Influence factor grading quantification method for complex scene remote sensing target detection
By constructing a hierarchical quantification method for influencing factors of remote sensing target detection, and utilizing uniform sampling and performance function back-inference, the accuracy and robustness issues of remote sensing target detection in complex scenarios are solved, and efficient engineering implementation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA ACADEMY OF SPACE TECHNOLOGY
- Filing Date
- 2025-12-26
- Publication Date
- 2026-04-17
AI Technical Summary
Existing remote sensing target detection technologies lack scientific methods for classifying and quantifying influencing factors in complex scenarios, resulting in unstable detection accuracy, poor robustness, high costs of ineffective debugging, and difficulties in engineering implementation.
By uniformly sampling to obtain the parameter values of influencing factors, constructing the functional relationship between performance indicators and factor parameter values, and inversely deducing the performance range to achieve factor classification, a causal forward deduction logic is established to scientifically quantify the influence of parameters on performance.
It achieves accurate identification and improved robustness of remote sensing target detection, reduces debugging costs, and improves the feasibility and efficiency of engineering implementation.
Smart Images
Figure CN121884046A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to data processing methods, and more particularly to a hierarchical quantification method for influencing factors in remote sensing target detection in complex scenarios. Background Technology
[0002] Remote sensing target detection technology serves as a core support for fields such as surveying and mapping, emergency rescue, and national defense reconnaissance. It must achieve accurate target identification in complex scenarios, including those obscured by clouds and fog, undulating terrain, varying target scales, and interference from ground features. In the entire process of "data acquisition—preprocessing—model training—detection output," several sub-factors (such as payload resolution and target camouflage level) exert significant influence. Even minor changes in these factors can lead to substantial fluctuations in detection model performance (such as accuracy and recall). Therefore, scientifically classifying and precisely quantifying individual sub-factors to clarify their impact on model performance is a crucial prerequisite for optimizing the detection process and improving adaptability to complex scenarios.
[0003] However, existing hierarchical quantization methods mostly employ empirical, uniform sampling across parameter intervals. This approach treats all parameter intervals equally, failing to focus on critical regions sensitive to model performance, resulting in significant discrepancies between quantization results and actual impacts. These discrepancies further hinder the precise basis for optimizing the detection process, easily misleading parameter adjustments. This not only leads to unstable detection accuracy and poor robustness but also significantly increases the cost of ineffective debugging and prolongs project cycles, severely restricting the engineering implementation and application of remote sensing target detection technology in complex scenarios.
[0004] The aforementioned problems are caused by the deficiencies of existing methods:
[0005] 1. Single uniform sampling only pursues equal-interval coverage of parameter ranges, without establishing a clear causal relationship between sampling points and performance indicators. The obtained data can only reflect a rough trend and is difficult to accurately characterize the quantitative correspondence between parameters and performance. Therefore, it cannot provide a mathematical basis for inference.
[0006] 2. Existing classifications often start from parameter sensitivity ranges or empirical thresholds, rather than being guided by model performance requirements. This creates a causal inversion between "performance requirements" and "parameter classification," resulting in classification results that are disconnected from actual detection needs and are highly subjective. Summary of the Invention
[0007] To address the technical problems existing in the prior art, the present invention aims to provide a hierarchical quantification method for influencing factors in remote sensing target detection in complex scenarios. This method can achieve scientific hierarchical and accurate quantification of influencing factors, providing a basis for the entire chain of optimization, such as data preprocessing and model tuning, thereby improving the accuracy and robustness of remote sensing target detection and recognition.
[0008] To achieve the above-mentioned objectives, this invention provides a hierarchical quantification method for influencing factors in remote sensing target detection in complex scenes, the method comprising:
[0009] At least one influencing factor in the entire remote sensing target detection chain is identified; the entire remote sensing target detection chain includes data acquisition, data preprocessing, remote sensing target detection model training, and target detection output.
[0010] Uniform sampling is performed within the range of values of the influencing factors to obtain multiple parameter values of the influencing factors;
[0011] For each influencing factor parameter value, a perturbation or transformation corresponding to that influencing factor parameter value is applied to the standard test dataset to generate a corresponding test sample set; then the test sample set is input into a pre-configured remote sensing target detection model for inference to obtain the corresponding performance index.
[0012] Based on the parameter values of the multiple influencing factors and the corresponding performance indicators, a functional relationship between the performance indicators and the parameter values of the influencing factors is obtained by fitting the performance indicators.
[0013] The range of values for the performance index is evenly divided into N performance intervals; N is a positive integer greater than 1.
[0014] Based on the aforementioned functional relationship, the corresponding range of influencing factor parameter values is obtained by reverse deduction from each performance range, thereby dividing the influencing factors into N levels.
[0015] According to one technical solution of the present invention, the influencing factors include platform-related factors, load-related factors, environmental factors, target-related factors, and target camouflage-related factors;
[0016] The platform-related factors include platform type, platform height, etc.
[0017] The load-related factors include load type and load resolution;
[0018] The environmental factors mentioned include atmospheric attenuation and light intensity;
[0019] The target factors include target scale and target operating conditions;
[0020] The target camouflage factors include optical camouflage and infrared camouflage.
[0021] According to one technical solution of the present invention, the performance indicators include at least one of accuracy, recall, average precision and F1 score.
[0022] According to one technical solution of the present invention, N is 4;
[0023] The performance index is evenly divided into four performance intervals, which correspond to the first level, the second level, the third level and the fourth level, respectively; and the performance interval corresponding to the first level has the highest value and the performance interval corresponding to the fourth level has the lowest value.
[0024] According to one technical solution of the present invention, for multiple influencing factor parameter values of any influencing factor, the corresponding performance index is obtained by reasoning through the same pre-configured remote sensing target detection model.
[0025] According to one technical solution of the present invention, when the influencing factor is a non-numerical factor, multiple types are selected from the optional types of the non-numerical factor, and for each type, the corresponding performance index is obtained by reasoning through a pre-configured remote sensing target detection model; then, each type is classified into corresponding levels according to the level of the performance index.
[0026] According to one technical solution of the present invention, it further includes:
[0027] Output a quantitative report; the quantitative report includes influencing factors, corresponding levels, corresponding influencing factor parameter value ranges, and corresponding level labels; the level labels include at least one of critical level, important level, minor level, and level to be optimized.
[0028] The critical level corresponds to the highest performance range, the important level corresponds to the second-highest performance range, the minor level corresponds to the medium performance range, and the level to be optimized corresponds to the low performance range.
[0029] The present invention also provides an electronic device, comprising: one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to enable the electronic device to perform the above-described method for hierarchical quantification of influencing factors for remote sensing target detection in complex scenes.
[0030] The present invention also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the above-described method for hierarchical quantification of influencing factors for remote sensing target detection in complex scenarios.
[0031] This invention provides a hierarchical quantification method for influencing factors in remote sensing target detection in complex scenes, which has the following beneficial effects:
[0032] 1. Overcoming the limitations of uniform sampling, a closed-loop logic of "uniform sampling—performance function—performance grading—parameter inverse solution" is constructed. Data across the entire range is obtained through uniform sampling, and a correlation function between performance and influencing factors is established. Using uniform segmentation of model performance as the causal starting point, the parameter grading interval is inferred backwards, achieving a forward causal derivation from performance requirements to parameter grading. This abandons the empirical uniform sampling grading logic and provides a scientific, accurate, and engineering-feasible grading and quantification scheme for influencing factors in complex remote sensing target detection scenarios.
[0033] 2. By taking model performance as the core of classification, performance-driven approaches replace experience-based debugging, enabling parameter classification to accurately match the refined needs of remote sensing target detection. This scientifically quantifies the impact of parameters on performance, effectively reducing costs and significantly improving the feasibility and efficiency of engineering implementation. Attached Figure Description
[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.
[0035] Figure 1 The flowchart illustrates a method for hierarchical quantification of influencing factors in remote sensing target detection in complex scenes according to an embodiment of the present invention. Detailed Implementation
[0036] The description of the embodiments in this specification should be taken in conjunction with the accompanying drawings, which should form part of the complete specification. In the drawings, the shape or thickness of the embodiments may be exaggerated and may be indicated in a simplified or convenient manner. Furthermore, parts of the various structures in the drawings will be described separately; it is worth noting that elements not shown in the figures or not described in words are in a form known to those skilled in the art.
[0037] The descriptions of the embodiments herein, including any references to directions and orientations, are for ease of description only and should not be construed as limiting the scope of the invention. The following description of preferred embodiments involves combinations of features, which may exist independently or in combination; the invention is not particularly limited to the preferred embodiments. The scope of the invention is defined by the claims. Figure 1 As shown; Specific Implementation Method 1
[0039] This embodiment provides a hierarchical quantification method for influencing factors in remote sensing target detection in complex scenes, the method comprising:
[0040] To identify at least one influencing factor in the entire remote sensing target detection chain; the entire remote sensing target detection chain includes data acquisition, data preprocessing, remote sensing target detection model training, and target detection output;
[0041] Uniform sampling is performed within the range of influencing factors to obtain multiple parameter values for influencing factors;
[0042] For each influencing factor parameter value, a perturbation or transformation corresponding to that influencing factor parameter value is applied to the standard test dataset to generate a corresponding test sample set; then the test sample set is input into a pre-configured remote sensing target detection model for inference to obtain the corresponding performance index.
[0043] Based on the parameter values of multiple influencing factors and their corresponding performance indicators, a functional relationship between the performance indicators and the parameter values of the influencing factors is obtained by fitting the data.
[0044] The performance index is evenly divided into N performance intervals; N is a positive integer greater than 1.
[0045] Based on the functional relationship, the corresponding range of influencing factor parameter values is obtained by back-calculating each performance range, thereby dividing the influencing factors into N levels.
[0046] In this embodiment, the detailed process of the influencing factor classification and quantification method is as follows:
[0047] a. Factor analysis: From the entire chain, we screen out the subdivided influencing factors such as payload resolution, target scale, and terrestrial environment;
[0048] Specifically, starting from the performance of remote sensing target detection models (accuracy, recall, AP, etc.), we sorted out and defined multiple major categories of core influencing factors across the entire chain, clarified whether there is a correlation between various factors and model performance, and based on the core principle of "whether the factor directly affects the model performance", combined with technical specifications and engineering practices in the field of remote sensing, we divided the influencing factors across the entire chain into five major categories of core influencing factors: platform factors, payload factors, environmental factors, target factors, and target camouflage factors.
[0049] b. Uniform sampling settings: For each subdivided influencing factor, determine the range of its influencing factor parameter values, and use uniform sampling to select multiple sample points at equal intervals within the range;
[0050] Performance evaluation: Substitute the influencing factor parameter values corresponding to each sample point into the preset remote sensing target detection model (such as YOLO, Faster R-CNN), run it on the standard test dataset (such as DOTA) (i.e. apply perturbations or transformations corresponding to the influencing factor parameter values to the standard test dataset, such as adjusting the image spatial resolution, superimposing cloud and fog occlusion, simulating target camouflage, etc.), and record the model's accuracy, recall and other performance indicators.
[0051] Specifically, by uniformly sampling and traversing the parameter range, complete correlation data between "parameters and performance" is obtained. The core purpose is to construct the functional relationship between performance and subdivided factors, providing a direct basis for subsequent hierarchical quantification, while balancing data completeness and ease of implementation.
[0052] b1. Uniform Sampling Scheme Design: For a single sub-factor, sample points are selected using an equally spaced uniform sampling method within its defined parameter value range. The number of sample points is set to 12-18 to ensure coverage of key nodes across the entire parameter range, avoiding sampling redundancy while capturing performance variation patterns.
[0053] For example: within the load resolution range of 0.3m-6m, 15 sample points (0.3m, 0.7m, 1.1m, ..., 6.0m) are selected at 0.4m intervals.
[0054] b2. Algorithm training: Based on the parameter values of the influencing factors corresponding to the selected sample points, several remote sensing images with the characteristics of the corresponding sample points are generated through simulation (including typical complex scenes such as cloud and fog obscuration and target camouflage), and the remote sensing target detection and recognition algorithm is trained.
[0055] For example, regarding the resolution of influencing factor loads, remote sensing images with resolutions of 0.3m, 0.7m, 1.1m, ..., 6.0m can be generated through simulation. Then, a remote sensing target detection model (such as the YOLOv5 algorithm) can be used to train the remote sensing image data to achieve the identification of specific targets (such as aircraft).
[0056] b3. Model testing: A certain number of remote sensing images are generated through simulation as a test set for target detection and recognition, and then the trained remote sensing target detection model is used for inference.
[0057] b4. Data Acquisition: The acquired performance indicators are evaluation indicators related to the remote sensing target detection model, such as average precision (AP), forming a raw data matrix of "parameter values - performance indicators".
[0058] b5. Construction of Performance Correlation Function: Based on the complete original data matrix, a fitting algorithm (such as multinomial fitting, nonlinear regression, etc.) is used to construct the functional relationship between the performance indicators of the remote sensing target detection model and the parameter values of influencing factors.
[0059]
[0060] in, For performance indicators, The parameters of the influencing factors are used to ensure that the function can accurately reflect the influence of changes in the parameters of the influencing factors on the performance indicators throughout the entire range.
[0061] c. Quantitative analysis: Based on the fluctuation trend of performance indicators with the changes in the parameter values of influencing factors, the degree of influence of the subdivided factor on the remote sensing target detection model is qualitatively or semi-quantitatively determined, and then classified into simple levels such as "high influence" and "low influence" to provide a reference for process optimization.
[0062] Specifically, through the function By dividing the data into uniform segments, the influencing factor parameter values can be deduced. The classification range enables a shift from performance-oriented to parameter-based classification, addressing the issues of existing classifications being too subjective and lacking specificity.
[0063] c1. Evenly segmented performance metrics: for the constructed functions Core performance indicators (such as AP) are selected as the grading benchmark, taking into account both differentiation and engineering practicality, and their value range is evenly divided into several levels.
[0064] For example: if the AP value ranges from 50% to 90%, then it is evenly divided into , , , Four intervals;
[0065] c2. Solving the graded parameter interval: Based on the uniform segmentation results of the performance index, through a function... Inverse kinematics yields the corresponding influencing factor parameter values. The range, that is, the range for each performance metric. Solve the inequality
[0066] ,
[0067] Obtain the parameter values of influencing factors The range of values forms a one-to-one correspondence between "performance level" and "parameter range".
[0068] For example: the performance parameter of payload resolution is AP= (Here, payload resolution refers to remote sensing images with the corresponding resolution.) When the AP interval is... When the inverse solution is used, the resolution range is obtained as follows: This means that the resolution range corresponds to a high-performance level.
[0069] c3. Establishing Grading Standards: Based on engineering practice needs, define clear grading attributes for each parameter's grading range. For example, it can be divided into multiple levels, from critical to optimizable.
[0070] The method of this implementation first constructs an influencing factor system. When analyzing the influencing factors, it proposes a scheme of uniform sampling modeling plus performance-oriented grading to achieve scientific grading and accurate quantification of influencing factors. This can provide a basis for the whole-chain optimization such as data preprocessing and model tuning, thereby improving the accuracy and robustness of remote sensing target detection and recognition. Specific Implementation Method Two
[0072] This embodiment is a further explanation of embodiment one. In this embodiment, the influencing factors include platform factors, load factors, environmental factors, target factors, and target camouflage factors.
[0073] Platform-related factors include platform type and platform sophistication;
[0074] Load-related factors include load type and load resolution;
[0075] Environmental factors include atmospheric decay and light intensity;
[0076] Target-related factors include target scale and target operating conditions;
[0077] Target camouflage factors include optical camouflage and infrared camouflage.
[0078] This embodiment involves the specific composition of the influencing factors.
[0079] Step a involves identifying five core influencing factors from the entire chain:
[0080] a1. Platform Characteristics: These correspond to the hardware and operational characteristics of the remote sensing data acquisition platform, which directly affect the stability and integrity of data acquisition.
[0081] a2. Payload characteristics: These are the technical parameters of the remote sensing imaging payload and are the core factors that determine the quality of the original image.
[0082] a3. Environmental characteristics: These correspond to the external scene environment in which the target is located, and directly interfere with the distinction between the target and the background in the image.
[0083] a4. Target characteristics: These correspond to the inherent attributes of the target itself and determine the target's identifiability in the image;
[0084] a5. Target Camouflage Category: Corresponds to the artificial or natural camouflage attributes of the target, specifically addressing the pain point of detecting camouflaged targets in complex scenarios.
[0085] Based on the five major categories of core influencing factors identified in step a, we accurately extracted the sub-factors that significantly impact model performance and defined their parameter ranges, laying the foundation for subsequent quantitative analysis. Typical examples of these sub-factors are as follows:
[0086] Platform characteristics: platform type, platform height, etc.;
[0087] Load characteristics include: load type, load resolution, etc.
[0088] Environmental characteristics: atmospheric attenuation, light intensity, etc.;
[0089] Target characteristics: target size, target operating conditions, etc.
[0090] Target camouflage: optical camouflage, infrared camouflage, etc.;
[0091] In defining the range of parameter values, for each selected sub-factor, the reasonable range of values is determined by combining engineering practice and industry standards.
[0092] For example, the payload resolution (spatial) ranges from 0.3m to 6m (covering the resolution space of commonly used remote sensing equipment). Specific Implementation Method 3
[0094] This embodiment is a further explanation of embodiment one or two. In this embodiment, the performance indicators include at least one of accuracy, recall, average precision, and F1 score.
[0095] In this embodiment, the specific structure of step b4 for constructing the original data matrix is described.
[0096] In b4, the collected performance metrics include precision, recall, average precision (AP), and F1 score, forming a raw data matrix of "parameter value - precision - recall - AP - F1 score". Specific Implementation Method Four
[0098] This embodiment is a further explanation of embodiment three. In this embodiment, N is 4.
[0099] The performance index values are evenly divided into four performance intervals, which correspond to the first, second, third and fourth levels, respectively; the first level corresponds to the highest performance interval value, and the fourth level corresponds to the lowest performance interval value.
[0100] In this embodiment, the grading standard established in step c3 is involved.
[0101] Step c3, based on engineering practice requirements, defines clear level attributes for each parameter's grading range. For example:
[0102] Critical level: Corresponds to the highest performance range (e.g., AP ≥ 80%), with optimal performance within the parameter range, and is the core focus area for engineering optimization;
[0103] Importance level: Corresponds to the second-highest performance range (e.g., 70%≤AP<80%), with good performance within the parameter range, requiring close attention;
[0104] Secondary level: Corresponds to the medium performance range (e.g., 60%≤AP<70%), performance meets the standard within the parameter range, and can maintain normal control;
[0105] Level to be optimized: This corresponds to the low performance range (e.g., AP < 60%), where performance is insufficient within the parameter range and needs adjustment and optimization. Detailed Implementation Method Five
[0107] This embodiment is a further explanation of embodiment one, two or four. In this embodiment, for multiple influencing factor parameter values of any influencing factor, the corresponding performance index is obtained through reasoning using the same pre-configured remote sensing target detection model. Specific Implementation Method Six
[0109] This embodiment is a further explanation of embodiment five. In this embodiment, when the influencing factor is a non-numerical factor, multiple types are selected from the available types of the non-numerical factor, and the corresponding performance index is obtained for each type through inference using a pre-configured remote sensing target detection model; then, each type is classified into corresponding levels according to the level of the performance index.
[0110] In this embodiment, when considering non-numerical factors such as platform type, payload type, optical camouflage, and infrared camouflage, multiple types are selected from the available types of these non-numerical factors, and the performance index corresponding to each type is obtained. Then, each type is classified into corresponding levels based on the performance index.
[0111] The platform type can be either satellite or unmanned aerial vehicle (UAV); the payload type can be either visible light, infrared, or synthetic aperture radar (SAR).
[0112] For each of the above values, the same pre-configured target detection model is used for inference, and the corresponding performance index is measured.
[0113] Taking performance metrics as an example, if AP = 0.82 for a satellite platform and AP = 0.76 for a drone platform, then "satellite" is classified as critical and "drone" as important in terms of platform type. Similarly, if SAR payload AP = 0.68 and visible light payload AP = 0.85, then SAR is classified as secondary and visible light is classified as critical.
[0114] Ultimately, this will achieve hierarchical quantification consistent with numerical factors. Detailed Implementation Method Seven
[0116] This embodiment is a further explanation of embodiment five. This embodiment also includes:
[0117] Output a quantitative report; the quantitative report includes influencing factors, corresponding levels, corresponding influencing factor parameter value ranges, and corresponding level labels; the level labels include at least one of critical level, important level, minor level, and level to be optimized.
[0118] The critical level corresponds to the highest performance range, the important level corresponds to the second highest performance range, the minor level corresponds to the medium performance range, and the level to be optimized corresponds to the low performance range.
[0119] In this embodiment, step e is also involved, which is to carry out the process of full-link adaptation and result summary. By outputting a quantitative report of "subdivision factors - performance level - parameter range - level attribute", the technical requirements and optimization direction of each level range are clarified.
[0120] Specifically, the above process is completed for all sub-factors across the entire chain, and the results are summarized to provide a practical and accurate basis for optimizing the testing process, thereby realizing the engineering value of the technical solution:
[0121] e1. Full-chain factor traversal execution: For all the subdivided influencing factors (covering the five major categories of platform, load, environment, target, and target camouflage) selected in step (a), repeat the process of steps (b) to (c) one by one to complete the uniform sampling, performance function construction, and hierarchical quantification of each factor to ensure that no factor is missed in the entire chain.
[0122] e2. Summary of multi-factor quantification results: Construct a comprehensive table of full-chain subdivided factors for hierarchical quantification, which horizontally covers all subdivided factors (marked with their respective core categories) and vertically includes fields such as "parameter range, performance level, and level attribute".
[0123] The present invention provides a hierarchical quantification method for influencing factors in remote sensing target detection in complex scenarios, comprising: acquiring at least one influencing factor in the entire chain of remote sensing target detection; uniformly sampling within the value range of the influencing factor to obtain multiple influencing factor parameter values; for each influencing factor parameter value, applying a perturbation or transformation corresponding to that parameter value to a standard test dataset to generate a corresponding test sample set; inputting the test sample set into a pre-configured remote sensing target detection model for inference to obtain corresponding performance indicators; fitting a functional relationship between the performance indicators and the influencing factor parameter values based on the multiple influencing factor parameter values and the corresponding performance indicators; uniformly dividing the value range of the performance indicators into N performance intervals; and, based on the functional relationship, back-calculating the corresponding influencing factor parameter value intervals through each performance interval, thereby classifying the influencing factors into N levels.
[0124] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.
[0125] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. 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, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0126] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate 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 functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal 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.
[0127] It should also be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device 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 terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0128] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.
Claims
1. A method for hierarchical quantification of influence factors for complex scene remote sensing target detection, characterized in that, The methods include: Obtain at least one influencing factor in the entire chain of remote sensing target detection; The entire chain of remote sensing target detection includes data acquisition, data preprocessing, remote sensing target detection model training, and target detection output; Uniform sampling is performed within the range of values of the influencing factors to obtain multiple parameter values of the influencing factors; For each influencing factor parameter value, a perturbation or transformation corresponding to that influencing factor parameter value is applied to the standard test dataset to generate a corresponding test sample set; then the test sample set is input into a pre-configured remote sensing target detection model for inference to obtain the corresponding performance index. Based on the parameter values of the multiple influencing factors and the corresponding performance indicators, a functional relationship between the performance indicators and the parameter values of the influencing factors is obtained by fitting the performance indicators. The range of values for the performance index is evenly divided into N performance intervals; N is a positive integer greater than 1. Based on the aforementioned functional relationship, the corresponding range of influencing factor parameter values is obtained by reverse deduction from each performance range, thereby dividing the influencing factors into N levels.
2. The influence factor grading quantization method for complex scene remote sensing target detection according to claim 1, characterized in that, The influencing factors include platform-related factors, payload-related factors, environmental factors, target-related factors, and target camouflage-related factors; The platform-related factors include platform type and platform level. The load-related factors include load type and load resolution; The environmental factors mentioned include atmospheric attenuation and light intensity; The target factors include target scale and target operating conditions; The target camouflage factors include optical camouflage and infrared camouflage.
3. The influence factor grading quantization method for complex scene remote sensing target detection according to claim 1 or 2, characterized in that, Performance metrics include at least one of accuracy, recall, average precision, and F1 score.
4. The influence factor grading quantization method for complex scene remote sensing target detection according to claim 1, characterized in that, The value of N is 4; The performance index is evenly divided into four performance intervals, which correspond to the first level, the second level, the third level and the fourth level, respectively; and the performance interval corresponding to the first level has the highest value and the performance interval corresponding to the fourth level has the lowest value.
5. The method of claim 1, 2 or 4, wherein, For any given influencing factor, multiple influencing factor parameter values are inferred using the same pre-configured remote sensing target detection model to obtain the corresponding performance indicators.
6. The hierarchical quantification method for influencing factors in remote sensing target detection in complex scenes according to claim 5, characterized in that, When the influencing factor is a non-numerical factor, multiple types are selected from the available types of the non-numerical factor, and for each type, the corresponding performance index is obtained through inference using a pre-configured remote sensing target detection model; then, each type is classified into corresponding levels according to the level of the performance index.
7. The hierarchical quantification method for influencing factors in remote sensing target detection in complex scenes according to claim 5, characterized in that, Also includes: Output a quantitative report; the quantitative report includes influencing factors, corresponding levels, corresponding ranges of influencing factor parameter values, and corresponding level labels; The level labels include at least one of critical level, important level, minor level, and level to be optimized. The critical level corresponds to the highest performance range, the important level corresponds to the second highest performance range, the minor level corresponds to the medium performance range, and the level to be optimized corresponds to the low performance range.
8. An electronic device, characterized in that, include: One or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, and the one or more computer programs are stored in the memory, and when the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the hierarchical quantification method for influencing factors of remote sensing target detection in complex scenes as described in any one of claims 1 to 7.
9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, implement the hierarchical quantification method for influencing factors in remote sensing target detection in complex scenes as described in any one of claims 1 to 7.