A SAR ship detection method based on distance compression domain characteristics guidance

Abstract

The present application relates to the technical field of SAR image target detection, and particularly relates to a SAR ship detection method based on distance compression domain characteristics guidance; comprising: step 1, obtaining spaceborne SAR echo data after BAQ decompression processing; step 2, performing matching filtering processing in the distance direction, thereby obtaining corresponding distance compression domain preliminary imaging; step 3, using a 1024*1024 sliding window for sliding window; step 4, adding a strip attention mechanism module in the main network ResNet50 of the RT-DETR benchmark model; step 5, adding a SimVSS module in the hybrid encoder of the RT-DETR benchmark model; step 6, inputting the features into the IoU perception query selection of the RT-DETR; step 7, performing multi-round optimization on the screened candidate boxes; step 8, adding a shape perception IoU regression positioning loss. The amplitude gradient and geometric scale characteristics of the ship target in the distance compression domain are utilized, the detection capability of the ship target is effectively enhanced, the detection performance is significantly improved, and the ship target can be more accurately identified.

Description

Technical Field

[0001] This invention relates to the field of SAR image target detection technology, and specifically to a SAR ship detection method guided by range compression domain characteristics. Background Technology

[0002] Oceans cover 71.8% of the Earth's surface, constituting humanity's second living space beyond land. Ships, as the core tools of water transportation, bear multiple functions including transportation, fishing activities, and national defense security; therefore, effective ship surveillance is particularly important. Currently, Synthetic Aperture Radar (SAR) technology occupies a key position in the field of ship target monitoring and identification, mainly due to its ability to generate high-resolution images with global coverage and to achieve continuous observation under all-weather, all-time conditions. Therefore, developing a reliable and efficient SAR ship detection algorithm is of paramount strategic importance for enhancing maritime safety management and monitoring effectiveness.

[0003] Traditional SAR image target detection first performs two-dimensional imaging processing on the acquired raw SAR echo domain data to generate a focused image similar to a visible light image. Then, ship targets are automatically detected and located from the focused image using traditional or deep learning target detection techniques. However, SAR ship detection methods based on imaging paradigms face challenges in practical applications of wide-area maritime ship detection, including time-consuming full-scene imaging calculations and poor image quality caused by defocusing of moving targets and sidelobe interference. Summary of the Invention

[0004] This invention provides a SAR ship detection method guided by range compression domain characteristics. Its purpose is to improve the timeliness of maritime ship target situational awareness by first completing the detection of a specific area with a target in the range compression domain, and then rapidly imaging only a specific local area. This solves the problem of low efficiency caused by invalid data processing in traditional technologies.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: This invention provides a SAR ship detection method guided by range compression domain characteristics, comprising the following steps: Step 1: Obtain the spaceborne SAR echo data after BAQ decompression. Step 2: Based on the above spaceborne SAR echo data, perform matched filtering in the range direction to obtain the corresponding range compression domain preliminary imaging; Step 3: In the distance compression domain image, use a sliding window of 1024×1024; Step 4: Add a strip attention mechanism module to the ResNet50 backbone network of the RT-DETR baseline model, and extract three scale features from the target in the 1024×1024 distance compression domain image, namely multi-scale feature maps sampled at 8, 16, and 32 times. ; Step 5: Add the SimVSS module to the hybrid encoder of the RT-DETR baseline model, and combine the above-mentioned 8, 16, and 32-fold multi-scale feature maps. The data is fed into a hybrid encoder based on the SimVSS module for multi-scale feature extraction, and then the multi-scale features are fused across scales. Step 6: Input the features output by the hybrid encoder into the IoU perceptual query selection of RT-DETR, and filter out the regions with high overlap (IoU) with the real target as the initial candidate boxes; Step 7: The Transformer decoder of RT-DETR performs multiple rounds of optimization on the selected candidate boxes. It analyzes the relationship between targets through a self-attention mechanism, gradually adjusts the position and size of the candidate boxes, and finally outputs accurate detection results. Step 8: In the RT-DETR model, in order to accurately locate the slender ship target in the range compression domain, a shape-aware IoU regression localization loss is added to the auxiliary prediction head structure.

[0006] Furthermore, the strip attention mechanism module described in step four specifically includes the following steps: Step 4.1, Define the strip attention mechanism module: Given input features First, perform three 1×1 convolution operations on it to obtain... , and These three feature maps; for feature maps Perform matrix dimension transformation and transpose operations to generate , for Then, the feature map Strip Pooling is used, which involves averaging the pixels in each row of the feature map to generate... Subsequently, the feature map and feature map The product operation is performed, and finally the position attention mapping result is obtained through the Softmax operation. Essentially, feature maps and The matrix multiplication operation is the product of the feature vector corresponding to a position and the feature vectors of all other positions. Therefore, the formula for calculating the similarity between positions is: (1.1)

[0007] in, Representation of feature map The Each position element; Representation of feature map The Each position element express and The degree of correlation between them; Furthermore, the feature map Perform Strip Pooling along the horizontal direction, then transform the dimension. , and matrix After matrix multiplication, the dimension is transformed ; Final feature map Input features Adding corresponding elements together yields the final feature map. ; Step 4.2, the embedding position of the SAM module: In the BasicBlock or Bottleneck structure of ResNet, the SAM module is inserted after the last convolutional layer of the residual path, and the channel weighting operation is performed before the addition operation with the skip connection. Step 4.3, Multi-level SAM module configuration: Add a SAM module to the output of each stage of ResNet to form a global-local joint attention mechanism.

[0008] Furthermore, the SimVSS module described in step five specifically includes the following steps: Step 5.1, define the SimVSS module: its design is mainly based on SSM and a feedforward network; this structure includes residual connections and normalization layers, designed to stabilize the gradient training process and accelerate model convergence; the traditional SSM is a linear time-invariant system function, which uses intermediate variables Univariate sequences Mapping to output sequence Specifically, given a state transition matrix As evolutionary factors, the weight matrix and the observation matrix These are respectively used as projection factors and defined skip connections. Its mathematical expression is as follows: (1.2)

[0009] In addition, the system function introduces a time-scale parameter. To discretize the data for processing discrete time series data; this transformation is defined as follows: (1.3)

[0010] in, This represents the mapping from the state space to the output space; it is used for refinement using a first-order Taylor series expansion. Approximation: (1.4)

[0011] For input The processing steps within the SimVSS block are described as follows: (1.5)

[0012] in, , , These represent the inputs at different layers of the SimVSS block. The output state; Step 5.2, the SimVSS module is embedded in the middle position between attention-based internal scale feature interaction and convolutional neural network-based cross-scale feature fusion.

[0013] Furthermore, the shape-aware IoU regression loss described in step eight is specifically as follows: Shape-aware IoU regression loss calculates the loss by focusing on the shape and scale of the bounding box itself, thus enabling more accurate bounding box regression. The calculation formula is shown below: (1.6)

[0014] in, It is a scale factor. and These are the prediction box and the truth box. and These are the prediction boxes. truth boxes width, and These are the prediction boxes. truth boxes height, and These are the weighting coefficients in the horizontal and vertical directions, respectively.

[0015] The beneficial effects achieved by this invention are as follows: 1) Performance improvement: By utilizing the amplitude gradient and geometric scale characteristics of ship targets in the range compression domain, the detection capability of ship targets is effectively enhanced, the detection performance is significantly improved, and ship targets can be identified more accurately.

[0016] 2) Feature Representation: The design of the strip attention mechanism module is well-suited to the elongated shape of ship targets, effectively representing their shape features. This helps to accurately distinguish ship targets in complex backgrounds and avoids detection errors caused by shape misjudgment.

[0017] 3) Receptive field expansion and global information learning: Feature fusion based on the SimVSS module is added to the hybrid encoder to expand the receptive field of the detection network, enabling the network to learn global contextual information between the ship target and other targets and the background, thereby better understanding the environment in which the target is located and improving the reliability and accuracy of detection.

[0018] 4) Enhanced localization capability: The introduction of shape-aware IoU regression loss optimizes the detection network's prediction of target location, significantly enhancing the network's localization capability and ensuring more accurate detection of ship target locations.

[0019] 5) Highly efficient data processing: Compared with the traditional SAR processing method of imaging first and then detecting, this invention can complete the localization of suspected ship target areas only in the range compression domain of SAR echoes, eliminate a large number of background areas in advance, greatly reduce the amount of data processing in subsequent imaging steps, and realize low-latency and rapid extraction of information of interest such as target slices from SAR satellite echoes.

[0020] 6) Broad application prospects: This method lays a solid foundation for the future on-orbit imaging, intelligent level improvement and on-orbit application efficiency of SAR satellites, and helps to promote the more efficient and intelligent application of SAR satellites in fields such as ocean monitoring and target reconnaissance. Attached Figure Description

[0021] 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 the structures shown in these drawings without creative effort.

[0022] Figure 1 The diagram shows a target detection grid structure of the method of the present invention.

[0023] Figure 2 The diagram shown is a structural diagram of the strip attention mechanism module of the method of the present invention.

[0024] Figure 3 The diagram shown is a structural diagram of the SimVSS module in the method of this invention. Detailed Implementation

[0025] The technical solutions in 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 a part of the embodiments of the present invention, and not all of them. 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.

[0026] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0027] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, if the word "and / or" appears throughout the text, it means including three parallel solutions; for example, "A and / or B" includes solution A, solution B, or a solution that simultaneously satisfies A and B. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0028] like Figures 1-3 As shown, this invention provides a SAR ship detection method guided by range compression domain characteristics, comprising the following steps: Step 1: Obtain spaceborne SAR (synthetic aperture radar) echo data after BAQ (block adaptive quantization) decompression processing; Step 2: Based on the above spaceborne SAR echo data, perform matched filtering in the range direction to obtain the corresponding range compression domain preliminary imaging; Step 3: In the distance compression domain image, use a sliding window of 1024×1024; Step 4, as follows Figure 1 As shown, a strip attention mechanism module is added to the ResNet50 backbone network of the RT-DETR baseline model, and three scale features are extracted from the target in the 1024×1024 distance compression domain image, namely multi-scale feature maps sampled at 8, 16, and 32 times. ; Step 5, as follows Figure 1As shown, a SimVSS module is added to the hybrid encoder of the RT-DETR baseline model, and the aforementioned 8, 16, and 32-fold multi-scale feature maps are used. The data is fed into a hybrid encoder based on the SimVSS module for multi-scale feature extraction, and then the multi-scale features are fused across scales. Step 6, as follows Figure 1 As shown, the features output by the hybrid encoder are input into the IoU-aware query selection of RT-DETR, and regions with high IoU with the real target are selected as initial candidate boxes. Generally speaking, in object detection tasks, an IoU greater than 0.5 can be considered a relatively high degree of overlap, but the specific threshold will vary depending on the dataset and task requirements.

[0029] Step 7, as follows Figure 1 As shown, RT-DETR's Transformer decoder performs multiple rounds of optimization on the selected candidate boxes, analyzes the relationships between targets through a self-attention mechanism, and gradually adjusts the position and size of the candidate boxes to finally output accurate detection results; Step 8, as follows Figure 1 As shown, in the RT-DETR model, to accurately locate slender, range-compressed ship targets, a shape-aware IoU regression localization loss is added to the auxiliary prediction head structure. The specific process is as follows: In step 7, the Transformer decoder optimizes the position and size information of the candidate boxes to obtain initially optimized candidate boxes; then, in step 8, the auxiliary prediction head structure uses these optimized candidate boxes as the base region for calculating the shape-aware IoU regression localization loss; based on this, the auxiliary prediction head performs more refined localization adjustments, thereby achieving accurate detection and localization of slender, range-compressed ship targets.

[0030] As a further preferred embodiment of the SAR ship detection method based on range compression domain characteristics guided by the present invention, the strip attention mechanism module described in step four specifically includes the following steps: Step 4.1, define the Strip Attention Module (SAM). For example... Figure 2 As shown, the SAM module can fully utilize the Strip Pooling technique to explore features in depth along the same spatial dimension, thereby significantly enhancing the ability to express the strip contextual features of ship targets in the range compression domain.

[0031] Given input features ( and These are the height and width, respectively. For the number of channels, To represent the real number field, first perform three 1×1 convolution operations on it to obtain... , and These three feature maps. (Regarding the feature maps...) Perform matrix dimension transformation and transpose operations to generate , for Then, the feature map Strip Pooling is used, which involves averaging the pixels in each row of the feature map to generate... Then, the feature map and feature map The product operation is performed, and finally the position attention mapping result is obtained through the Softmax operation. Essentially, feature maps and The matrix multiplication operation is the product of the feature vector corresponding to a position and the feature vectors of all other positions. Therefore, the formula for calculating the similarity between positions is: (1.1)

[0032] in, Representation of feature map The Each position element; Representation of feature map The Each position element express and The degree of correlation between them.

[0033] Furthermore, the feature map Perform Strip Pooling along the horizontal direction, then transform the dimension. , and matrix After matrix multiplication, the dimension is transformed Final feature map Input features Adding corresponding elements together yields the final feature map. .

[0034] Step 4.2, the embedding position of the SAM module: In the BasicBlock or Bottleneck structure of ResNet, the SAM module is inserted after the last convolutional layer of the residual path, and the channel weighting operation is performed before the addition operation with the Skip Connection.

[0035] Step 4.3, Multi-level SAM Module Configuration: A SAM module is added to the output of each stage of ResNet to form a global-local joint attention mechanism. For example, the dimensionality reduction ratio is set to [value missing] for layers 2 to 4 of ResNet50. The SAM module.

[0036] As a further preferred embodiment of the SAR ship detection method based on range compression domain characteristics guided by the present invention, the SimVSS module mentioned in step five specifically includes the following steps: Step 5.1, define the SimVSS module. For example... Figure 3 As shown, the SimVSS module in the hybrid encoder is used to extend the receptive field of the detection network, enabling it to learn global contextual information between the ship target and other targets and the background.

[0037] Its design is primarily based on a State-Space Model (SSM) and a feedforward network. This structure includes residual connections and normalization layers, designed to stabilize the gradient training process and accelerate model convergence. A traditional SSM can be viewed as a linear time-invariant system function, which uses intermediate variables... Univariate sequences Mapping to output sequence Specifically, given a state transition matrix As evolutionary factors, the weight matrix and the observation matrix These are respectively used as projection factors and defined skip connections. Its mathematical expression is as follows: (1.2)

[0038] in, The dimension of the state variable; It is a time variable; State variables describe the system's state over time. The internal state of.

[0039] Furthermore, the system function can be modified by introducing a time-scale parameter. To discretize the data for processing discrete-time series data. This transformation can be defined as follows: (1.3)

[0040] in, Represents the discrete-time state vector, at time step State variables; The discretized state transition matrix is ​​represented by the continuous-time state matrix. Through matrix exponent The resulting value is used to describe the evolution of states in a discrete-time system. Represents the state vector at time k−2; This represents the discretized input matrix, derived from the continuous-time input matrix. Time scale parameters ,parameter and The decision is made jointly, reflecting the impact of inputs on the state. For time steps The input vector is the external signal that drives the changes in the system state; For time steps The output vector of the system is the observable. To output the matrix, the state vector Mapping to output The output matrix is ​​the same as that in a continuous-time system. ). This represents the skip connection matrix after discretization, used to represent the direct influence of the input on the output. The time scale parameter (sampling period) is used to discretize continuous-time systems and determine the discrete time step. It is a continuous-time state matrix that describes the continuous-time dynamic changes of the system state. It is a continuous-time input matrix that reflects the influence of the input signal on the continuous-time state. It is a coefficient used to adjust the input terms during the discretization process. The matrix exponent is the core operation used to convert a continuous-time system into a discrete-time system.

[0041] During the discretization of the system function Let represent the mapping from the state space to the output space, where D represents the number of output features and N represents the dimension of the state space. To refine this, a first-order Taylor series expansion is used. Approximation: (1.4)

[0042] in, yes This is a discretized approximation. When discretizing a continuous-time system, it is usually necessary to approximate it. Here, a first-order Taylor expansion is used for approximation. Specifically, The Taylor expansion can approximate it as follows: ,Right now The impact of state changes on the output under certain conditions. This approximation can help simplify calculations and make the model easier to handle discrete time series data. In some cases, It can also be understood as the incremental change introduced into the input signal during the discretization process.

[0043] For input The processing steps within the SimVSS block can be described as follows: (1.5)

[0044] in, , , These represent the inputs at different layers of the SimVSS block. The output status.

[0045] Step 5.2, as follows Figure 1 As shown, the SimVSS module is embedded in the middle of attention-based internal scale feature interaction (AIFI) and convolutional neural network-based cross-scale feature fusion (CCFM).

[0046] As a further preferred embodiment of the SAR ship detection method based on range compression domain characteristics guided by the present invention, the shape-aware IoU regression loss in step eight is specifically as follows: Shape-aware IoU regression loss (Shape-IoU) calculates the loss by focusing on the shape and scale of the bounding box itself, thus enabling more accurate bounding box regression. Its calculation formula is shown below: (1.6)

[0047] in, Shape-aware IoU regression loss is used to evaluate the predicted bounding box. ) and true bounding box ( The smaller the value of the loss function, which measures the difference in shape and position, the higher the degree of matching between the predicted box and the ground truth box. These are weighting coefficients used to measure the horizontal weighting of the bounding box. ) and vertical ( The degree of mismatch in direction; To predict the x-coordinate of the center of the bounding box; The x-coordinate of the center of the true bounding box; To predict the center ordinate of the bounding box; The center ordinate of the true bounding box; It is a constant used to control the scaling ratio of the distance function (to prevent underflow or overflow problems in numerical calculations), and it determines the weight of distance in loss calculation; It is a scale factor. and These are the prediction box and the truth box. and These are the prediction boxes. truth boxes width, and These are the prediction boxes. truth boxes height, and These are the weighting coefficients in the horizontal and vertical directions, respectively.

[0048] Through the above steps, suspected target areas can be quickly located on a wide sea surface, effectively reducing the time and computing resources required for subsequent focusing imaging. Furthermore, this invention can learn the strong distinguishing characteristics between ship targets and sea clutter based on the amplitude characteristics and geometric features of ship echo energy in the range-direction focusing and azimuth-direction defocusing directions, thereby quickly locating suspected target areas.

[0049] This invention provides a new method and technical approach for wide-area maritime ship target detection using spaceborne SAR, and has significant practical application value.

[0050] The above description is merely an optional embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A SAR ship detection method guided by range compression domain characteristics, characterized in that, Includes the following steps: Step 1: Obtain the spaceborne SAR echo data after BAQ decompression. Step 2: Based on the above spaceborne SAR echo data, perform matched filtering in the range direction to obtain the corresponding range compression domain preliminary imaging; Step 3: In the distance compression domain image, use a sliding window of 1024×1024; Step 4: Add a strip attention mechanism module to the ResNet50 backbone network of the RT-DETR baseline model, and extract three scale features from the target in the 1024×1024 distance compression domain image, namely multi-scale feature maps sampled at 8, 16, and 32 times. ; Step 5: Add the SimVSS module to the hybrid encoder of the RT-DETR baseline model, and combine the 8, 16, and 32x multi-scale feature maps mentioned above. The data is fed into a hybrid encoder based on the SimVSS module for multi-scale feature extraction, and then the multi-scale features are fused across scales. Step 6: Input the features output by the hybrid encoder into the IoU perceptual query selection of RT-DETR, and filter out the regions with high overlap with the real target as the initial candidate boxes; Step 7: The Transformer decoder of RT-DETR performs multiple rounds of optimization on the selected candidate boxes. It analyzes the relationship between targets through a self-attention mechanism, gradually adjusts the position and size of the candidate boxes, and finally outputs accurate detection results. Step 8: In the RT-DETR model, in order to accurately locate the slender ship target in the range compression domain, a shape-aware IoU regression localization loss is added to the auxiliary prediction head structure.

2. The SAR ship detection method based on range compression domain characteristics as described in claim 1, characterized in that, The strip attention mechanism module described in step four specifically includes the following steps: Step 4.1, Define the strip attention mechanism module: Given input features First, perform three 1×1 convolution operations on it to obtain... , and These three feature maps; for feature maps Perform matrix dimension transformation and transpose operations to generate , for Then, the feature map Strip Pooling is used, which involves averaging the pixels in each row of the feature map to generate... Subsequently, the feature map and feature map The product operation is performed, and finally the position attention mapping result is obtained through the Softmax operation. Essentially, feature maps and The matrix multiplication operation is the product of the feature vector corresponding to a position and the feature vectors of all other positions. Therefore, the formula for calculating the similarity between positions is: （1.1） in, Representation of feature map The Each position element; Representation of feature map The Each position element express and The degree of correlation between them; Furthermore, the feature map Perform Strip Pooling along the horizontal direction, then transform the dimension. , and matrix After matrix multiplication, the dimension is transformed ; Final feature map Input features Adding corresponding elements together yields the final feature map. ; Step 4.2, the embedding position of the SAM module: In the BasicBlock or Bottleneck structure of ResNet, the SAM module is inserted after the last convolutional layer of the residual path, and the channel weighting operation is performed before the addition operation with the skip connection. Step 4.3, Multi-level SAM module configuration: Add a SAM module to the output of each stage of ResNet to form a global-local joint attention mechanism.

3. The SAR ship detection method based on range compression domain characteristics as described in claim 1, characterized in that, Step 5 of the SimVSS module specifically includes the following steps: Step 5.1, define the SimVSS module: its design is mainly based on SSM and a feedforward network; this structure includes residual connections and normalization layers, designed to stabilize the gradient training process and accelerate model convergence; the traditional SSM is a linear time-invariant system function, which uses intermediate variables Univariate sequences Mapping to output sequence Specifically, given a state transition matrix As evolutionary factors, the weight matrix and the observation matrix These are respectively used as projection factors and defined skip connections. Its mathematical expression is as follows: （1.2） In addition, the system function introduces a time-scale parameter. To discretize the data for processing discrete time series data; this transformation is defined as follows: （1.3） in, This represents the mapping from the state space to the output space; it is used for refinement using a first-order Taylor series expansion. Approximation: （1.4） For input The processing steps within the SimVSS block are described as follows: （1.5） in, , , These represent the inputs at different layers of the SimVSS block. The output state; The discretized state transition matrix is ​​represented by the continuous-time state matrix. Through matrix exponent The result is obtained through conversion and is used to describe the evolution of states in a discrete-time system. This represents the discretized input matrix, derived from the continuous-time input matrix. Time scale parameters ,parameter and The decision is made jointly, reflecting the impact of inputs on the state; Step 5.2, the SimVSS module is embedded in the middle position between attention-based internal scale feature interaction and convolutional neural network-based cross-scale feature fusion.

4. The SAR ship detection method based on range compression domain characteristics as described in claim 1, characterized in that, The shape-aware IoU regression loss described in step eight is specifically as follows: Shape-aware IoU regression loss calculates the loss by focusing on the shape and scale of the bounding box itself, thus enabling more accurate bounding box regression. The calculation formula is shown below: （1.6） in, It is a scale factor. and These are the prediction box and the truth box. and These are the prediction boxes. truth boxes width, and These are the prediction boxes. truth boxes height, and These are the weighting coefficients in the horizontal and vertical directions, respectively; For shape-aware IoU regression loss, it is a loss function used to evaluate the difference in shape and position between the predicted bounding box and the ground truth bounding box. The smaller the value, the higher the degree of matching between the predicted box and the ground truth box. These are weighting coefficients used to measure the degree of mismatch between the bounding boxes in the horizontal and vertical directions; To predict the x-coordinate of the center of the bounding box; The x-coordinate of the center of the true bounding box; To predict the center ordinate of the bounding box; The center ordinate of the true bounding box; It is a constant used to control the scaling ratio of the distance function.

Images