are read, and the coastline and target point are projected onto the UTM coordinate system. The shortest distance from the target point to the coastline is calculated using a geometric algorithm, and the unit is converted and the result is output.

2. The remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering according to claim 1, characterized in that, Step S2 specifically includes: Obtain remote sensing image block size parameters and the overlap rate parameter of the slice ; The sliding window's step size is calculated based on the block size and block overlap ratio: in, This indicates a floor function that ensures the step size is an integer pixel value. , indicating the overlap ratio of adjacent windows; For the original image size is For remote sensing imagery, calculate the number of grid rows and columns that can be completely divided: For grid index Calculate the pixel coordinate range of the segmented region: in, ; Using the affine transformation parameters of the image pixel coordinates Convert to geographic projection coordinates : Then, the projected coordinates are converted to WGS84 latitude and longitude coordinates through coordinate system transformation. : in, This represents a transformation function from a specific projected coordinate system to the WGS84 geographic coordinate system; To ensure complete coverage of the image edge areas, three edge cases are processed: Right edge processing conditions: Bottom edge processing conditions: Processing the bottom right corner area: Both of the above conditions must be met simultaneously; in, ; For each segmented slice, update its geographic transformation parameters:

3. The remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering according to claim 1, characterized in that, Step S3 specifically includes: Load the global land and sea database with a resolution of 0.01°, containing land and sea attribute information of 648,054,001 grid points; For any latitude and longitude coordinates Calculate its index position in the global land and sea database: Among them, the longitude range is The latitude range is ; Read the value at the corresponding position in the global land and sea database based on the index position. : For the coordinates of the four corner points of each preprocessed slice Count the number of ocean-attribute corner points: in, For indicator functions; The filtering criteria are:

4. The remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering according to claim 1, characterized in that, Step S4 specifically includes: For input images Perform linear grayscale stretching: in, ; For input images Perform optimal linear stretching based on percentiles: For input images Perform square root stretching: Calculate information entropy: in, grayscale value The probability distribution; Calculate the standard deviation (contrast index): Calculate the average gradient (sharpness index): Histogram uniformity: Design a comprehensive evaluation function to assess image quality:

5. The remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering according to claim 1, characterized in that, Step S5 specifically includes: Based on image spatial resolution Select detection model: For input image size Adaptive detection strategy: For large images, a sliding window block detection method with a 30% overlap rate is used: in, Input dimensions for the model; For each detected target, calculate the key geometric properties: Calculate the center pixel coordinates: in, These are the coordinates of the four vertices of the rotated bounding box; Calculate the target length: Convert the pixel coordinates of the target center point to geographic coordinates:

6. The remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering according to claim 1, characterized in that, Step S6 specifically includes: For the center point coordinates of each detected target Calculate its land and sea database index: Step 5.2: Target type matching and filtering. Based on configuration parameters. Target filtering: in, .

7. The remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering according to claim 1, characterized in that, Step S7 specifically includes: Determine the cutting dimensions based on the width and height of the target object: in, and These are the width and height of the target, respectively. To amplify the magnification factor, ensure that the cropped area contains sufficient background information surrounding the target; Calculate the boundary of the clipping region with the target center point as the center: When the cut area is smaller than the specified size, a zero-fill strategy is used: like or Then create The zero matrix is ​​obtained, and the actual cropped data is copied to the corresponding position; Update the geographic transformation parameters of the cropped tiles:

8. The remote sensing image ship target detection and adaptive cropping method based on dual land-sea segmentation filtering according to claim 1, characterized in that, Step S8 specifically includes: Based on the latitude and longitude coordinates of the target point, calculate its UTM projection zone number and determine the corresponding EPSG code: Project the target point and coastline data into the same UTM coordinate system: in, Coastline data in the WGS84 coordinate system. For the corresponding UTM coordinates; Calculate the shortest Euclidean distance from the target point to the coastline: in, These are the coordinates of discrete points on the coastline. Convert distance from meters to kilometers:

Images