Data-driven visible light imaging satellite recognition efficiency evaluation method and device
By accessing and preprocessing visible light imaging satellite data, quantifying the recognition performance factor and using BP neural network to train the recognition efficiency metric model, the problems of low computing timeliness and inaccurate recognition accuracy in traditional methods are solved, and fast and accurate recognition efficiency evaluation is achieved.
Patent Information
- Application Number
- CN202411596793.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-11
- Publication Date
- 2025-03-21
AI Technical Summary
When evaluating the recognition performance of visible light imaging satellites, the traditional Monte Carlo test method has low calculation timeliness, and the calculation of target recognition accuracy is inaccurate when the number of samples is small, and the performance gap between theoretical models and practical applications is large.
By connecting to visible light imaging satellite image data, environmental data and satellite state data, preprocessing is carried out to form comprehensive data, quantifying the recognition performance factor, generating an identification performance metric model based on BP neural network training, and dynamically assessing the identification performance in real time.
It realizes the recognition efficiency of visible light imaging sensors quickly and accurately, reduces the computing timeliness, and improves the scientificity and close to reality of the recognition accuracy.
Smart Images
Figure CN119672552A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the intersection of visible light image processing and artificial intelligence, and in particular to a data-driven visible light imaging satellite recognition efficiency evaluation method and device. Background Art
[0002] With the rapid development of aerospace remote sensing technology, a large number of sub-meter resolution space visible light imaging satellites have emerged, making visible light imaging technology play an increasingly important role. At present, the recognition accuracy index is usually used to evaluate the recognition performance of visible light imaging satellites. Traditional recognition accuracy measurement methods are mostly static quantification, statistical quantification based on Monte Carlo experiments, etc., especially the Monte Carlo experiment method has a large number of samples and low calculation timeliness. If the number of samples is small, the target recognition accuracy calculation will be inaccurate.
[0003] In order to solve the shortcomings of Monte Carlo calculation of target recognition accuracy, considering the influencing factors such as platform, sensor and environment, a quantitative analytical formula for target recognition accuracy was constructed through theoretical modeling, which greatly improved the timeliness of calculation. However, there is a large performance gap between the theoretical model and the actual application, and it is impossible to accurately evaluate and quantify the recognition capability of visible light imaging satellites. Therefore, how to accurately and quickly measure the recognition performance of imaging satellites is a difficult problem that needs to be studied and overcome urgently. Summary of the invention
[0004] The present invention aims to provide a data-driven visible light imaging satellite recognition efficiency evaluation method and device, so as to accurately and quickly measure the recognition efficiency of visible light imaging sensors.
[0005] The present invention provides a data-driven visible light imaging satellite recognition efficiency evaluation method, comprising the following steps:
[0006] Step 1, accessing visible light imaging satellite image data, environmental data and satellite status data, and preprocessing the visible light imaging satellite image data, environmental data and satellite status data to form comprehensive data;
[0007] Step 2, quantify the recognition efficiency factor of the comprehensive data;
[0008] Step 3: Generate model training data based on the recognition efficiency factor and the target recognition algorithm label annotation results, and generate a recognition efficiency measurement model of the visible light imaging sensor for different types of targets through BP neural network training;
[0009] Step 4: Access the image data, environmental data, and satellite status data of the visible light imaging sensor in real time, and dynamically evaluate the recognition efficiency of the visible light imaging sensor for different types of targets based on the recognition efficiency measurement model.
[0010] Furthermore, in step 1:
[0011] The preprocessing of visible light satellite image data includes: dividing the visible light sensor imaging data into target background areas by selecting a surrounding background area of 9 times the target area with the target as the center, and classifying and sorting the target background image data formed according to the target type and time sequence;
[0012] The preprocessing of environmental data and satellite status data includes: organizing the environmental data according to geographical location, and then sorting them in chronological order for different geographical locations; sorting the satellite status data in chronological order.
[0013] In one embodiment, in step 2, the identification performance factors of the comprehensive data are quantified including exposure performance, cloud performance, color performance, and satellite vibration performance.
[0014] In one embodiment, the exposure efficiency is: obtaining exposure quantity through image data to quantify the exposure efficiency factor.
[0015] In one embodiment, the cloud layer effectiveness is: quantifying the cloud layer effectiveness through the correspondence between cloud layer thickness and cloud amount level.
[0016] In one embodiment, according to the degree of distinction between the target and the background determined by human subjectivity, the color contrast level of the target and the background in the image data is quantified, and the color effectiveness is calculated based on the color contrast level.
[0017] In one embodiment, the satellite vibration effectiveness is: based on the influence of satellite lateral vibration, considering different situations of satellite low-frequency vibration and high-frequency vibration, to quantify the satellite vibration effectiveness.
[0018] Furthermore, the satellite vibration effectiveness needs to be calculated differently based on the satellite vibration frequency and amplitude.
[0019] Further, step 3 includes:
[0020] Based on the comprehensive data of target image background data, environmental data, and satellite status data, the exposure effectiveness factor, atmospheric effectiveness factor, color effectiveness factor, and satellite vibration effectiveness factor of comprehensive data of different target types are calculated;
[0021] The value range of each performance factor is equally divided into several intervals, and different performance factor interval combination categories for different target types are generated through permutation and combination;
[0022] Use typical target recognition algorithms to identify targets, and count the target recognition accuracy rates under different target types and different performance factor interval combinations;
[0023] Assign the corresponding target recognition accuracy rate to the data under each combination category as a label to form model training data;
[0024] According to the variable step-size sliding time window method, with the newly added data time period as the variable step-size, several target types of visible light imaging samples are selected in reverse order from the latest time data as model training data;
[0025] The model training data is divided into a training set and a test set, a BP neural network is constructed, and the network parameters including the target error, learning rate and maximum number of iterations are set. Finally, the training set and the test set are input into the constructed BP neural network training. After the training is completed, a recognition efficiency measurement model is generated.
[0026] The present invention also provides a data-driven visible light imaging satellite recognition efficiency evaluation device, comprising:
[0027] A visible light sensor imaging data preprocessing module is used to access visible light imaging satellite image data, environmental data and satellite status data, and preprocess the visible light imaging satellite image data, environmental data and satellite status data to form comprehensive data;
[0028] The recognition effectiveness factor measurement module is used to quantify the recognition effectiveness factor of the comprehensive data;
[0029] The recognition efficiency measurement model training module is used to generate model training data based on the recognition efficiency factor and the target recognition algorithm label annotation results, and to generate the recognition efficiency measurement model of the visible light imaging sensor for different types of targets through BP neural network training;
[0030] The recognition accuracy calculation module is used to access the image data, environmental data and satellite status data of the visible light imaging sensor in real time, and dynamically evaluate the recognition efficiency of the visible light imaging sensor for different types of targets based on the recognition efficiency measurement model.
[0031] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0032] 1. Based on the principle of visible light imaging and target recognition, the present invention takes into account the influencing factors such as sensors, platforms and environment, analyzes and constructs measurement models of exposure efficiency factor, atmospheric efficiency factor, color efficiency factor and satellite vibration efficiency factor, making the quantification of satellite recognition efficiency for different targets more scientific.
[0033] 2. The present invention uses the sensor imaging recognition efficiency factor as the input of the recognition efficiency measurement model, and based on the associated annotation results of actual image data, environmental data, satellite status data, etc., through offline training of the BP neural network, the generated quantitative model is faster and closer to reality when calculating the recognition accuracy than the traditional method of Monte Carlo experiment. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 The flowchart of the data-driven visible light imaging satellite recognition performance evaluation method in an embodiment of the present invention. DETAILED DESCRIPTION
[0035] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0036] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0037] Example
[0038] like Figure 1 As shown, this embodiment proposes a data-driven visible light imaging satellite recognition performance evaluation method, comprising the following steps:
[0039] Step 1: The visible light sensor imaging data preprocessing module accesses the visible light imaging satellite image data, environmental data and satellite status data, and preprocesses the visible light imaging satellite image data, environmental data and satellite status data to form comprehensive data; the preprocessing operation in this embodiment includes target background extraction, time sorting, etc. The specific method is as follows:
[0040] (1) Preprocessing of visible light satellite image data
[0041] The visible light sensor imaging data is divided into the target background area by selecting a surrounding background area of 9 times the target area with the target as the center. The resulting target background image data is classified and sorted according to the target type and time sequence.
[0042] (2) Preprocessing of environmental data and satellite status data
[0043] The environmental data are organized according to geographical location, and then sorted in chronological order for different geographical locations; the satellite status data are sorted in chronological order.
[0044] Step 2: The recognition efficiency factor measurement module quantifies the recognition efficiency factor of the comprehensive data from the dimensions of exposure efficiency, cloud efficiency, color efficiency, satellite vibration efficiency, etc. The specific method flow is as follows:
[0045] (1) Exposure efficiency
[0046] The exposure efficiency factor is quantified by obtaining the exposure amount through image data. The specific quantification formula is:
[0047]
[0048] Among them, E0 is the optimal exposure, σ is the camera sensitivity, and E is the exposure of the linear array CCD device.
[0049] (2) Cloud performance
[0050] The cloud layer efficiency is quantified through the correspondence between cloud thickness and cloud amount level. The specific formula is:
[0051]
[0052] Among them, N cloud It is the cloud cover level, which is divided into 9 levels and can be checked according to the cloud thickness and level comparison table.
[0053]
[0054] (3) Color performance
[0055] According to the subjective judgment of the degree of distinction between the target and the background, the color contrast level of the target and the background in the image data is quantified, and the color effectiveness f is calculated based on the color contrast level r , the specific formula is as follows:
[0056]
[0057] Among them, N color It is the color contrast level, which is divided into 9 levels and is determined by human subjective judgment.
[0058] (4) Satellite vibration performance
[0059] Based on the influence of satellite lateral vibration, the satellite vibration efficiency f is quantified by considering different situations of satellite low-frequency vibration and high-frequency vibration. s The specific formula is as follows:
[0060] ① Satellite low-frequency vibration with small amplitude
[0061] f s =|sinc(πN c d m )|
[0062] in, is the Symplectic function, N c is the spatial frequency, d m is the diffuse spot diameter, in um.
[0063] ② Satellite low-frequency vibration with large amplitude
[0064]
[0065] in, is the displacement distance caused by vibration, is the sine integral function.
[0066] ③ Satellite high-frequency vibration
[0067] f s =J0(2πN c d m )
[0068] in, is the zero-order Bessel function.
[0069] Step 3: The recognition efficiency measurement model training module generates model training data based on the recognition efficiency factor and the target recognition algorithm label annotation results, and generates the recognition efficiency measurement model of the visible light imaging sensor for different types of targets through BP neural network training. The specific method flow is as follows:
[0070] (1) The visible light sensor imaging data preprocessing module associates the visible light imaging satellite image data, environmental data and satellite status data according to time and geographical location to form comprehensive data of various target image background data, environmental data and satellite status data.
[0071] (2) Call the recognition effectiveness factor measurement module to calculate the exposure effectiveness factor, atmospheric effectiveness factor, color effectiveness factor and satellite vibration effectiveness factor of the comprehensive data of different target types.
[0072] (3) The value range of each performance factor is divided into several intervals, and different performance factor interval combination categories for different target types are generated through permutations and combinations.
[0073] (4) Use typical target recognition algorithms to identify targets and calculate the target recognition accuracy under different target types and different performance factor interval combination categories.
[0074] (5) The data under each combination category are assigned corresponding target recognition accuracy as labels to form model training data.
[0075] (6) In order to ensure the efficiency of training, a variable step-size sliding time window is used, with the newly added data time period as the variable step-size, and the data is selected in reverse order from the latest time. The first t Visible light imaging samples of target types are used as model training data.
[0076] (7) Divide the model training data into a training set and a test set according to a certain ratio. For example, use 80% of the samples in the model training data as the training set and 20% of the samples in the model training data as the test set. Construct a BP neural network with 2 hidden layers and no more than 30 nodes. Set network parameters such as target error, learning rate, and maximum number of iterations. Finally, generate a recognition efficiency measurement model through training.
[0077] Step 4: The recognition accuracy calculation module accesses the image data, environmental data, and satellite status data of the visible light imaging sensor in real time, and dynamically evaluates the recognition efficiency of the visible light imaging sensor for different types of targets based on the recognition efficiency measurement model.
[0078] Based on the same concept, this embodiment also provides a data-driven visible light imaging satellite recognition efficiency evaluation device, including:
[0079] A visible light sensor imaging data preprocessing module is used to access visible light imaging satellite image data, environmental data and satellite status data, and preprocess the visible light imaging satellite image data, environmental data and satellite status data to form comprehensive data;
[0080] The recognition effectiveness factor measurement module is used to quantify the recognition effectiveness factor of the comprehensive data;
[0081] The recognition efficiency measurement model training module is used to generate model training data based on the recognition efficiency factor and the target recognition algorithm label annotation results, and to generate the recognition efficiency measurement model of the visible light imaging sensor for different types of targets through BP neural network training;
[0082] The recognition accuracy calculation module is used to access the image data, environmental data and satellite status data of the visible light imaging sensor in real time, and dynamically evaluate the recognition efficiency of the visible light imaging sensor for different types of targets based on the recognition efficiency measurement model.
[0083] The working methods of the above modules can refer to the description of the aforementioned data-driven visible light imaging satellite recognition efficiency evaluation method, which will not be repeated here.
[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A data-driven visible light imaging satellite recognition performance evaluation method, characterized in that: The steps include: Step 1, accessing visible light imaging satellite image data, environmental data and satellite status data, and preprocessing the visible light imaging satellite image data, environmental data and satellite status data to form comprehensive data; Step 2, quantify the recognition efficiency factor of the comprehensive data; Step 3: Generate model training data based on the recognition efficiency factor and the target recognition algorithm label annotation results, and generate a recognition efficiency measurement model of the visible light imaging sensor for different types of targets through BP neural network training; Step 4: Access the image data, environmental data, and satellite status data of the visible light imaging sensor in real time, and dynamically evaluate the recognition efficiency of the visible light imaging sensor for different types of targets based on the recognition efficiency measurement model.
2. The data-driven visible light imaging satellite recognition efficiency evaluation method according to claim 1, characterized in that: In step 1: The preprocessing of visible light satellite image data includes: dividing the visible light sensor imaging data into target background areas by selecting a surrounding background area of 9 times the target area with the target as the center, and classifying and sorting the target background image data formed according to the target type and time sequence; The preprocessing of environmental data and satellite status data includes: organizing the environmental data according to geographical location, and then sorting them in chronological order for different geographical locations; sorting the satellite status data in chronological order.
3. The data-driven visible light imaging satellite recognition efficiency evaluation method according to claim 1, characterized in that: In step 2, the recognition efficiency factors of the comprehensive data are quantified, including exposure efficiency, cloud efficiency, color efficiency, and satellite vibration efficiency.
4. The data-driven visible light imaging satellite recognition efficiency evaluation method according to claim 4 is characterized in that: The exposure efficiency is: obtaining the exposure quantity through image data to quantify the exposure efficiency factor.
5. The data-driven visible light imaging satellite recognition efficiency evaluation method according to claim 4, characterized in that: The cloud layer efficiency is to quantify the cloud layer efficiency through the corresponding relationship between cloud layer thickness and cloud amount level.
6. The data-driven visible light imaging satellite recognition efficiency evaluation method according to claim 4, characterized in that: According to the subjective judgment of the degree of distinction between the target and the background, the color contrast level of the target and the background in the image data is quantified, and the color effectiveness is calculated based on the color contrast level.
7. The data-driven visible light imaging satellite recognition efficiency evaluation method according to claim 4, characterized in that: The satellite vibration effectiveness is: based on the influence of satellite lateral vibration, considering different situations of satellite low-frequency vibration and high-frequency vibration, and quantifying the satellite vibration effectiveness.
8. The data-driven visible light imaging satellite recognition efficiency evaluation method according to claim 7, characterized in that: The satellite vibration effectiveness needs to be calculated based on the satellite vibration frequency and amplitude.
9. The data-driven visible light imaging satellite recognition efficiency evaluation method according to claim 1, characterized in that: Step 3 includes: Based on the comprehensive data of target image background data, environmental data, and satellite status data, the exposure effectiveness factor, atmospheric effectiveness factor, color effectiveness factor, and satellite vibration effectiveness factor of comprehensive data of different target types are calculated; The value range of each performance factor is equally divided into several intervals, and different performance factor interval combination categories for different target types are generated through permutation and combination; Use typical target recognition algorithms to identify targets, and count the target recognition accuracy rates under different target types and different performance factor interval combinations; Assign the corresponding target recognition accuracy rate to the data under each combination category as a label to form model training data; According to the variable step-size sliding time window method, with the newly added data time period as the variable step-size, several target types of visible light imaging samples are selected in reverse order from the latest time data as model training data; The model training data is divided into a training set and a test set, a BP neural network is constructed, and the network parameters including the target error, learning rate and maximum number of iterations are set. Finally, the training set and the test set are input into the constructed BP neural network training. After the training is completed, a recognition efficiency measurement model is generated.
10. A data-driven visible light imaging satellite recognition efficiency evaluation device, characterized in that: include: A visible light sensor imaging data preprocessing module is used to access visible light imaging satellite image data, environmental data and satellite status data, and preprocess the visible light imaging satellite image data, environmental data and satellite status data to form comprehensive data; The recognition effectiveness factor measurement module is used to quantify the recognition effectiveness factor of the comprehensive data; The recognition efficiency measurement model training module is used to generate model training data based on the recognition efficiency factor and the target recognition algorithm label annotation results, and to generate the recognition efficiency measurement model of the visible light imaging sensor for different types of targets through BP neural network training; The recognition accuracy calculation module is used to access the image data, environmental data and satellite status data of the visible light imaging sensor in real time, and dynamically evaluate the recognition efficiency of the visible light imaging sensor for different types of targets based on the recognition efficiency measurement model.