A small sample multi-scale spatial fragment target detection method based on illumination transfer

Abstract

The embodiment of the application provides a small sample multi-scale space debris target detection method based on illumination migration, which comprises the following steps: collecting an image to be processed; identifying space debris in the image to be processed based on a preset space debris detection model; the space debris detection model is used for multiple feature extraction, in each feature extraction, multi-scale features are extracted first, and then based on global average pooling and attention mechanism, the correlation between the features and the background of the image to be processed is strengthened to obtain first features; based on a gating residual mechanism, the illumination component in the first features is identified; according to the illumination component, the detection threshold of the gating unit is adjusted; according to the detection threshold, second features are obtained; based on the first features and the second features, feature fusion is performed to identify the space debris. The technical scheme provided by the application is used to solve the problems of high detection difficulty, low detection precision and large calculation amount in the prior art.

Description

Technical Field

[0001] This document relates to the field of space debris management technology, and in particular to a small-sample, multi-scale space debris target detection method based on illumination migration. Background Technology

[0002] With the rapid development of space technology, the number of satellites, probes, and other spacecraft launched by humans is constantly increasing, but this has also led to an increasingly serious problem of space debris.

[0003] In recent years, researchers have begun to explore space debris detection methods based on deep learning and computer vision technologies. Deep learning models, especially convolutional neural networks (CNNs), have performed exceptionally well in tasks such as image classification and object detection.

[0004] However, space debris varies greatly in size and shape, ranging from a few centimeters to several meters, and its attitude is constantly changing, which increases the difficulty of detection. In space, lighting conditions are highly variable, from direct sunlight to the weak light reflected from Earth, reducing the detection accuracy of the model. Acquiring high-precision images and data of space debris is difficult, and the scarcity of data limits the training effect of deep learning models. At the same time, in order to improve detection accuracy, a large amount of additional computation is required, thereby increasing the power consumption and complexity of on-orbit equipment. Summary of the Invention

[0005] In view of the above-mentioned solutions, this application aims to propose a small-sample, multi-scale spatial debris target detection method based on illumination migration to solve at least one of the above problems.

[0006] In a first aspect, one or more embodiments of this application provide a small-sample, multi-scale spatial debris target detection method based on illumination migration, including:

[0007] Acquire the image to be processed;

[0008] The space debris in the image to be processed is identified based on a preset space debris detection model;

[0009] The space debris detection model is used to perform multiple feature extractions. In each feature extraction, multi-scale features are first extracted, and then the correlation between the features and the background of the image to be processed is strengthened based on global average pooling and attention mechanisms to obtain a first feature. Based on a gated residual mechanism, the illumination component in the first feature is identified. According to the illumination component, the detection threshold of the gated unit is adjusted. According to the detection threshold, a second feature is obtained. Based on the first feature and the second feature, feature fusion is performed to identify the space debris.

[0010] Furthermore, the space debris detection model includes: a small-sample, multi-scale space debris target detection sub-model based on illumination migration;

[0011] The small-sample multi-scale spatial debris target detection sub-model based on illumination migration includes, in order of information flow direction, the following modules: first module, second module, third module, and fourth module.

[0012] The first module is used to extract a third feature from the acquired image to be processed, the third feature corresponding to a shallow feature;

[0013] The second module is used to extract the first feature from the third feature, wherein the first feature corresponds to the middle-layer feature;

[0014] The third module is used to adjust the channel weights corresponding to the first feature;

[0015] The fourth module is used to adjust the intensity of the first feature based on the channel weight and the gated residual mechanism to obtain the second feature.

[0016] Furthermore, the first module, the second module, the third module, and the fourth module each contain multiple sub-modules, and the ratio of the number of sub-modules in the first module, the second module, the third module, and the fourth module is 2:2:4:2.

[0017] Furthermore, the sub-modules of the first module, in accordance with the information flow direction, sequentially include: an image block embedding layer and a spatial module;

[0018] The second module's sub-modules, in order of information flow direction, include: an image block embedding layer and a spatial module;

[0019] The sub-modules of the third module, in order of information flow direction, include: an image block embedding layer, a spatial module, and a spatial module containing a compression-excitation sub-module;

[0020] The sub-modules of the fourth module, in order of information flow direction, include: an image block embedding layer, a spatial module, and a gate control module;

[0021] The image block embedding layer is used to segment the image into small blocks and vectorize them;

[0022] The space module is used to extract features of space debris and the relative relationship between the space debris and other subjects in the image;

[0023] The spatial module containing the compression-excitation submodule is used to adaptively adjust the channel weights of the features;

[0024] The gate control module is used to dynamically adjust the feature intensity.

[0025] Furthermore, the space module includes:

[0026] Depth-separable convolutional layers are used to extract features from the space debris;

[0027] A batch normalization layer is used to normalize the features of the space debris;

[0028] A global average pooling layer is used to reduce the dimensionality of the normalized features of space debris;

[0029] A deformable self-attention mechanism layer is used to enhance the features of the space debris in relation to other subjects in the image;

[0030] A feedforward network layer is used to receive the output of the deformable self-attention mechanism layer and perform a nonlinear transformation on the output of the deformable self-attention mechanism layer.

[0031] Furthermore, the spatial module containing the compression-excitation submodule includes:

[0032] Depth-separable convolutional layers are used to extract features from the space debris;

[0033] A batch normalization layer is used to normalize the features of the space debris;

[0034] A global average pooling layer is used to reduce the dimensionality of the normalized features of space debris;

[0035] A deformable self-attention mechanism layer is used to enhance the features of the space debris in relation to other subjects in the image;

[0036] A feedforward network layer is used to receive the output of the deformable self-attention mechanism layer and perform a nonlinear transformation on the output of the deformable self-attention mechanism layer.

[0037] The compression-excitation submodule is used to adjust the weights of the features of the space debris through global pooling.

[0038] Furthermore, the door control module includes: a reset door and an attention door;

[0039] The reset gate is used to generate a weight matrix for the first feature based on the gated residual;

[0040] The attention gate is used to dynamically adjust the first feature to obtain the second feature based on the weight matrix of the reset gate and the weight matrix of the attention gate; wherein the output of the gate control module is the sum of the first feature and the second feature.

[0041] Furthermore, the space debris detection model also includes: a fast pyramid pooling module and a feature fusion module;

[0042] The fast pyramid pooling module is used to extract the fourth feature from the second feature;

[0043] The feature fusion module is used to perform feature fusion on the third feature, the first feature, and the fourth feature.

[0044] Secondly, one or more embodiments of this application provide a small-sample, multi-scale spatial debris target detection device based on illumination migration, comprising:

[0045] The acquisition module is used to acquire images to be processed.

[0046] The data processing module is used to identify space debris in the image to be processed based on a preset space debris detection model;

[0047] The space debris detection model is used to perform multiple feature extractions. In each feature extraction, multi-scale features are first extracted, and then the correlation between the features and the background of the image to be processed is strengthened based on global average pooling and attention mechanisms to obtain a first feature. Based on a gated residual mechanism, the illumination component in the first feature is identified. According to the illumination component, the detection threshold of the gated unit is adjusted. According to the detection threshold, a second feature is obtained. Based on the first feature and the second feature, feature fusion is performed to identify the space debris.

[0048] Thirdly, embodiments of this application provide a storage medium for storing computer-executable instructions, characterized in that, when executed, the computer-executable instructions implement the steps of the small-sample multi-scale spatial debris target detection method based on illumination migration as described in the first aspect.

[0049] Compared with the prior art, this application can achieve at least the following technical effects:

[0050] This invention is based on a lightweight approach to first extract multi-scale features, and then strengthens the correlation between the features and the background of the image to be processed based on global average pooling and attention mechanisms. This enables the capture of global and local features of space debris and accurate detection of tiny debris. At the same time, a gated residual mechanism is introduced to identify the illumination component from the features and adjust the detection threshold of the gated unit according to the illumination component. This allows for adaptive adjustment of the detection threshold based on illumination conditions, thereby adjusting the intensity of the first feature and enhancing the robustness and adaptability of the model under complex illumination conditions. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in one or more embodiments of this application 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 recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A flowchart illustrating a small-sample, multi-scale spatial debris target detection method based on illumination migration, provided for one or more embodiments of this application;

[0053] Figure 2 A schematic diagram of the structure of a space debris detection model provided in one or more embodiments of this application;

[0054] Figure 3 A schematic diagram of the structure of a small-sample multi-scale spatial debris target detection sub-model based on illumination transfer, provided for one or more embodiments of this application;

[0055] Figure 4 A schematic diagram of the structure of a space module provided in one or more embodiments of this application;

[0056] Figure 5 A schematic diagram of the structure of a space module containing a compression-excitation submodule is provided for one or more embodiments of this application;

[0057] Figure 6 A schematic diagram of the structure of a door control module provided in one or more embodiments of this application;

[0058] Figure 7 A schematic diagram of transfer training provided for one or more embodiments of this application;

[0059] Figure 8 This is a schematic diagram of the structure of a small-sample, multi-scale spatial debris target detection device based on illumination migration, provided for one or more embodiments of this application. Detailed Implementation

[0060] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this application, the technical solutions in one or more embodiments of this application 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 this application, and not all of the embodiments. Based on one or more embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the protection scope of this document.

[0061] This application provides a small-sample, multi-scale spatial debris target detection method based on illumination migration, such as... Figure 1 As shown, it includes the following steps:

[0062] Step 1: Acquire the image to be processed.

[0063] In this embodiment, space debris, including defunct satellites, rocket remnants, and malfunctioning spacecraft components, travels at high speeds, posing a serious threat to spacecraft and astronauts in orbit and potentially triggering a chain reaction. Ground-based radar and optical telescopes can track and measure the trajectory and reflectivity of debris, thereby detecting and predicting its location.

[0064] Step 2: Identify space debris in the image to be processed based on a preset space debris detection model.

[0065] In this embodiment, the space debris detection model performs multiple feature extractions. Each feature extraction first extracts multi-scale features, then strengthens the association between the features and the background of the image to be processed based on global average pooling and attention mechanisms to obtain a first feature. Based on a gating residual mechanism, the illumination component in the first feature is identified. The detection threshold of the gating unit is adjusted according to the illumination component. A second feature is obtained based on the detection threshold. Feature fusion is performed based on the first and second features to identify the space debris.

[0066] Specifically, the core idea of ​​the space debris detection model is to improve detection accuracy through the synergy of local and global features. Regarding local features, multi-scale features are extracted multiple times to capture subtle local characteristics such as surface texture and edge shape of space debris, including details like surface bumps and cracks. At the global feature level, after reducing feature dimensionality using global average pooling, a deformable self-attention mechanism is employed to consider global information such as the debris's position in the entire space environment and its relative relationship with other celestial bodies or space objects. When these two mechanisms work together, the deformable self-attention mechanism adaptively focuses, enabling the model to comprehensively consider the unique local details of the debris and its global space context. For example, by combining the debris's irregular shape with its position in its orbit, the model can more accurately determine whether it is threatening space debris, thereby improving detection accuracy.

[0067] To further enhance subtle local features such as surface texture and edge shape of space debris, a gated residual mechanism is introduced. This mechanism strengthens the signal intensity of strong shape features, allowing even relatively weaker but important features (such as the unique irregular shape of small debris) to be fully captured. This enables the model to consider different features of space debris and avoid missing crucial information. Simultaneously, the gated residual mechanism dynamically adjusts the threshold based on current global feature information, weighting channels containing debris features to enhance their responses and weaken the influence of background interference, thereby strengthening global features. Through these methods, the synergistic effect of local and global features can be further leveraged.

[0068] Finally, feature fusion is achieved based on the synergy between local and global features.

[0069] It should be noted that both the first and second features contain local and global features. The relationship between the first and second features is that the second feature is obtained by adjusting the local and global features in the first feature using a gating residual mechanism.

[0070] This application provides a space debris detection model, such as Figure 2 As shown, it includes: a small-sample multi-scale spatial debris target detection sub-model 1 based on illumination transfer, a fast pyramid pooling module 2, and a feature fusion module 3;

[0071] The small-sample multi-scale spatial debris target detection sub-model 1 based on illumination transfer is used to extract the first feature, the second feature, and the third feature.

[0072] Fast pyramid pooling module 2 is used to extract the fourth feature from the second feature;

[0073] Feature fusion module 3 is used to perform feature fusion on the third feature, the first feature and the fourth feature.

[0074] Specifically, the small-sample multi-scale spatial debris target detection sub-model 1 based on illumination transfer, such as... Figure 3 As shown, along the direction of information flow, it includes, in sequence: first module 11, second module 12, third module 13 and fourth module 14;

[0075] The first module 11 is used to extract the third feature from the acquired image to be processed, the third feature corresponding to the shallow feature;

[0076] The second module 12 is used to extract the first feature from the third feature, where the first feature corresponds to the middle layer feature;

[0077] The third module 13 is used to adjust the channel weights corresponding to the first feature;

[0078] The fourth module 14 is used to adjust the intensity of the first feature based on the channel weight and the gated residual mechanism to obtain the second feature.

[0079] Module 1, Module 2, Module 3, and Module 4 each contain multiple sub-modules. The ratio of the number of sub-modules in Module 1, Module 2, Module 3, and Module 4 is as follows:

[0080] N1:N2:N3:N4 = 2:2:4:2.

[0081] In this embodiment of the application, the sub-modules of the first module, in accordance with the information flow direction, include: an image patch embedding layer and a spatial module (s-Block);

[0082] The second module's sub-modules, in order of information flow direction, include: an image block embedding layer and a spatial module;

[0083] The third module's sub-modules, in order of information flow direction, include: an image block embedding layer, a spatial module, and a spatial module (s-se-Block) containing a compression-excitation sub-module;

[0084] The fourth module's sub-modules, in order of information flow direction, include: image block embedding layer, spatial module, and gate-block module;

[0085] The image patch embedding layer is used to segment the image into small patches and vectorize them;

[0086] The spatial module is used to extract features of space debris and its relative relationship to other subjects in the image;

[0087] A spatial module containing a compression-excitation submodule is used to adaptively adjust the channel weights of features;

[0088] The gate control module is used to dynamically adjust the feature intensity.

[0089] Among them, spatial modules, such as Figure 4 As shown, it includes:

[0090] Depth-separable convolutional layers (3x3 DW and 1x1 DW) are used to extract features from the space debris;

[0091] A batch normalization layer (BN) is used to normalize the features of the space debris;

[0092] A global average pooling layer (Avgpool) is used to reduce the dimensionality of the normalized features of space debris;

[0093] A deformable attention mechanism layer is used to enhance the features of the space debris in relation to other subjects in the image;

[0094] The feedforward network layer (FFN) is used to receive the output of the deformable self-attention mechanism layer and perform a nonlinear transformation on the output of the deformable self-attention mechanism layer.

[0095] Specifically, the s-Block module incorporates depthwise separable convolution (DW), which has fewer parameters and lower computational cost compared to ordinary convolution, thus facilitating network deepening. The s-Block module first extracts local features at different scales through 3x3 DW and 1x1 DW convolution operations. Let the input feature be X1, then after passing through the convolutional layers:

[0096] X I =DW 3×3 (X1)+DW 1×1 (X1)

[0097] Among them, DW 3×3 This represents a depthwise separable convolution result with a 3x3 kernel. (DW) 1×1 This represents a depthwise separable convolution result with a 1x1 kernel.

[0098] Subsequently, the features are processed by Batch Normalization (BN) and the feature dimensionality is reduced by global average pooling (AvgPool).

[0099] X BN =BN(X) I )

[0100] X d =DeformableAttention(X) BN ·W Q ,P s (X BN ·W K ),P s (X BN ·W V ))

[0101] Where BN(·) represents the batch normalization function, DeformableAttention represents the operator corresponding to the deformable self-attention mechanism, and X BN W represents the features after batch normalization. Q W K W VX represents the weight matrix in the deformable attention mechanism, corresponding to the projection matrices of the query, key, and value, respectively. BN ·W Q : Maps input features to the query space for calculating attention weights. X BN ·W K : Mapped to the key space and matched with the query to determine feature relevance. X BN ·W V : Mapped to the value space, providing feature information for aggregation, X d This represents the output features after processing by the deformable self-attention mechanism. P_s(·) means: positional encoding function. Its function is to add spatial location information to the key and value features, enhancing the model's ability to perceive the spatial location of debris targets, especially suitable for multi-scale debris detection.

[0102] To enhance the synergy between local and global features, a Deformable Attention mechanism is introduced to adaptively focus on space debris features. Finally, the output features are generated through a feedforward network (FFN).

[0103] X s-Block =FFN(X2+X) d )

[0104] Here, FFN(·) represents the corresponding operator of the feedforward neural network. X2 is the original feature input to the s-Block module. X2+X d This represents a residual connection, which connects the original feature X2 with the attention-enhanced feature X. d Add elements one by one.

[0105] This design enables the model to adaptively adjust attention weights based on different input features, better capturing the global and local features of small space debris.

[0106] The spatial module containing the compression-excitation submodule, such as Figure 5 As shown, it includes:

[0107] Depth-separable convolutional layers (3x3 DW and 1x1 DW) are used to extract features from the space debris;

[0108] A batch normalization layer (BN) is used to normalize the features of the space debris;

[0109] A global average pooling layer (Avgpool) is used to reduce the dimensionality of the normalized features of space debris;

[0110] A deformable attention mechanism layer is used to enhance the features of the space debris in relation to other subjects in the image;

[0111] The feedforward network layer (FFN) is used to receive the output of the deformable self-attention mechanism layer and perform a nonlinear transformation on the output of the deformable self-attention mechanism layer.

[0112] The compression-excitation submodule (SE) is used to adjust the weights of the features of the space debris through global pooling.

[0113] Specifically, s-se-Block is an enhanced version of s-Block, incorporating a Squeeze-and-Excitation (SE) block into its structure. The SE module uses global pooling to weight and adjust features, improving the model's adaptability in space debris detection, especially in low-light environments. The key formula is:

[0114]

[0115] Here, Z represents the result of global average pooling of feature map X. H and W represent the height and width of the feature map, respectively. X(i,j) represents the element value of feature map X at spatial coordinates (i,j), indicating the feature intensity at that location (such as the local features of space debris). i corresponds to the height coordinate, and j corresponds to the width coordinate.

[0116] SE module features are generated by adjusting the activation function and weights:

[0117] X se =σ(W2·ReLU(W1·Z))·

[0118] X se This represents the output characteristics of the SE module (Squeeze-and-Excitation).

[0119] W1 and W2 represent the weight matrices in the SE module, used to perform linear transformations on the globally pooled Z (dimensionality reduction → activation → dimensionality increase), and to facilitate the model learning the attention weights between channels (such as which channels are more critical for fragment detection).

[0120] σ and ReLU are table activation functions, namely sigmoid (channel weight normalization) and ReLU (introducing nonlinearity), which enhance the model's ability to model nonlinearities of channels and features.

[0121] X represents the original feature map input to the SE module.

[0122] Finally, the output characteristics of the s-se-Block module are:

[0123] X s-se-Block =FFN(X se +X d )

[0124] That is, the s-se-Block module integrates channel attention and spatial attention, and then enhances features through FFN.

[0125] Door control module, such as Figure 6 As shown, this includes: the reset gate (r-Gate) and the attention gate (a-Gate);

[0126] The reset gate is used to generate the weight matrix of the first feature based on the gated residual;

[0127] The attention gate is used to dynamically adjust the first feature to obtain the second feature based on the weight matrix of the reset gate and the weight matrix of the attention gate; wherein, the output of the gate control module is the sum of the first feature and the second feature.

[0128] Specifically, the gate control module performs threshold adjustment through two sub-gating units, r-Gate and a-Gate. r-Gate generates a residual signal based on the input features and performs global adjustment through a linear layer to ensure that different features of space debris are fully captured. The signal features are represented as follows:

[0129] X r-Gate =σ(σ(FC(X3)+X3))

[0130] Where X3 is the input feature of the gating module, FC(·) is the operator corresponding to the fully connected layer, and σ is the activation function.

[0131] a-Gate uses an adaptive gating mechanism to dynamically adjust the threshold based on global feature information, ensuring the model's flexibility in low-light environments. a-Gate uses feature weighting:

[0132] X a-Gate =AvgPool(W r ·tanh(σ(FFN(X r-Gate )+W a ·X r-Gate )))

[0133] The final output feature can be represented as:

[0134] X Gate-Block =X r-Gate +X a-Gate

[0135] Among them, W aFor the weight matrix, perform a linear transformation on sigma and tanh, W r Perform a linear transformation on X^{r-Gate} using the weight matrix. W a and W r It helps the model learn the dependencies between channels or feature dimensions, and adapts the gating mechanism to the weighting requirements of multi-scale features of space debris (such as enhancing the weak features of small debris).

[0136] Tanh is a hyperbolic tangent activation function that introduces nonlinearity, thereby enhancing the nonlinear expressive power of features.

[0137] AvgPool(·) is the operator corresponding to Average Pooling, which aggregates spatial dimensions (height H, width W). By compressing spatial information, AvgPool(·) generates channel-level global features (such as the average response of each channel), which is suitable for capturing global context (such as the overall distribution of background and debris) in space debris detection.

[0138] In this embodiment, the gate control module can identify the illumination components (such as brightness and contrast) in the input features and determine the current illumination conditions based on these components, such as a low-light scene or a bright-light scene. Then, based on the current illumination conditions, it dynamically generates a detection threshold for the gate control unit that adapts to the current illumination conditions. For example, in a low-light scene, the gate control module automatically lowers the detection threshold to avoid missed detections; in a bright-light scene, it raises the threshold to reduce false detections, thereby achieving adaptive adjustment to changes in illumination.

[0139] In this embodiment, a multi-layer Backbone network (11 to 14) extracts features layer by layer, and then further compresses and enhances the feature representation through a Spatial Pyramid Pooling-Fast (SPPF) module 2. The SPPF module is a lightweight feature extraction module designed to compress high-dimensional features while preserving important spatial information. It maps features to multiple resolutions by performing pooling operations at different scales, and then concatenates these features to generate a fused feature map, which is the fourth feature.

[0140] In this embodiment, the features in each stage of the feature fusion module 3 are refined by the C3 module and the CBG module. The core idea of ​​the C3 module is to divide the features into blocks, with one part passing directly through the network and the other part undergoing convolution operations, and then concatenating the two parts. The CBG module consists of "convolution (Conv) + batch normalization + GeLU activation function". Upsampling is used to map low-resolution features to a high-resolution space and fuse them with shallow features to achieve cross-scale feature combination. The fused features are then convolved to generate the final detection result.

[0141] In this embodiment of the application, a transfer learning model is applied when constructing the model, and the specific process is as follows: Figure 7 As shown: First, the object detection model is pre-trained on a large dataset. Then, the pre-trained backbone network is transferred to the space debris detection task through fine-tuning, which is particularly suitable for small sample datasets of space debris. During the fine-tuning stage, by adaptively adjusting parameters, the model can maintain efficient detection capabilities in complex environments with varying lighting conditions. That is, transfer learning is used to transfer the feature extraction network trained under normal lighting conditions to low lighting conditions to ensure the model's recognition accuracy. In addition, transfer learning can also reduce the parameters involved in model computation, thereby reducing the computational load. Here, the Backbone corresponds to the backbone network, which is used to extract features from the input data. The Head corresponds to the head network, which processes the features extracted by the Backbone to produce the final output. The Detect module corresponds to the detection module, which identifies and locates space debris in the image.

[0142] This application provides a small-sample, multi-scale spatial debris target detection device based on illumination migration, such as... Figure 8 As shown, it includes:

[0143] Acquisition module 701 is used to acquire images to be processed;

[0144] The data processing module 702 is used to identify space debris in the image to be processed based on a preset space debris detection model;

[0145] The space debris detection model is used to perform multiple feature extractions. In each feature extraction, multi-scale features are first extracted, and then the correlation between the features and the background of the image to be processed is strengthened based on global average pooling and attention mechanisms to obtain the first feature. Based on the gated residual mechanism, the intensity of the first feature is adjusted to obtain the second feature. Based on the first feature and the second feature, feature fusion is performed to identify the space debris.

[0146] This application provides a storage medium for storing computer-executable instructions, characterized in that, when executed, the computer-executable instructions implement the steps of the small-sample multi-scale spatial debris target detection method based on illumination migration as described in any one of the embodiments.

[0147] It should be noted that the embodiments concerning storage media in this application and the embodiments concerning blockchain-based service provision methods in this application are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the corresponding blockchain-based service provision methods described above, and the repeated parts will not be described again.

[0148] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0149] In the 1930s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many improvements to the methodology today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that an improvement to the methodology cannot be implemented using a hardware physical module. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program a digital system themselves to "integrate" it onto a PLD, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0150] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0151] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0152] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing the embodiments of this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0153] Those skilled in the art will understand that one or more embodiments of this application can be provided as a method, system, or computer program product. Therefore, one or more embodiments of this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0154] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0155] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0156] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0157] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0158] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0159] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0160] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0161] One or more embodiments of this application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. One or more embodiments of this application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can reside in local and remote computer storage media, including storage devices.

[0162] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0163] The above description is merely an embodiment of this document and is not intended to limit the scope of this document. Various modifications and variations can be made to this document by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this document should be included within the scope of the claims of this document.

Claims

1. A small-sample, multi-scale spatial debris target detection method based on illumination migration, characterized in that, Includes the following steps: Acquire the image to be processed; The space debris in the image to be processed is identified based on a preset space debris detection model; The space debris detection model is used to perform multiple feature extractions. In each feature extraction, multi-scale features are extracted first, and then the correlation between the features and the background of the image to be processed is strengthened based on global average pooling and attention mechanism to obtain the first feature. Based on the gated residual mechanism, the illumination component in the first feature is identified; The detection threshold of the gating unit is adjusted according to the illumination component; a second feature is obtained according to the detection threshold; feature fusion is performed based on the first feature and the second feature to identify the space debris.

2. The method according to claim 1, characterized in that, The space debris detection model includes: a small-sample, multi-scale space debris target detection sub-model based on illumination migration; The small-sample multi-scale spatial debris target detection sub-model based on illumination migration includes, in order of information flow direction, the following modules: first module, second module, third module, and fourth module. The first module is used to extract a third feature from the acquired image to be processed, the third feature corresponding to a shallow feature; The second module is used to extract the first feature from the third feature, wherein the first feature corresponds to the middle-layer feature; The third module is used to adjust the channel weights corresponding to the first feature; The fourth module is used to adjust the intensity of the first feature based on the channel weight and the gated residual mechanism to obtain the second feature.

3. The method according to claim 2, characterized in that, The first module, the second module, the third module, and the fourth module each contain multiple sub-modules, and the ratio of the number of sub-modules in the first module, the second module, the third module, and the fourth module is 2:2:4:

2.

4. The method according to claim 3, characterized in that, The sub-modules of the first module, in order of information flow direction, include: an image block embedding layer and a spatial module; The second module's sub-modules, in order of information flow direction, include: an image block embedding layer and a spatial module; The sub-modules of the third module, in order of information flow direction, include: an image block embedding layer, a spatial module, and a spatial module containing a compression-excitation sub-module; The sub-modules of the fourth module, in order of information flow direction, include: an image block embedding layer, a spatial module, and a gate control module; The image block embedding layer is used to segment the image into small blocks and vectorize them; The space module is used to extract features of space debris and the relative relationship between the space debris and other subjects in the image; The spatial module containing the compression-excitation submodule is used to adaptively adjust the channel weights of the features; The gate control module is used to dynamically adjust the feature intensity.

5. The method according to claim 4, characterized in that, The space module includes: Depth-separable convolutional layers are used to extract features from the space debris; A batch normalization layer is used to normalize the features of the space debris; A global average pooling layer is used to reduce the dimensionality of the normalized features of space debris; A deformable self-attention mechanism layer is used to enhance the features of the space debris in relation to other subjects in the image; A feedforward network layer is used to receive the output of the deformable self-attention mechanism layer and perform a nonlinear transformation on the output of the deformable self-attention mechanism layer.

6. The method according to claim 4, characterized in that, The space module containing the compression-excitation submodule includes: Depth-separable convolutional layers are used to extract features from the space debris; A batch normalization layer is used to normalize the features of the space debris; A global average pooling layer is used to reduce the dimensionality of the normalized features of space debris; A deformable self-attention mechanism layer is used to enhance the features of the space debris in relation to other subjects in the image; A feedforward network layer is used to receive the output of the deformable self-attention mechanism layer and perform a nonlinear transformation on the output of the deformable self-attention mechanism layer. The compression-excitation submodule is used to adjust the weights of the features of the space debris through global pooling.

7. The method according to claim 4, characterized in that, The door control module includes: a reset door and an attention door; The reset gate is used to generate a weight matrix for the first feature based on the gated residual; The attention gate is used to dynamically adjust the first feature to obtain the second feature based on the weight matrix of the reset gate and the weight matrix of the attention gate; wherein the output of the gate control module is the sum of the first feature and the second feature.

8. The method according to claim 2, characterized in that, The space debris detection model also includes: a fast pyramid pooling module and a feature fusion module; The fast pyramid pooling module is used to extract the fourth feature from the second feature; The feature fusion module is used to perform feature fusion on the third feature, the first feature, and the fourth feature.

9. A small-sample, multi-scale spatial debris target detection device based on illumination migration, characterized in that, include: The acquisition module is used to acquire images to be processed. The data processing module is used to identify space debris in the image to be processed based on a preset space debris detection model; The space debris detection model is used to perform multiple feature extractions. In each feature extraction, multi-scale features are extracted first, and then the correlation between the features and the background of the image to be processed is strengthened based on global average pooling and attention mechanism to obtain the first feature. Based on the gated residual mechanism, the illumination component in the first feature is identified; The detection threshold of the gating unit is adjusted according to the illumination component; a second feature is obtained according to the detection threshold; feature fusion is performed based on the first feature and the second feature to identify the space debris.

10. A storage medium for storing computer-executable instructions, characterized in that, When executed, the computer-executable instructions implement the steps of the small-sample multi-scale spatial debris target detection method based on illumination migration as described in any one of claims 1-7.

Images