SAR image ship detection and positioning method based on multi-slice integrated reasoning

Abstract

The SAR image ship detection and positioning method based on multi-slice integrated reasoning comprises the steps of performing slicing processing on a to-be-detected SAR image to obtain image slice sets of various sizes; selecting a corresponding pre-trained detection model according to a scene demand and a slice size demand of a to-be-detected SAR image; based on the image slice set of various sizes and a selected pre-trained detection model, obtaining a detection result on each image slice; performing false alarm suppression processing based on the detection result on each image slice to obtain a ship detection result; positioning based on a ship detection result to obtain longitude and latitude coordinates of the ship; wherein the pre-trained detection model is obtained by training the deep learning model based on the image and the ship in the image; according to the invention, accurate target detection on a large-scale SAR image is realized, and the false alarm rate of target detection on the SAR image is reduced; and accurate position information of the ship can be obtained.

Description

Technical Field

[0001] The invention relates to the field of intelligent interpretation of remote sensing image targets, and in particular to a SAR image ship detection and positioning method based on multi-slice integrated reasoning. Background Art

[0002] Ship detection technology in large-scale synthetic aperture radar (SAR) images is of great value in both military and civilian fields. However, due to the complexity of image background and simplicity of texture, ship detection in large-scale SAR images is very prone to false alarms, such as reefs, islands, sea clutter and land buildings with similar shapes.

[0003] Traditional SAR ship detection methods mainly detect ships by artificially designing features, such as the waves, texture, and scale of the ship. However, these methods have low detection efficiency and are not adaptable to the scene. The design of features needs to be modified according to the change of the scene, and false alarms are prone to occur in coastal areas. In addition, speckle noise and ship motion blur make some targets lose their inherent characteristics, which brings difficulties to traditional ship detection methods. At present, the research on various ship detection methods is mostly based on constant false alarm rate (CFAR) detection technology. In order to determine potential targets in the ocean, an efficient multi-scale CFAR detector is constructed using a generalized gamma distribution clutter model. For ship detection in large-scale SAR images, a super-pixel-level two-stage CFAR technology is developed. Traditional algorithms usually only provide high detection performance for specific scenes, and their flexibility is insufficient. In addition, these CFAR-based techniques rely heavily on the accuracy of clutter statistical models. Background contamination can cause inaccurate parameter calculations, resulting in serious missed detections.

[0004] With the development of deep learning technology, deep learning-based object detection has become more and more mature. Many researchers have applied deep learning object detection methods to ship detection in SAR images. In particular, the end-to-end network proposed in recent years has made it unnecessary to manually design features and enhanced the generalization ability of the model. Since its success in large-scale image detection, deep learning methods have been widely used in computer vision tasks such as image segmentation and object detection. In the field of object detection, deep learning object detection methods are mainly divided into anchor-based and anchor-free. The development of anchor-based methods started early, mainly including two-stage representative methods such as R-CNN and one-stage methods represented by SSD. The main feature of two-stage representative methods such as R-CNN is the accurate prediction position of the bounding box, and the main feature of one-stage methods represented by SSD is the fast detection speed. Later, because the design of using anchors is relatively inflexible for object detection, the idea of ​​replacing the anchor framework with key points emerged. For example, CornerNet and others are representative methods without anchors, which not only improve the detection accuracy but also reduce the number of model calculations.

[0005] In the early days, deep learning-based ship detection methods for SAR images were: patches were cropped from the entire SAR image, and then CNN was used to identify whether the cropped patches belonged to the target object. These methods could not overcome the impact of ship shape and size changes on detection accuracy. Later, with the development of target detection technology under deep learning, a series of anchor-based target detection frameworks were proposed. This problem was solved by using region-based CNN to fuse features at different levels. Recently, with the introduction of anchor-free technology, since anchor-free target detection technology has an inherent design advantage for target detection over a large scale span, which happens to be the target problem in large-scale SAR images. Therefore, many researchers have focused on applying anchor-free framework target detection technology to target detection under SAR images. However, compared with anchor-based target detection models, anchor-free target detection models have lower accuracy.

[0006] In large-scale SAR image ship detection, on the one hand, the size of the ship is relatively small relative to the entire image, so less information is mapped to the feature map, which will greatly reduce the performance of ship detection. On the other hand, large-scale SAR images have rich scenes, which can easily cause false alarms of ships, such as clutter on land, islands, reefs, waves, etc. In addition, rich ship scenes can also bring difficulties to detection, such as docking of ships and multiple ships moored side by side.

[0007] In summary, the existing technology for SAR image target detection has the problems of high false alarm rate and low accuracy. Summary of the invention

[0008] In order to solve the problems of high false alarm rate and low accuracy in SAR image target detection in the prior art, the present invention proposes a SAR image ship detection and positioning method based on multi-slice integrated reasoning, comprising:

[0009] Slice the SAR image to be detected to obtain image slice sets of various sizes;

[0010] According to the scene requirements and slice size requirements of the SAR image to be detected, select the corresponding pre-trained detection model;

[0011] Based on the image slice sets of multiple sizes and the selected pre-trained detection model, obtaining a detection result on each image slice;

[0012] Perform false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result;

[0013] Positioning is performed based on the ship detection results to obtain the longitude and latitude coordinates of the ship;

[0014] The pre-trained detection model is obtained by training a deep learning model based on remote sensing images or mission scene images, and ships in the remote sensing images or mission images.

[0015] Optionally, performing false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result includes:

[0016] Based on the detection results on each image slice, a pixel feature analysis method is combined to perform initial target screening;

[0017] The screened targets are subjected to multi-source information fusion to obtain fusion detection results, and the confidence of the fusion detection results is reset based on the confidence update formula;

[0018] Based on the confidence of the fusion detection results, the adaptive threshold screening method is used to perform secondary target screening to obtain the ship detection results.

[0019] Optionally, the initial target screening based on the detection results on each image slice combined with a pixel feature analysis method includes:

[0020] The detection results on each image slice are used to calculate the pixel average value of the target area and the background area in the bounding box, and the difference between the pixel average values ​​of the target area and the background area in the bounding box by using the pixel feature analysis method;

[0021] Eliminate the detection results whose difference is less than the threshold value;

[0022] The threshold is obtained by statistically analyzing the difference between the average pixel values ​​of the target area and the background area within the boundary box of the detection results on each image slice in history.

[0023] Optionally, fusing multi-source information of the screened target to obtain a fusion detection result includes:

[0024] According to the coordinate relationship between each image slice and the original image, the target bounding box in the local coordinate system is mapped to the global coordinate system;

[0025] The bounding boxes of the same target after mapping are subjected to weighted non-maximum suppression. On the basis of retaining the bounding box with the highest confidence, the confidences of all overlapping bounding boxes are weightedly calculated according to the intersection-over-union ratio of the overlapping bounding boxes and the optimal bounding box to obtain the fusion detection result.

[0026] Optionally, the confidence of the fusion detection result is based on the adaptive threshold screening method to perform secondary target screening to obtain the ship detection result, including:

[0027] Adopting the adaptive threshold screening method to eliminate the confidence of the fusion detection results that are less than the set confidence threshold;

[0028] The remaining fusion detection results are converted to the SAR image to be detected to obtain the ship detection results.

[0029] Optionally, the performing positioning based on the ship detection result to obtain the longitude and latitude coordinates of the ship includes:

[0030] Based on the pixel coordinates of the target center point in the ship detection results and the SAR radar parameters, the physical offset of the target center point in the distance and azimuth relative to the reference point is calculated;

[0031] Taking the reference point as the origin, based on the physical offset of the target center point in the distance direction and azimuth direction relative to the reference point, determine the coordinates of the target center point in the north-east coordinate system;

[0032] The coordinates of the target center point in the north-east coordinate system are transformed to obtain the longitude and latitude coordinates of the target center point.

[0033] Optionally, the training of the detection model includes:

[0034] Acquire remote sensing images or mission scene images in different scenes, and ships in remote sensing images or mission images;

[0035] Performing different slicing processes on the remote sensing image or the mission image, and the ship in the remote sensing image or the mission image, to obtain image slice sets of different sizes;

[0036] Divide each image slice set into a training set and a test set according to a set ratio;

[0037] Training the deep learning model based on the training set to obtain a preliminarily trained detection model;

[0038] The preliminarily trained detection model is tested based on the test set, and the preliminarily trained detection model that passes the test is used as the trained detection model corresponding to the slice size.

[0039] In another aspect, the present invention further provides a SAR image ship detection and positioning system based on multi-slice integrated reasoning, comprising:

[0040] A slicing module is used to slice the SAR image to be detected to obtain image slice sets of various sizes;

[0041] A selection module is used to select a corresponding pre-trained detection model according to the scene requirements and slice size requirements of the SAR image to be detected;

[0042] A detection module, configured to obtain a detection result on each image slice based on the image slice sets of multiple sizes and a selected pre-trained detection model;

[0043] A false alarm suppression module, used for performing false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result;

[0044] The positioning module is used to locate the ship based on the ship detection results and obtain the longitude and latitude coordinates of the ship;

[0045] The pre-trained detection model is obtained by training a deep learning model based on remote sensing images or mission scene images, and ships in the remote sensing images or mission images.

[0046] Optionally, the false alarm suppression module includes:

[0047] A pixel feature analysis submodule, used for performing initial target screening based on the detection results on each image slice in combination with a pixel feature analysis method;

[0048] The multi-source fusion submodule is used to fuse the multi-source information of the screened targets to obtain the fusion detection results, and reset the confidence of the fusion detection results based on the confidence update formula; and use the adaptive threshold screening method to perform secondary target screening based on the confidence of the fusion detection results to obtain the ship detection results. Optionally, the pixel feature analysis submodule is specifically used for:

[0049] The detection results on each image slice are used to calculate the pixel average value of the target area and the background area in the bounding box, and the difference between the pixel average values ​​of the target area and the background area in the bounding box by using the pixel feature analysis method;

[0050] Eliminate the detection results whose difference is less than the threshold value;

[0051] The threshold is obtained by statistically analyzing the difference between the average pixel values ​​of the target area and the background area within the boundary box of the detection results on each image slice in history.

[0052] Optionally, the multi-source fusion submodule is specifically used for:

[0053] According to the coordinate relationship between each image slice and the original image, the target bounding box in the local coordinate system is mapped to the global coordinate system;

[0054] The bounding boxes of the same target after mapping are subjected to weighted non-maximum suppression. On the basis of retaining the bounding box with the highest confidence, the confidences of all overlapping bounding boxes are weightedly calculated according to the intersection-over-union ratio of the overlapping bounding boxes and the optimal bounding box to obtain the fusion detection result.

[0055] The adaptive threshold screening method is used to eliminate the confidence of the fusion detection results that are less than the set confidence threshold; the remaining fusion detection results are converted to the SAR image to be detected to obtain the ship detection results.

[0056] Optionally, the positioning module is specifically used for:

[0057] Based on the pixel coordinates of the target center point in the ship detection results and the SAR radar parameters, the physical offset of the target center point in the distance and azimuth relative to the reference point is calculated;

[0058] Taking the reference point as the origin, based on the physical offset of the target center point in the distance direction and azimuth direction relative to the reference point, determine the coordinates of the target center point in the north-east coordinate system;

[0059] The coordinates of the target center point in the north-east coordinate system are transformed to obtain the longitude and latitude coordinates of the target center point.

[0060] Optionally, training modules are also included for:

[0061] Acquire remote sensing images or mission scene images in different scenes, and ships in remote sensing images or mission images;

[0062] Performing different slicing processes on the remote sensing image or the mission image, and the ship in the remote sensing image or the mission image, to obtain image slice sets of different sizes;

[0063] Divide each image slice set into a training set and a test set according to a set ratio;

[0064] Training the deep learning model based on the training set to obtain a preliminarily trained detection model;

[0065] The preliminarily trained detection model is tested based on the test set, and the preliminarily trained detection model that passes the test is used as the trained detection model corresponding to the slice size.

[0066] In another aspect, the present application further provides a computing device, comprising: at least one processor and a memory;

[0067] The memory is used to store one or more programs;

[0068] When the one or more programs are executed by the at least one processor, the SAR image ship detection and positioning method based on multi-slice integrated reasoning as described above is implemented.

[0069] On the other hand, the present application also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed, the method for ship detection and positioning in SAR images based on multi-slice integrated reasoning as described above is implemented.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] The present invention provides a SAR image ship detection and positioning method based on multi-slice integrated reasoning, comprising: slicing a SAR image to be detected to obtain an image slice set of various sizes; selecting a corresponding pre-trained detection model according to the scene requirements and slice size requirements of the SAR image to be detected; obtaining a detection result on each image slice based on the image slice set of various sizes and the selected pre-trained detection model; performing false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result; positioning based on the ship detection result to obtain the longitude and latitude coordinates of the ship; wherein the pre-trained detection model is obtained by training a deep learning model based on a remote sensing image or a mission scene image, and a ship in the remote sensing image or the mission image; the present invention realizes accurate detection of targets on large-scale SAR images, greatly reduces the false alarm rate of target detection on SAR images; and based on the ship detection result, accurate position information of the ship can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a flow chart of the SAR image ship detection and positioning method based on multi-slice integrated reasoning of the present invention;

[0073] Figure 2 It is a schematic diagram of the structure of the SAR image ship detection and positioning system based on multi-slice integrated reasoning of the present invention;

[0074] Figure 3 It is a schematic diagram of a false alarm suppression algorithm based on multi-source information fusion of the present invention;

[0075] Figure 4 Schematic diagram of a false alarm suppression algorithm based on pixel feature analysis of the present invention;

[0076] Figure 5 A visualization diagram of the detection results obtained by applying the SAR image ship detection method of the present invention;

[0077] Figure 6 It is a schematic diagram of the structure of the electronic device of the present invention. DETAILED DESCRIPTION

[0078] In the fields of maritime traffic management, marine environment monitoring, maritime search and rescue, etc., there is an urgent need for real-time monitoring and precise positioning of ship positions. To this end, this application proposes a SAR image ship detection and positioning method based on multi-slice integrated reasoning, which greatly improves the performance of ship detection in large-scale SAR images by fusing the detection results of different channels and suppressing false alarms.

[0079] The multi-size slice integrated reasoning framework corresponding to the SAR image ship detection method includes four modules: slicing module, detection module, false alarm suppression module and positioning module.

[0080] The slicing module cuts large-scale SAR images into slice sets of various sizes, which on the one hand prepares for feature scaling, and on the other hand can better keep the target from being cut.

[0081] By feeding slice sets of different sizes into the detection module, the target features can be scaled to a greater extent. The module also maintains a model library containing various types of algorithms that can be applied to different task scenarios and requirements.

[0082] In the false alarm suppression module, two sub-methods are proposed in a progressive structure: pixel feature analysis method (FAE-pff) and multi-source fusion method (FAE-msf). FAE-pff calculates the difference between the foreground and background of each bounding box to distinguish correct detections from false alarms. FAE-msf fuses the detection results from different sources and achieves false alarm suppression through weighted NMS and adaptive threshold screening.

[0083] In the positioning module, the detection results and SAR radar parameters of the target are input, and the physical offset of the target relative to the reference point is calculated. Then, the coordinates of the target point in the north-east coordinate system with the reference point as the origin are solved; finally, the longitude and latitude coordinates of the target are obtained.

[0084] Embodiment 1:

[0085] A ship detection method based on SAR image based on multi-size slice integrated reasoning, such as Figure 1 As shown, including:

[0086] Step S1: Slice the SAR image to be detected to obtain image slice sets of various sizes;

[0087] Step S2: Select a corresponding pre-trained detection model according to the scene requirements and slice size requirements of the SAR image to be detected;

[0088] Step S3: based on the image slice sets of various sizes and the selected pre-trained detection model, obtaining the detection results on each image slice;

[0089] Step S4: performing false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result;

[0090] Step S5: positioning the ship based on the ship detection result to obtain the longitude and latitude coordinates of the ship;

[0091] The pre-trained detection model is obtained by training a deep learning model based on remote sensing images or mission scene images, and ships in the remote sensing images or mission images.

[0092] This application proposes a SAR image ship detection and positioning method based on multi-slice integrated reasoning, which greatly improves the performance of ship detection in large-scale SAR images by fusing the detection results of different channels and suppressing false alarms. In this framework, a multi-size slicer is first used to slice the original large-scale image into slice sets of different sizes, which can expand the scale range of image feature expression.

[0093] Secondly, a detection model library was established to train a variety of models for use in order to cope with differentiated task scenarios such as different targets of interest, environmental backgrounds, and different slice sizes, including models trained using different parameter configurations and different data sets.

[0094] Then, a weighted non-maximum suppression (Weighted-NMS) method is proposed to achieve information complementarity and enhancement by performing bounding box fusion and confidence weighted update on the detection results of multiple channels.

[0095] Thirdly, two false alarm suppression methods are proposed, namely, pixel feature analysis and adaptive threshold screening, to reduce the false detection results in the output of the detection model. The pixel feature analysis method eliminates the detection results with unclear differences by calculating the difference between the foreground and background in each detection frame. The adaptive threshold screening method excludes the detection results with low confidence after fusion by dynamically updating the confidence threshold set by the original detection network. Experiments show that the proposed method can produce more correct detections while greatly reducing false detection results.

[0096] Finally, a target positioning method is proposed. Based on the detection result of the target and the SAR radar parameters, the physical offset of the target in the distance and azimuth relative to the reference point is solved, and then the coordinates of the target point in the north-east coordinate system with the reference point as the origin are solved. The coordinates of the target point in the north-east coordinate system are transformed to obtain the longitude and latitude coordinates of the target point. In this embodiment, the target center point is referred to as the target point.

[0097] SAR images are positioned based on ship detection results, which can meet the needs of these fields for high-precision, real-time positioning. Through ship detection results, the precise location information of the ship can be obtained, and combined with technical means such as geographic information system (GIS), real-time visualization and dynamic tracking of the ship's position can be achieved. This is of great significance for improving the efficiency of maritime traffic management, timely discovering and handling maritime safety incidents, and protecting the marine ecological environment.

[0098] The present invention will be further introduced below:

[0099] The present invention provides a SAR image ship detection and positioning method based on multi-slice integrated reasoning, comprising:

[0100] Step S1: Slice the SAR image to be detected to obtain image slice sets of various sizes, specifically including:

[0101] Large-scale SAR image slicing

[0102] 1-1: Select multiple cut image sizes according to the large-scale SAR image size.

[0103] 1-2: Use the slicing module to cut large-scale SAR images into SAR images of various sizes to form slice sets of various scales.

[0104] Step S2: According to the scene requirements of the SAR image to be detected and the slice size in step S1, select the corresponding pre-trained detection model, specifically including:

[0105] The pre-trained detection model here is obtained by training the deep learning model based on the remote sensing image or mission scene image and the ships in the remote sensing image or mission image. The specific training process is as follows:

[0106] Acquire remote sensing images or mission scene images in different scenes, and ships in remote sensing images or mission images;

[0107] constructing a data set with different slice sizes from the remote sensing image or the mission image and the ship in the remote sensing image or the mission image;

[0108] Dividing the data set into a training set and a test set according to a set ratio;

[0109] Training the deep learning model based on the training set to obtain a preliminarily trained detection model;

[0110] The preliminarily trained detection model is tested based on the test set, and the preliminarily trained detection model that passes the test is used as the trained detection model.

[0111] Step S3: based on the image slice set and the selected pre-trained detection model, obtaining the detection result on each image slice;

[0112] SAR image target detection:

[0113] 3-1: Deploy the trained image detection model to the detection module. The model is generated by a data set constructed from real remote sensing images or images of relevant mission scenes and other data in the early stage and a variety of machine learning training strategies. Select the most appropriate model for deployment based on the relevance of the target task and the training data.

[0114] 3-2: Input the sliced ​​SAR images into the corresponding pre-trained image detection models for detection.

[0115] Step S4: performing false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result, including:

[0116] Based on the detection results on each image slice, a pixel feature analysis method is combined to perform initial target screening;

[0117] The screened targets are subjected to multi-source information fusion to obtain fusion detection results, and the confidence of the fusion detection results is reset based on the confidence update formula;

[0118] Based on the confidence of the fusion detection results, the adaptive threshold screening method is used to perform secondary target screening to obtain the ship detection results.

[0119] Furthermore, the initial target screening based on the detection results on each image slice combined with the pixel feature analysis method includes:

[0120] The detection results on each image slice are used to calculate the pixel average value of the target area and the background area in the bounding box, and the difference between the pixel average values ​​of the target area and the background area in the bounding box by using the pixel feature analysis method;

[0121] Eliminate the detection results whose difference is less than the threshold value;

[0122] The threshold is obtained by statistically analyzing the difference between the average pixel values ​​of the target area and the background area within the boundary box of the detection results on each image slice in history.

[0123] Furthermore, the screened targets are subjected to multi-source information fusion to obtain fusion detection results, including:

[0124] According to the coordinate relationship between each image slice and the original image, the target bounding box in the local coordinate system is mapped to the global coordinate system;

[0125] The bounding boxes of the same target after mapping are subjected to weighted non-maximum suppression. On the basis of retaining the bounding box with the highest confidence, the confidences of all overlapping bounding boxes are weightedly calculated according to the intersection-over-union ratio of the overlapping bounding boxes and the optimal bounding box to obtain the fusion detection result.

[0126] Furthermore, the confidence level based on the fusion detection result is used to perform secondary target screening using an adaptive threshold screening method to obtain a ship detection result, including:

[0127] Adopting the adaptive threshold screening method to eliminate the confidence of the fusion detection results that are less than the set confidence threshold;

[0128] The remaining fusion detection results are converted to the SAR image to be detected to obtain the ship detection results.

[0129] Step S4 specifically includes false alarm suppression:

[0130] 4-1: In each channel, the pixel averages of the target area and the background area in the bounding box are calculated for the detection results on the slice.

[0131] 4-2: In each channel, calculate the difference between the average pixel values ​​of the target area and the background area within the bounding box.

[0132] 4-3: In each channel, set the threshold value to eliminate target detection results with lower difference.

[0133] 4-4: Through the weighted non-maximum suppression (Weighted-NMS) method, the target bounding boxes generated in different channels are first non-maximum suppressed, and the confidence of the optimal candidate box is updated using the intersection over union (IoU) between each overlapping box and the optimal candidate box as the confidence weight, thereby fusing the detection results of multiple channels and updating the domain value to eliminate targets with low confidence.

[0134] 4-5: According to the positional relationship between the slices retained when slicing the image and the original image, the detection results after false alarm suppression are converted to the original large-scale SAR image to obtain the final detection results.

[0135] Step S5: Based on the detection results and SAR radar parameters, target positioning is performed to obtain the longitude and latitude coordinates of the ship target;

[0136] 5-1: Input the ship detection results and SAR radar parameters. The detection results include the coordinates of the upper left corner of the target and the width and height of the target frame; the radar parameters include the longitude and latitude of the reference point, the aircraft yaw angle, the ideal heading angle, the azimuth sampling interval, the range sampling interval, the aircraft height, the ground altitude, the scene center slant range, the aircraft speed, the pulse repetition frequency, the downward viewing angle, etc.

[0137] 5-2: Based on the ship detection results, calculate the pixel center coordinates of the target.

[0138] 5-3: Calculate the physical offset between the target point and the reference point based on the target pixel center coordinates and sampling interval, aircraft altitude, and ground altitude.

[0139]

[0140] 5-4: Solve the coordinates of the target point in the north-east coordinate system with the reference point as the origin.

[0141] 5-5: Through coordinate system conversion, the coordinates of the target point in the geodetic coordinate system, namely longitude, latitude and altitude, are obtained.

[0142] The present invention firstly processes the large-scale SAR image by multi-size slices, then uses a variety of detection models to detect the sliced ​​image, then suppresses false alarms for the detection results in each channel, then fuses the multi-channel detection results and further suppresses false alarms, and finally uses the positioning module to obtain the longitude and latitude coordinates of the detection results. The problem of accurate detection of targets on large-scale SAR images is solved, the false alarm rate of target detection on SAR images is greatly reduced, and accurate location information of the detection results can be obtained.

[0143] Embodiment 2:

[0144] The main flow chart of a SAR image ship detection and positioning method based on multi-slice integrated reasoning is as follows: Figure 2 As shown, the false alarm suppression module in this method is mainly composed of two algorithms, and the specific details of the process are as follows Figure 3 and Figure 4 shown.

[0145] Step 1: Large-scale image slicing:

[0146] 1-1: Given a large-scale SAR image I.

[0147] 1-2: Export the large-scale SAR image to the slicer module to obtain image slice sets of three sizes: in Represents all image slices at three slice sizes.

[0148] Step 2: Image Detection:

[0149] 2-1: The trained detection model {m1,m2…,m n}Deployed to the detection module.

[0150] 2-2: According to the scene requirements and slice size of large-scale SAR images, select the corresponding models m1, m2, and m3. For example, when the target task is to detect fishing boats on the sea surface, select the model trained using fishing boats in the detection model library.

[0151] 2-3: Input the sliced ​​images into three detection models to obtain detection results of three sizes in Representative detection results in all image slices at three slice sizes.

[0152] Step 3: False alarm suppression, such as Figure 3 As shown:

[0153] 3-1: Input the detection results on the slice into the pixel feature analysis algorithm to calculate the pixel difference between the target area and the background area in each target bounding box.

[0154] 3-2: According to the optimal threshold n of pixel difference calculated in the training set, set the domain value to eliminate targets with small pixel difference.

[0155] 3-3: Input the three sizes of objects after elimination into the multi-source information fusion module, and perform weighted fusion on the bounding box confidence of the same object. And update the confidence weight according to the following formula:

[0156] c i =max(c1,c2,...,c n )

[0157] c update =c i +∑c j IoU ij

[0158] In the formula, c i is the maximum value of the confidence of the detection box when a target is detected n times, which corresponds to the optimal detection box, where i takes values ​​of 1, 2…, j, n, c update is the updated confidence of the optimal detection box, IoU ij is the intersection-over-union ratio of the j-th detection box and the i-th detection box.

[0159] 3-4: Reset the confidence of the model detection result output. Tunnel represents the number of channels set. See the formula:

[0160]

[0161] Where, T update represents the updated confidence screening threshold, T origin Represents the confidence filter threshold before updating.

[0162] 3-5: Eliminate targets with confidence lower than the threshold value to obtain the final target result.

[0163] 3-6: Map the final detection results to a large-scale image for image result display, such as Figure 5 shown.

[0164] Step 4: Positioning the Module

[0165] Input the detection results and SAR radar parameters to locate the target and obtain the latitude and longitude coordinates of the target.

[0166] Embodiment 3:

[0167] The present invention based on the same inventive concept also provides a SAR image ship detection and positioning system based on multi-slice integrated reasoning, comprising:

[0168] A slicing module is used to slice the SAR image to be detected to obtain image slice sets of various sizes;

[0169] A selection module is used to select a corresponding pre-trained detection model according to the scene requirements and slice size requirements of the SAR image to be detected;

[0170] A detection module, configured to obtain a detection result on each image slice based on the image slice sets of multiple sizes and a selected pre-trained detection model;

[0171] A false alarm suppression module, used for performing false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result;

[0172] The positioning module is used to locate the ship based on the ship detection results and obtain the longitude and latitude coordinates of the ship;

[0173] The pre-trained detection model is obtained by training a deep learning model based on remote sensing images or mission scene images, and ships in the remote sensing images or mission images.

[0174] Optionally, the false alarm suppression module includes:

[0175] A pixel feature analysis submodule, used for performing initial target screening based on the detection results on each image slice in combination with a pixel feature analysis method;

[0176] The multi-source fusion submodule is used to fuse the multi-source information of the screened targets to obtain the fusion detection results, and reset the confidence of the fusion detection results based on the confidence update formula; and use the adaptive threshold screening method to perform secondary target screening based on the confidence of the fusion detection results to obtain the ship detection results. Optionally, the pixel feature analysis submodule is specifically used for:

[0177] The detection results on each image slice are used to calculate the pixel average value of the target area and the background area in the bounding box, and the difference between the pixel average values ​​of the target area and the background area in the bounding box by using the pixel feature analysis method;

[0178] Eliminate the detection results whose difference is less than the threshold value;

[0179] The threshold is obtained by statistically analyzing the difference between the average pixel values ​​of the target area and the background area within the boundary box of the detection results on each image slice in history.

[0180] Optionally, the multi-source fusion submodule is specifically used for:

[0181] According to the coordinate relationship between each image slice and the original image, the target bounding box in the local coordinate system is mapped to the global coordinate system;

[0182] The bounding boxes of the same target after mapping are subjected to weighted non-maximum suppression. On the basis of retaining the bounding box with the highest confidence, the confidences of all overlapping bounding boxes are weightedly calculated according to the intersection-over-union ratio of the overlapping bounding boxes and the optimal bounding box to obtain the fusion detection result.

[0183] The adaptive threshold screening method is used to eliminate the confidence of the fusion detection results that are less than the set confidence threshold; the remaining fusion detection results are converted to the SAR image to be detected to obtain the ship detection results.

[0184] Optionally, the positioning module is specifically used for:

[0185] Based on the pixel coordinates of the target center point in the ship detection results and the SAR radar parameters, the physical offset of the target center point in the distance and azimuth relative to the reference point is calculated;

[0186] Taking the reference point as the origin, based on the physical offset of the target center point in the distance direction and azimuth direction relative to the reference point, determine the coordinates of the target center point in the north-east coordinate system;

[0187] The coordinates of the target center point in the north-east coordinate system are transformed to obtain the longitude and latitude coordinates of the target center point.

[0188] Optionally, training modules are also included for:

[0189] Acquire remote sensing images or mission scene images in different scenes, and ships in remote sensing images or mission images;

[0190] Performing different slicing processes on the remote sensing image or the mission image, and the ship in the remote sensing image or the mission image, to obtain image slice sets of different sizes;

[0191] Divide each image slice set into a training set and a test set according to a set ratio;

[0192] Training the deep learning model based on the training set to obtain a preliminarily trained detection model;

[0193] The preliminarily trained detection model is tested based on the test set, and the preliminarily trained detection model that passes the test is used as the trained detection model corresponding to the slice size.

[0194] Embodiment 4:

[0195] like Figure 6 As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0196] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding functions, so as to implement the steps of the SAR image ship detection and positioning method based on multi-slice integrated reasoning in the above-mentioned embodiment.

[0197] Embodiment 5:

[0198] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in the electronic device for storing programs and data. It can be understood that the storage medium here can include both the built-in storage medium in the electronic device and the extended storage medium supported by the electronic device. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of the SAR image ship detection and positioning method based on multi-slice integrated reasoning in the above embodiment.

[0199] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0200] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0201] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0202] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0203] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of the claims of the present invention to be approved.

Claims

1. A SAR image ship detection and positioning method based on multi-slice integrated reasoning, characterized in that: include: Slice the SAR image to be detected to obtain image slice sets of various sizes; According to the scene requirements and slice size requirements of the SAR image to be detected, select the corresponding pre-trained detection model; Based on the image slice sets of multiple sizes and the selected pre-trained detection model, obtaining a detection result on each image slice; Perform false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result; Positioning is performed based on the ship detection results to obtain the longitude and latitude coordinates of the ship; The pre-trained detection model is obtained by training a deep learning model based on remote sensing images or mission scene images, and ships in the remote sensing images or mission images.

2. The method according to claim 1, characterized in that The performing false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result includes: Based on the detection results on each image slice, a pixel feature analysis method is combined to perform initial target screening; The screened targets are subjected to multi-source information fusion to obtain fusion detection results, and the confidence of the fusion detection results is reset based on the confidence update formula; Based on the confidence of the fusion detection results, the adaptive threshold screening method is used to perform secondary target screening to obtain the ship detection results.

3. The method according to claim 2, characterized in that The initial target screening based on the detection results on each image slice combined with the pixel feature analysis method includes: The detection results on each image slice are used to calculate the pixel average value of the target area and the background area in the bounding box, and the difference between the pixel average values ​​of the target area and the background area in the bounding box by using the pixel feature analysis method; Eliminate the detection results whose difference is less than the threshold value; The threshold is obtained by statistically analyzing the difference between the average pixel values ​​of the target area and the background area within the boundary box of the detection results on each image slice in history.

4. The method according to claim 2, characterized in that The screened targets are subjected to multi-source information fusion to obtain fusion detection results, including: According to the coordinate relationship between each image slice and the original image, the target bounding box in the local coordinate system is mapped to the global coordinate system; The bounding boxes of the same target after mapping are subjected to weighted non-maximum suppression. On the basis of retaining the bounding box with the highest confidence, the confidences of all overlapping bounding boxes are weightedly calculated according to the intersection-over-union ratio of the overlapping bounding boxes and the optimal bounding box to obtain the fusion detection result.

5. The method according to claim 2, characterized in that The confidence of the fusion detection result is based on the adaptive threshold screening method to perform secondary target screening to obtain the ship detection result, including: Adopting the adaptive threshold screening method to eliminate the confidence of the fusion detection results that are less than the set confidence threshold; The remaining fusion detection results are converted to the SAR image to be detected to obtain the ship detection results.

6. The method according to claim 1, characterized in that The positioning based on the ship detection result to obtain the longitude and latitude coordinates of the ship includes: Based on the pixel coordinates of the target center point in the ship detection results and the SAR radar parameters, the physical offset of the target center point in the distance and azimuth relative to the reference point is calculated; Taking the reference point as the origin, based on the physical offset of the target center point in the distance direction and azimuth direction relative to the reference point, determine the coordinates of the target center point in the north-east coordinate system; The coordinates of the target center point in the north-east coordinate system are transformed to obtain the longitude and latitude coordinates of the target center point.

7. The method according to claim 1, characterized in that The training of the detection model includes: Acquire remote sensing images or mission scene images in different scenes, and ships in remote sensing images or mission images; Performing different slice processing on the remote sensing image or the mission image, and the ship in the remote sensing image or the mission image, to obtain image slice sets of different sizes; Divide each image slice set into a training set and a test set according to a set ratio; Training the deep learning model based on the training set to obtain a preliminarily trained detection model; The preliminarily trained detection model is tested based on the test set, and the preliminarily trained detection model that passes the test is used as the trained detection model corresponding to the slice size.

8. A SAR image ship detection and positioning system based on multi-slice integrated reasoning, characterized in that: include: A slicing module is used to slice the SAR image to be detected to obtain image slice sets of various sizes; A selection module is used to select a corresponding pre-trained detection model according to the scene requirements and slice size requirements of the SAR image to be detected; A detection module, configured to obtain a detection result on each image slice based on the image slice sets of multiple sizes and a selected pre-trained detection model; A false alarm suppression module, used for performing false alarm suppression processing based on the detection results on each image slice to obtain a ship detection result; The positioning module is used to locate the ship based on the ship detection results and obtain the longitude and latitude coordinates of the ship; The pre-trained detection model is obtained by training a deep learning model based on remote sensing images or mission scene images, and ships in the remote sensing images or mission images.

9. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a SAR image ship detection and positioning method based on multi-slice integrated reasoning as described in any one of claims 1 to 7 is implemented.

10. A readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, a SAR image ship detection and positioning method based on multi-slice integrated reasoning as described in any one of claims 1 to 7 is implemented.

Images