A method for full-process quality control of an automated production line for remote sensing data
By using metadata pre-inspection and automatic quality inspection methods with node quality gates, the problem of random transmission and tampering of quality issues in the automated production line of remote sensing data is solved, achieving traceability and reproducibility of data quality and forming a closed-loop feedback mechanism for sensor performance.
Patent Information
- Application Number
- CN202611518439.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-29
- Publication Date
- 2026-11-17
AI Technical Summary
Existing automated production lines for remote sensing data lack effective quality transfer and interception mechanisms, causing data quality issues to be passed on to downstream nodes, resulting in high rework costs. Furthermore, the lack of systematic records and cross-batch quality traceability makes product quality files susceptible to tampering, and sensor performance drift lacks closed-loop feedback.
Pre-inspection feature vectors are generated through metadata pre-inspection, the processing node sequence is determined and parameters are bound, node quality gates are set for automatic quality inspection, correction parameters and failure parameters are recorded, a quality inspection evidence chain is constructed, and cross-batch quality trend analysis and on-orbit parameter correction are realized.
It enables the interception and processing of remote sensing data quality within nodes, reduces invalid calculations, ensures the traceability and reproducibility of the processing process, prevents the tampering of quality records, and forms a closed-loop feedback between ground quality inspection and satellite operation and control.
Smart Images

Figure CN123118034A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of remote sensing data processing and quality control technology, specifically to a method for full-process quality control of an automated remote sensing data production line. Background Technology
[0002] Existing automated remote sensing data production lines typically perform radiometric correction, geometric correction, orthorectification, fusion, and mosaicking on input data in a fixed node sequence. Quality inspection is often located at the end of production or only involves simple spot checks after each node is completed. In this approach, metadata is often only used for archiving, retrieval, and basic information recording, and is not involved in the decision-making of the processing path before production begins. This results in data with excessive side-view limits, abnormal sensor gain, or high cloud cover still being processed according to standard procedures, consuming computing resources and potentially affecting the quality of subsequent products. Furthermore, there is a lack of effective quality transfer and interception mechanisms between processing nodes. Deviations from one node can flow to downstream nodes with the data, sometimes not being discovered until after product output, leading to high rework costs. When quality issues arise, existing solutions rely heavily on manual judgment and parameter tuning, making reprocessing difficult to reproduce, and failure parameters lack systematic recording. Product quality archives are usually stored in ordinary metadata files, and node quality inspection results are independent of each other, making it difficult to prove that they have not been tampered with, and posing challenges for third-party verification and cross-batch quality traceability. Furthermore, the quality variation trends among multiple batches of the same sensor are rarely used for on-orbit parameter correction, and there is a lack of closed-loop feedback between the ground production line and the satellite operation and control and ground calibration systems. Therefore, the existing technology has significant shortcomings. Summary of the Invention
[0003] The purpose of this invention is to provide a method for full-process quality control of an automated production line for remote sensing data, so as to solve the problems mentioned in the background art.
[0004] To address the aforementioned technical problems, this invention provides the following technical solution: a method for full-process quality control of an automated remote sensing data production line, comprising:
[0005] S1. Receive the remote sensing data to be processed and the metadata corresponding to the remote sensing data to be processed, assign a data identifier to the remote sensing data to be processed, and bind the remote sensing data to be processed, the metadata and the data identifier.
[0006] S2. Perform metadata pre-inspection on the obtained metadata. The metadata pre-inspection does not read the pixel content of the remote sensing data to be processed. Extract at least two pre-inspection features from the obtained metadata to form a pre-inspection feature vector. Match the obtained pre-inspection feature vector with the pre-inspection rule table to obtain the path identifier and the pre-inspection parameter set. Generate a pre-inspection record. The pre-inspection record includes the data identifier, the path identifier, the pre-inspection parameter set, the pre-inspection feature vector, and the pre-inspection timestamp.
[0007] S3. Determine the processing node sequence from the node library of the automated production line based on the obtained path identifier; write the parameters in the obtained pre-inspection parameter set that match the processing type of each processing node in the processing node sequence into the parameter register of the corresponding processing node; and make the upstream quality inspection record set received by the first processing node contain the obtained pre-inspection records.
[0008] S4. For each processing node in the obtained processing node sequence, initialize the reprocessing count of the corresponding processing node to 0. Each processing node receives the input data of the corresponding processing node and the upstream quality inspection record set. The input data of the corresponding processing node is the output data of the previous processing node, and the input data of the first processing node is the remote sensing data to be processed. Based on at least one of the quality inspection records of the preceding nodes in the upstream quality inspection record set and the parameters that match the processing node in the obtained pre-inspection parameter set, determine the activation parameters of the processing node. Process the input data of the processing node according to the node processing logic and the activation parameters to obtain the output data of the processing node. The node quality gate module corresponding to each processing node calculates the node quality inspection index of the corresponding processing node based on the node quality criterion and compares the obtained node quality inspection index with the corresponding node quality criterion.
[0009] If the obtained node quality inspection indicators meet the corresponding node quality criteria, a node quality inspection record for the corresponding processing node is generated. The node quality inspection record includes at least the node quality inspection indicators and a pass flag. The obtained node quality inspection record is added to the upstream quality inspection record set of the corresponding processing node to obtain the downstream quality inspection record set of the corresponding processing node. The output data and downstream quality inspection record set of the corresponding processing node are passed to the next processing node, and the downstream quality inspection record set of the corresponding processing node becomes the upstream quality inspection record set of the next processing node. If the processing node is the last processing node in the obtained processing node sequence, the corresponding output data is used as a candidate product, and the obtained downstream quality inspection record set is passed to the end quality gate verification.
[0010] If the obtained node quality inspection index does not meet the corresponding node quality criteria, then the deviation type and deviation magnitude are determined based on the deviation between the obtained node quality inspection index and the corresponding node quality criteria; the obtained deviation type and deviation magnitude are used to query the parameter adjustment rule table corresponding to the processing node to obtain the correction parameter corresponding to the processing node; the parameter register of the processing node is updated according to the obtained correction parameter, the obtained correction parameter is recorded in the current failure parameter combination corresponding to the processing node, the reprocessing count of the processing node is incremented by 1, and step S4 is re-executed; if the reprocessing count of the processing node reaches the maximum reprocessing count and the obtained node quality inspection index still does not meet the corresponding node quality criteria, then a node-level non-conforming mark is generated, the output data of the processing node, the node-level non-conforming mark and the current failure parameter combination are transferred to the manual review queue, and the transmission to the next processing node is stopped;
[0011] S5. Perform end-of-line quality gate verification on the obtained candidate products to obtain end-of-line quality inspection indicators and end-of-line quality inspection records. If the obtained end-of-line quality inspection indicators do not meet the end-of-line quality criteria, mark the obtained candidate products as end-of-line unqualified and transfer them to the manual review queue. If the obtained end-of-line quality inspection indicators meet the end-of-line quality criteria, construct a quality inspection evidence chain in sequence based on the obtained pre-inspection records, the node quality inspection records corresponding to each processing node, and the obtained end-of-line quality inspection records. Adjacent records in the quality inspection evidence chain are associated with each other through hash values. Output the obtained candidate products and the quality inspection evidence chain together.
[0012] Furthermore, the pre-detection feature vector in S2 includes at least two of the following: side view angle, solar altitude angle, sensor gain status flag, data packet integrity flag, and cloud cover estimate; the pre-detection rule table contains the correspondence between pre-detection feature conditions, path identifiers, and pre-detection parameter sets; if the obtained pre-detection feature vector matches multiple rules simultaneously, the path identifier and pre-detection parameter set corresponding to the rule with the highest priority are selected according to the priority order of the rules in the obtained pre-detection rule table; if the obtained pre-detection feature vector does not match any rule, the path identifier is set to the standard product line, and the pre-detection parameter set is set to the default parameter set; the value of the path identifier includes the standard product line, the downgraded product line, the radiation correction enhancement line, and the delay processing line.
[0013] Furthermore, the rule configuration of the pre-inspection rule table is as follows:
[0014] If the obtained side view is greater than the first threshold, the path is identified as a downgraded product line, and the screening threshold for the geometric correction control point of the pre-inspection parameter set is set to a lenient threshold.
[0015] If the obtained sensor gain status flag is an abnormal value, the path is identified as a radiation correction enhancement line, the initial value of the radiation correction coefficient in the pre-detection parameter set is set to the enhancement coefficient value, and a radiation correction verification node is inserted after the radiation correction node in the obtained processing node sequence.
[0016] If the estimated cloud cover is greater than the second threshold, the path is identified as a delayed processing line, and the cloud mask threshold of the pre-detection parameters is set to a strict threshold.
[0017] If the obtained side view is less than or equal to the first threshold, the obtained sensor gain status flag is normal, and the obtained cloud cover estimate is less than or equal to the second threshold, then the path identifier is determined to be the standard product line.
[0018] The priority order of the rules in the pre-inspection rule table is as follows: radiation correction enhancement line is higher than degraded product line, degraded product line is higher than delay processing line, and delay processing line is higher than standard product line.
[0019] Furthermore, the specific method for determining the processing node sequence from the node library of the automated production line based on the obtained path identifier in S3 includes:
[0020] If the obtained path identifier is a standard product line, the processing node sequence includes, in sequence, a radiation correction node, a geometric correction node, an orthorectification node, a fusion node, and a mosaic node.
[0021] If the obtained path is identified as a radiometric correction enhancement line, the processing node sequence includes, in sequence, a radiometric correction node, a radiometric correction verification node, a geometric correction node, an orthorectification node, a fusion node, and a mosaic node.
[0022] If the obtained path identifier is a downgraded product line, the processing node sequence includes, in sequence, a radiation correction node, a geometric correction node, and an orthophoto correction node;
[0023] If the obtained path identifier is a delayed processing line, the processing node sequence includes cloud mask node, radiation correction node, geometric correction node, orthorectification node, fusion node and mosaic node in sequence.
[0024] The cloud mask node performs a cloud mask operation on the input data according to the cloud mask threshold, and uses the data after the cloud mask operation as the input data of the radiation correction node;
[0025] The parameters in the pre-detection parameter set are written as follows: the initial value of the radiation correction coefficient is written to the parameter register of the radiation correction node; the screening threshold of the geometric correction control point, the DEM interpolation method, and the resampling method are written to the parameter registers of the geometric correction node and the orthorectification node; the fusion weight is written to the parameter register of the fusion node; and the cloud mask threshold is written to the parameter register of the cloud mask node.
[0026] Furthermore, in step S4, the processing node selects and activates corresponding compensation parameters from the obtained pre-inspection parameter set based on the quality inspection records of the preceding nodes in the upstream quality inspection record set. The specific execution method includes:
[0027] If the deviation type of the preceding node is geometric deviation and the deviation magnitude is greater than the third threshold, then the processing node will enable the more stringent control point screening threshold in the obtained pre-detection parameter set.
[0028] If the deviation type of the preceding node is radiation deviation and the deviation magnitude is greater than the fourth threshold, then the processing node will enable the radiation correction coefficient fine-tuning amount in the obtained pre-detection parameter set and add the obtained radiation correction coefficient fine-tuning amount to the initial value of the radiation correction coefficient of the processing node.
[0029] If the deviation type of the preceding node is spectral deviation and the deviation magnitude is greater than the fifth threshold, then the processing node will use the fusion weight adjusted from the obtained pre-detection parameters and use the adjusted fusion weight as the fusion weight of the processing node.
[0030] If the deviation type of the preceding node is edge deviation and the deviation magnitude is greater than the sixth threshold, then the processing node will enable a more stringent resampling method in the obtained pre-detection parameter set, and resample the input data of the previous processing node according to the more stringent resampling method.
[0031] Furthermore, the correspondence between the node quality inspection indicators obtained in S4 and the corresponding node quality criteria includes:
[0032] When the processing node is a radiometric correction node, the obtained node quality inspection index is the correlation coefficient of the pixel value sequence between each band after radiometric correction. The node quality criterion is the interval in which the correlation coefficient should fall. Satisfying the criterion means that the correlation coefficient falls into the interval.
[0033] Specifically, the node quality criterion at this time is the interval of inter-band correlation coefficients, denoted as [r]. min , r max The node quality inspection index is the Pearson correlation coefficient between each band after radiometric correction, denoted as r. Where ba and bb are pixel value sequences of two bands, Let ba and bb be the covariances. and Let ba and bb be the standard deviations of respectively, and let r satisfy the criterion that r falls within [r]. min , r max ].
[0034] When the processing node is a geometric correction node, the obtained node quality inspection index is the root mean square of the residual between the fitted coordinates and the measured coordinates of each control point after geometric correction. The node quality criterion is that the root mean square should not be greater than the first residual threshold. Satisfying the criterion means that the root mean square is not greater than the first residual threshold.
[0035] Specifically, the node quality criterion at this time is the RPC fitting residual threshold, denoted as ε. rpc The node quality inspection index is the root mean square of the RPC fitting residuals of the control points after geometric correction, denoted as RMSE. rpc ,
[0036]
[0037] in, To control the number of points, This represents the x-coordinate value in the fitted coordinates of the k-th control point; This represents the ordinate value in the fitted coordinates of the k-th control point. This represents the x-coordinate value in the measured coordinates of the k-th control point; Let y = the measured coordinate value of the k-th control point, satisfying the criterion. .
[0038] When the processing node is an orthorectification node, the obtained node quality inspection index is the root mean square of the residual between the corrected coordinates and the reference coordinates of each inspection point after orthorectification. The node quality criterion is that the root mean square should not be greater than the second residual threshold. Satisfying the criterion means that the root mean square is not greater than the second residual threshold.
[0039] Specifically, the node quality criterion at this time is the residual threshold of the corresponding points, denoted as ε. gcp The node quality inspection index is the plane error of the inspection point after orthorectification, denoted as RMSE. xy ,
[0040]
[0041] in, For the number of checkpoints, Let x be the x-coordinate value of the orthorectified coordinates of the k-th checkpoint. This represents the ordinate value in the orthorectified coordinate system of the k-th checkpoint. This is the x-coordinate value in the reference coordinates of the k-th checkpoint; The ordinate value is the reference coordinate of the k-th checkpoint; satisfying the criterion. .
[0042] When the processing node is a fusion node, the obtained node quality inspection index includes the spectral fidelity index and spatial sharpness index of the fused image. The node quality criterion includes a spectral threshold that the spectral fidelity index should not be greater than and a sharpness threshold that the spatial sharpness index should not be less than. Satisfying the criterion means that the spectral fidelity index is not greater than the spectral threshold and the spatial sharpness index is not less than the sharpness threshold.
[0043] Specifically, the node quality criterion at this time includes the spectral fidelity threshold ε. spec Spatial sharpness threshold ε spat The node quality inspection indicators include the spectral angle mapping value (SAM) and spatial frequency value (SF) of the fused image.
[0044]
[0045]
[0046] in, To fuse the multispectral band pixel values, The pixel values of the corresponding band after fusion. Let be the pixel value of the merged image at coordinates (x, y). Let (x-1, y) be the pixel value of the merged image at coordinates (x-1, y). Let be the pixel value of the merged image at coordinates (x, y-1). The number of rows in the image. Let SAM be the number of columns in the image, satisfying the criterion SAM≤ε. spec And SF≥ε spat .
[0047] When the processing node is a mosaic node, the obtained node quality inspection index is the difference in grayscale mean between the two images on both sides of the mosaic edge in the edge buffer. The node quality criterion is that the grayscale mean difference should not be greater than the edge grayscale difference threshold. Satisfying the criterion means that the grayscale mean difference is not greater than the edge grayscale difference threshold.
[0048] Specifically, the node quality criterion at this time is the edge grayscale difference threshold ε. seam The node quality inspection index is the grayscale mean difference Δg between the images on both sides of the mosaic edge, where Δg = |μ left -μ right |, where μ left μ represents the average grayscale value of the left-side image at the edge within the edge buffer. right Let Δg be the average grayscale value of the right-side image at the junction within the junction buffer, satisfying the criterion Δg≤ε. seam .
[0049] Furthermore, the method for determining the deviation magnitude in S4 based on the deviation between the obtained node quality inspection index and the corresponding node quality criterion includes:
[0050] If the obtained node quality criterion is an upper limit criterion, then the obtained deviation range is the ratio of the difference between the obtained node quality inspection index and the corresponding threshold to the corresponding threshold; the upper limit criterion means that the node quality inspection index is not greater than the corresponding threshold and the criterion is satisfied.
[0051] If the obtained node quality criterion is a lower limit criterion, then the obtained deviation amplitude is the quotient of the difference between the corresponding threshold and the obtained node quality inspection index divided by the corresponding threshold; the lower limit criterion means that the node quality inspection index is not less than the corresponding threshold and thus meets the criterion.
[0052] If the obtained node quality criterion is an interval-type criterion, then the obtained deviation amplitude is the ratio of the deviation of the obtained node quality inspection index from the upper or lower limit of the corresponding interval to the interval width; the interval-type criterion means that the node quality inspection index falls into the corresponding interval and thus satisfies the criterion.
[0053] If the obtained node quality criteria include multiple sub-criteria, then calculate the sub-item deviation magnitude corresponding to each sub-criteria, and take the maximum value among the sub-item deviation magnitudes as the obtained deviation magnitude.
[0054] The parameter adjustment rule table includes the mapping relationship between deviation type, deviation amplitude range, and correction parameters; the deviation type includes radiometric deviation, geometric deviation, spectral deviation, and edge deviation; the deviation amplitude range includes at least a first amplitude range or a second amplitude range, where the deviation in the second amplitude range is greater than the deviation in the first amplitude range; the correction parameters include at least one of DEM interpolation method, control point screening threshold, resampling kernel function, radiometric calibration coefficient fine-tuning amount, and fusion weight adjustment amount; depending on the type of correction parameter, for method-type parameters, the corresponding parameter in the parameter register of the processing node is updated using a replacement method, and for numerical-type parameters, the corresponding parameter in the parameter register of the processing node is updated using an overlay method.
[0055] Furthermore, in the process of constructing the quality inspection evidence chain in sequence according to the obtained pre-inspection record, the node quality inspection record corresponding to each processing node, and the obtained end quality inspection record in S5, the obtained pre-inspection record contains the previous hash and the current hash. The previous hash is a preset initial hash, and the current hash is generated based on the content of the obtained pre-inspection record except for the current hash field and the previous hash. For the i-th node quality inspection record, the node quality inspection record contains the previous hash and the current hash. The previous hash in the i-th node quality inspection record is taken as the current hash of the last record in the upstream quality inspection record set when the node quality inspection record is generated. The current hash in the i-th node quality inspection record is generated based on the content of the node quality inspection record except for the current hash field and the current hash in the (i-1)-th node quality inspection record.
[0056] For the obtained end-of-line quality inspection record, the current hash in the end-of-line quality inspection record is generated based on the content of the end-of-line quality inspection record excluding the current hash field and the previous hash. The previous hash in the end-of-line quality inspection record is taken as the current hash of the last record in the obtained downstream quality inspection record set.
[0057] The obtained pre-inspection records, the node quality inspection records of each processing node in the obtained processing node sequence, and the end quality inspection records are sequentially concatenated to obtain the corresponding quality inspection evidence chain. When verifying the obtained quality inspection evidence chain, the current hash of each record is recalculated according to the same hash generation method. If the current hash of each recalculated record is consistent with the current hash stored in the obtained quality inspection evidence chain, it is confirmed that the obtained quality inspection evidence chain has not been tampered with. If the current hash of a certain record is inconsistent with the current hash stored in the obtained quality inspection evidence chain, then that record and all subsequent records are marked as tampered with.
[0058] Furthermore, the method also includes a cross-batch quality trend analysis step, used to collect node quality inspection records, node quality inspection indicators, and end-of-batch quality inspection records of multiple batches of products of the same sensor; for each batch, radiation quality inspection index values and geometric quality inspection index values are extracted from the obtained node quality inspection indicators, and arranged in batch time order to form radiation index sequence and geometric index sequence; the sliding window mean method is used to perform trend analysis on the obtained radiation index sequence and geometric index sequence respectively, and calculate the mean and variance within the sliding window;
[0059] When the mean offset within the obtained sliding window exceeds the preset statistical control limit, it is determined that the same sensor has radiometric performance drift or geometric performance drift. When radiometric performance drift is determined, a radiometric calibration coefficient correction is generated based on the obtained mean offset and offset direction. When geometric performance drift is determined, an attitude control offset correction or orbit parameter correction is generated based on the obtained mean offset and offset direction. The obtained radiometric calibration coefficient correction or attitude control offset correction is encapsulated as an on-orbit parameter correction suggestion and pushed to the satellite operation and control system or ground calibration system through a standard interface. After the satellite operation and control system or ground calibration system performs parameter correction, the node quality inspection indicators and end quality inspection records of subsequent batches of products are continuously tracked. If the radiometric quality inspection indicator values or geometric quality inspection indicator values of subsequent batches return to within the statistical control limit, the correction is determined to be effective. Otherwise, a correction effect failure prompt is generated, and the width of the statistical control limit is reduced by a preset ratio to improve the sensitivity of subsequent trend analysis.
[0060] Furthermore, the method also includes setting a quality-efficiency trade-off coefficient, and adjusting the stringency of the obtained node quality criterion and the maximum number of reprocessing based on the obtained quality-efficiency trade-off coefficient; the value range of the quality-efficiency trade-off coefficient is [0,1]; when the obtained quality-efficiency trade-off coefficient increases, the node quality criterion becomes more stringent and the maximum number of reprocessing increases; when the obtained quality-efficiency trade-off coefficient decreases, the node quality criterion becomes more lenient and the maximum number of reprocessing decreases, but the node quality criterion is not lower than the minimum acceptable standard;
[0061] The specific adjustment method for the stringency of the node quality criterion includes: for upper limit criterion, dividing the corresponding threshold by the adjustment factor to obtain the adjusted criterion; for lower limit criterion, multiplying the corresponding threshold by the adjustment factor to obtain the adjusted criterion; for interval criterion, keeping the interval center unchanged, dividing the interval half-width by the adjustment factor to obtain the adjusted interval upper and lower limits; the adjustment factor is equal to the sum of the difference between the obtained quality and efficiency trade-off coefficient and 0.5, multiplied by the preset adjustment sensitivity coefficient, and then added to 1. In this embodiment, the adjustment sensitivity coefficient is greater than 0 and less than 2; the quality and efficiency trade-off coefficient is determined based on at least one of the current production line's computing resource load status, product quality level requirements, and historical node quality inspection pass rate; when the product quality level requirement is high precision level, or the historical node quality inspection pass rate is lower than the pass rate threshold, the quality and efficiency trade-off coefficient is increased; when the computing resource load status is higher than the load threshold, the quality and efficiency trade-off coefficient is decreased.
[0062] Compared with the prior art, the beneficial effects achieved by the present invention are:
[0063] (1) The metadata pre-inspection of the present invention uses only metadata and does not read pixel content, which can divert data to the processing node sequence that matches its quality status before production begins, reducing invalid calculations;
[0064] (2) In this invention, a node quality gate is set at each processing node. When the node quality inspection index does not meet the node quality criterion, it will not be passed downstream. Instead, the parameter adjustment rule table will be queried according to the deviation type and deviation magnitude and automatic reprocessing will be performed, so that the quality problem can be intercepted and processed inside the node.
[0065] (3) In this invention, by recording the combination of correction parameters and failure parameters during the reprocessing process, the maximum number of reprocessing times is reached before the process is transferred to manual review, making the processing process traceable and reproducible; at the same time, the pre-inspection record, node quality inspection record and end quality inspection record are linked by hash to form a quality inspection evidence chain, which is convenient to verify whether the quality file has been tampered with.
[0066] (4) In cross-batch quality trend analysis, the present invention uses statistical methods such as sliding window to identify sensor performance drift and generates on-orbit parameter correction suggestions, forming a closed-loop feedback between ground quality inspection and satellite operation and control, and ground calibration. Attached Figure Description
[0067] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0068] Figure 1 This is a flowchart illustrating the full-process quality control method for an automated production line of remote sensing data according to the present invention. Detailed Implementation
[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0070] Please see Figure 1 This embodiment provides a method for full-process quality control of an automated production line for remote sensing data, including:
[0071] S1. Receive the remote sensing data to be processed and the metadata corresponding to the remote sensing data to be processed, assign a data identifier to the remote sensing data to be processed, and bind the remote sensing data to be processed, the metadata and the data identifier.
[0072] S2. Perform metadata pre-inspection on the obtained metadata. The metadata pre-inspection does not read the pixel content of the remote sensing data to be processed. Extract at least two pre-inspection features from the obtained metadata to form a pre-inspection feature vector. Match the obtained pre-inspection feature vector with the pre-inspection rule table to obtain the path identifier and the pre-inspection parameter set. Generate a pre-inspection record. The pre-inspection record includes the data identifier, the path identifier, the pre-inspection parameter set, the pre-inspection feature vector, and the pre-inspection timestamp.
[0073] The pre-detection feature vector in S2 includes at least two of the following: side view angle, solar altitude angle, sensor gain status flag, data packet integrity flag, and cloud cover estimate; the pre-detection rule table contains the correspondence between pre-detection feature conditions, path identifiers, and pre-detection parameter sets; if the obtained pre-detection feature vector matches multiple rules simultaneously, the path identifier and pre-detection parameter set corresponding to the rule with the highest priority are selected according to the priority order of the rules in the obtained pre-detection rule table; if the obtained pre-detection feature vector does not match any rule, the path identifier is set to the standard product line, and the pre-detection parameter set is set to the default parameter set; the value of the path identifier includes the standard product line, the downgraded product line, the radiation correction enhancement line, and the delay processing line.
[0074] The rule configuration for the pre-inspection rule table is as follows:
[0075] If the obtained side view is greater than the first threshold, the path is identified as a downgraded product line, and the screening threshold for the geometric correction control point of the pre-inspection parameter set is set to a lenient threshold.
[0076] If the obtained sensor gain status flag is an abnormal value, the path is identified as a radiation correction enhancement line, the initial value of the radiation correction coefficient in the pre-detection parameter set is set to the enhancement coefficient value, and a radiation correction verification node is inserted after the radiation correction node in the obtained processing node sequence.
[0077] If the estimated cloud cover is greater than the second threshold, the path is identified as a delayed processing line, and the cloud mask threshold of the pre-detection parameters is set to a strict threshold.
[0078] If the obtained side view is less than or equal to the first threshold, the obtained sensor gain status flag is normal, and the obtained cloud cover estimate is less than or equal to the second threshold, then the path identifier is determined to be the standard product line.
[0079] The priority order of the rules in the pre-inspection rule table is as follows: radiation correction enhancement line is higher than degraded product line, degraded product line is higher than delay processing line, and delay processing line is higher than standard product line.
[0080] S3. Determine the processing node sequence from the node library of the automated production line based on the obtained path identifier; write the parameters in the obtained pre-inspection parameter set that match the processing type of each processing node in the processing node sequence into the parameter register of the corresponding processing node; and make the upstream quality inspection record set received by the first processing node contain the obtained pre-inspection records.
[0081] The specific method for determining the processing node sequence from the node library of the automated production line based on the obtained path identifier in S3 includes:
[0082] If the obtained path identifier is a standard product line, the processing node sequence includes, in sequence, a radiation correction node, a geometric correction node, an orthorectification node, a fusion node, and a mosaic node.
[0083] If the obtained path is identified as a radiometric correction enhancement line, the processing node sequence includes, in sequence, a radiometric correction node, a radiometric correction verification node, a geometric correction node, an orthorectification node, a fusion node, and a mosaic node.
[0084] If the obtained path identifier is a downgraded product line, the processing node sequence includes, in sequence, a radiation correction node, a geometric correction node, and an orthophoto correction node;
[0085] If the obtained path identifier is a delayed processing line, the processing node sequence includes cloud mask node, radiation correction node, geometric correction node, orthorectification node, fusion node and mosaic node in sequence.
[0086] The cloud mask node performs a cloud mask operation on the input data according to the cloud mask threshold, and uses the data after the cloud mask operation as the input data of the radiation correction node;
[0087] The parameters in the pre-detection parameter set are written as follows: the initial value of the radiation correction coefficient is written to the parameter register of the radiation correction node; the screening threshold of the geometric correction control point, the DEM interpolation method, and the resampling method are written to the parameter registers of the geometric correction node and the orthorectification node; the fusion weight is written to the parameter register of the fusion node; and the cloud mask threshold is written to the parameter register of the cloud mask node.
[0088] S4. For each processing node in the obtained processing node sequence, initialize the reprocessing count of the corresponding processing node to 0. Each processing node receives the input data of the corresponding processing node and the upstream quality inspection record set. The input data of the corresponding processing node is the output data of the previous processing node, and the input data of the first processing node is the remote sensing data to be processed. Based on at least one of the quality inspection records of the preceding nodes in the upstream quality inspection record set and the parameters that match the processing node in the obtained pre-inspection parameter set, determine the activation parameters of the processing node. Process the input data of the processing node according to the node processing logic and the activation parameters to obtain the output data of the processing node. The node quality gate module corresponding to each processing node calculates the node quality inspection index of the corresponding processing node based on the node quality criterion and compares the obtained node quality inspection index with the corresponding node quality criterion.
[0089] Specifically, in step S4, the processing node selects and activates corresponding compensation parameters from the obtained pre-inspection parameter set based on the quality inspection records of the preceding nodes in the upstream quality inspection record set. The specific execution method includes:
[0090] If the deviation type of the preceding node is geometric deviation and the deviation magnitude is greater than the third threshold, then the processing node will enable the more stringent control point screening threshold in the obtained pre-detection parameter set.
[0091] If the deviation type of the preceding node is radiation deviation and the deviation magnitude is greater than the fourth threshold, then the processing node will enable the radiation correction coefficient fine-tuning amount in the obtained pre-detection parameter set and add the obtained radiation correction coefficient fine-tuning amount to the initial value of the radiation correction coefficient of the processing node.
[0092] If the deviation type of the preceding node is spectral deviation and the deviation magnitude is greater than the fifth threshold, then the processing node will use the fusion weight adjusted from the obtained pre-detection parameters and use the adjusted fusion weight as the fusion weight of the processing node.
[0093] If the deviation type of the preceding node is edge deviation and the deviation magnitude is greater than the sixth threshold, then the processing node will enable a more stringent resampling method in the obtained pre-detection parameter set, and resample the input data of the previous processing node according to the more stringent resampling method.
[0094] If the obtained node quality inspection indicators meet the corresponding node quality criteria, a node quality inspection record for the corresponding processing node is generated. The node quality inspection record includes at least the node quality inspection indicators and a pass flag. The obtained node quality inspection record is added to the upstream quality inspection record set of the corresponding processing node to obtain the downstream quality inspection record set of the corresponding processing node. The output data and downstream quality inspection record set of the corresponding processing node are passed to the next processing node, and the downstream quality inspection record set of the corresponding processing node becomes the upstream quality inspection record set of the next processing node. If the processing node is the last processing node in the obtained processing node sequence, the corresponding output data is used as a candidate product, and the obtained downstream quality inspection record set is passed to the end quality gate verification.
[0095] Specifically, the correspondence between the node quality inspection indicators obtained in S4 and the corresponding node quality criteria includes:
[0096] When the processing node is a radiometric correction node, the obtained node quality inspection index is the correlation coefficient of the pixel value sequence between each band after radiometric correction. The node quality criterion is the interval in which the correlation coefficient should fall. Satisfying the criterion means that the correlation coefficient falls into the interval.
[0097] Specifically, the node quality criterion at this time is the interval of inter-band correlation coefficients, denoted as [r]. min , r max The node quality inspection index is the Pearson correlation coefficient between each band after radiometric correction, denoted as r. Where ba and bb are pixel value sequences of two bands, Let ba and bb be the covariances. and Let ba and bb be the standard deviations of respectively, and let r satisfy the criterion that r falls within [r]. min , r max ].
[0098] When the processing node is a geometric correction node, the obtained node quality inspection index is the root mean square of the residual between the fitted coordinates and the measured coordinates of each control point after geometric correction. The node quality criterion is that the root mean square should not be greater than the first residual threshold. Satisfying the criterion means that the root mean square is not greater than the first residual threshold.
[0099] Specifically, the node quality criterion at this time is the RPC fitting residual threshold, denoted as ε. rpc The node quality inspection index is the root mean square of the RPC fitting residuals of the control points after geometric correction, denoted as RMSE. rpc ,
[0100]
[0101] in, To control the number of points, This represents the x-coordinate value in the fitted coordinates of the k-th control point; This represents the ordinate value in the fitted coordinates of the k-th control point. This represents the x-coordinate value in the measured coordinates of the k-th control point; Let y = the measured coordinate value of the k-th control point, satisfying the criterion. .
[0102] When the processing node is an orthorectification node, the obtained node quality inspection index is the root mean square of the residual between the corrected coordinates and the reference coordinates of each inspection point after orthorectification. The node quality criterion is that the root mean square should not be greater than the second residual threshold. Satisfying the criterion means that the root mean square is not greater than the second residual threshold.
[0103] Specifically, the node quality criterion at this time is the residual threshold of the corresponding points, denoted as ε. gcp The node quality inspection index is the plane error of the inspection point after orthorectification, denoted as RMSE. xy ,
[0104]
[0105] in, For the number of checkpoints, Let x be the x-coordinate value of the orthorectified coordinates of the k-th checkpoint. This represents the ordinate value in the orthorectified coordinate system of the k-th checkpoint. This is the x-coordinate value in the reference coordinates of the k-th checkpoint; The ordinate value is the reference coordinate of the k-th checkpoint; satisfying the criterion. .
[0106] When the processing node is a fusion node, the obtained node quality inspection index includes the spectral fidelity index and spatial sharpness index of the fused image. The node quality criterion includes a spectral threshold that the spectral fidelity index should not be greater than and a sharpness threshold that the spatial sharpness index should not be less than. Satisfying the criterion means that the spectral fidelity index is not greater than the spectral threshold and the spatial sharpness index is not less than the sharpness threshold.
[0107] Specifically, the node quality criterion at this time includes the spectral fidelity threshold ε. spec Spatial sharpness threshold ε spat The node quality inspection indicators include the spectral angle mapping value (SAM) and spatial frequency value (SF) of the fused image.
[0108]
[0109]
[0110] in, To fuse the multispectral band pixel values, The pixel values of the corresponding band after fusion. Let be the pixel value of the merged image at coordinates (x, y). Let (x-1, y) be the pixel value of the merged image at coordinates (x-1, y). Let be the pixel value of the merged image at coordinates (x, y-1). The number of rows in the image. Let SAM be the number of columns in the image, satisfying the criterion SAM≤ε. spec And SF≥ε spat .
[0111] When the processing node is a mosaic node, the obtained node quality inspection index is the difference in grayscale mean between the two images on both sides of the mosaic edge in the edge buffer. The node quality criterion is that the grayscale mean difference should not be greater than the edge grayscale difference threshold. Satisfying the criterion means that the grayscale mean difference is not greater than the edge grayscale difference threshold.
[0112] Specifically, the node quality criterion at this time is the edge grayscale difference threshold ε. seam The node quality inspection index is the grayscale mean difference Δg between the images on both sides of the mosaic edge, where Δg = |μ left -μ right |, where μ left μ represents the average grayscale value of the left-side image at the edge within the edge buffer. right Let Δg be the average grayscale value of the right-side image at the junction within the junction buffer, satisfying the criterion Δg≤ε. seam .
[0113] If the obtained node quality inspection index does not meet the corresponding node quality criteria, then the deviation type and deviation magnitude are determined based on the deviation between the obtained node quality inspection index and the corresponding node quality criteria; the obtained deviation type and deviation magnitude are used to query the parameter adjustment rule table corresponding to the processing node to obtain the correction parameter corresponding to the processing node; the parameter register of the processing node is updated according to the obtained correction parameter, the obtained correction parameter is recorded in the current failure parameter combination corresponding to the processing node, the reprocessing count of the processing node is incremented by 1, and step S4 is re-executed; if the reprocessing count of the processing node reaches the maximum reprocessing count and the obtained node quality inspection index still does not meet the corresponding node quality criteria, then a node-level non-conforming mark is generated, the output data of the processing node, the node-level non-conforming mark and the current failure parameter combination are transferred to the manual review queue, and the transmission to the next processing node is stopped;
[0114] The method for determining the deviation magnitude in S4 based on the deviation between the obtained node quality inspection index and the corresponding node quality criterion includes:
[0115] If the obtained node quality criterion is an upper limit criterion, then the obtained deviation range is the ratio of the difference between the obtained node quality inspection index and the corresponding threshold to the corresponding threshold; the upper limit criterion means that the node quality inspection index is not greater than the corresponding threshold and the criterion is satisfied.
[0116] If the obtained node quality criterion is a lower limit criterion, then the obtained deviation amplitude is the quotient of the difference between the corresponding threshold and the obtained node quality inspection index divided by the corresponding threshold; the lower limit criterion means that the node quality inspection index is not less than the corresponding threshold and thus meets the criterion.
[0117] If the obtained node quality criterion is an interval-type criterion, then the obtained deviation amplitude is the ratio of the deviation of the obtained node quality inspection index from the upper or lower limit of the corresponding interval to the interval width; the interval-type criterion means that the node quality inspection index falls into the corresponding interval and thus satisfies the criterion.
[0118] If the obtained node quality criteria include multiple sub-criteria, then calculate the sub-item deviation magnitude corresponding to each sub-criteria, and take the maximum value among the sub-item deviation magnitudes as the obtained deviation magnitude.
[0119] The parameter adjustment rule table includes the mapping relationship between deviation type, deviation amplitude range, and correction parameters; the deviation type includes radiometric deviation, geometric deviation, spectral deviation, and edge deviation; the deviation amplitude range includes at least a first amplitude range or a second amplitude range, where the deviation in the second amplitude range is greater than the deviation in the first amplitude range; the correction parameters include at least one of DEM interpolation method, control point screening threshold, resampling kernel function, radiometric calibration coefficient fine-tuning amount, and fusion weight adjustment amount; depending on the type of correction parameter, for method-type parameters, the corresponding parameter in the parameter register of the processing node is updated using a replacement method, and for numerical-type parameters, the corresponding parameter in the parameter register of the processing node is updated using an overlay method.
[0120] S5. Perform end-of-line quality gate verification on the obtained candidate products to obtain end-of-line quality inspection indicators and end-of-line quality inspection records. If the obtained end-of-line quality inspection indicators do not meet the end-of-line quality criteria, mark the obtained candidate products as end-of-line unqualified and transfer them to the manual review queue. If the obtained end-of-line quality inspection indicators meet the end-of-line quality criteria, construct a quality inspection evidence chain in sequence based on the obtained pre-inspection records, the node quality inspection records corresponding to each processing node, and the obtained end-of-line quality inspection records. Adjacent records in the quality inspection evidence chain are associated with each other through hash values. Output the obtained candidate products and the quality inspection evidence chain together.
[0121] In step S5, during the process of constructing a quality inspection evidence chain in sequence based on the obtained pre-inspection records, the node quality inspection records corresponding to each processing node, and the obtained end quality inspection records, the obtained pre-inspection records contain the previous hash and the current hash. The previous hash is a preset initial hash, and the current hash is generated based on the content of the obtained pre-inspection records excluding the current hash field and the previous hash. For the i-th node quality inspection record, the node quality inspection record contains the previous hash and the current hash. The previous hash in the i-th node quality inspection record is taken as the current hash of the last record in the upstream quality inspection record set when the node quality inspection record is generated. The current hash in the i-th node quality inspection record is generated based on the content of the node quality inspection record excluding the current hash field and the current hash in the (i-1)-th node quality inspection record.
[0122] For the obtained end-of-line quality inspection record, the current hash in the end-of-line quality inspection record is generated based on the content of the end-of-line quality inspection record excluding the current hash field and the previous hash. The previous hash in the end-of-line quality inspection record is taken as the current hash of the last record in the obtained downstream quality inspection record set.
[0123] The obtained pre-inspection records, the node quality inspection records of each processing node in the obtained processing node sequence, and the end quality inspection records are sequentially concatenated to obtain the corresponding quality inspection evidence chain. When verifying the obtained quality inspection evidence chain, the current hash of each record is recalculated according to the same hash generation method. If the current hash of each recalculated record is consistent with the current hash stored in the obtained quality inspection evidence chain, it is confirmed that the obtained quality inspection evidence chain has not been tampered with. If the current hash of a certain record is inconsistent with the current hash stored in the obtained quality inspection evidence chain, then that record and all subsequent records are marked as tampered with.
[0124] The method further includes a cross-batch quality trend analysis step, which is used to collect node quality inspection records, node quality inspection indicators, and end-of-batch quality inspection records of multiple batches of products of the same sensor; for each batch, radiation quality inspection index values and geometric quality inspection index values are extracted from the obtained node quality inspection indicators, and arranged in batch time order to form radiation index sequence and geometric index sequence; the sliding window mean method is used to perform trend analysis on the obtained radiation index sequence and geometric index sequence respectively, and calculate the mean and variance within the sliding window;
[0125] When the mean offset within the obtained sliding window exceeds the preset statistical control limit, it is determined that the same sensor has radiometric performance drift or geometric performance drift. When radiometric performance drift is determined, a radiometric calibration coefficient correction is generated based on the obtained mean offset and offset direction. When geometric performance drift is determined, an attitude control offset correction or orbit parameter correction is generated based on the obtained mean offset and offset direction. The obtained radiometric calibration coefficient correction or attitude control offset correction is encapsulated as an on-orbit parameter correction suggestion and pushed to the satellite operation and control system or ground calibration system through a standard interface. After the satellite operation and control system or ground calibration system performs parameter correction, the node quality inspection indicators and end quality inspection records of subsequent batches of products are continuously tracked. If the radiometric quality inspection indicator values or geometric quality inspection indicator values of subsequent batches return to within the statistical control limit, the correction is determined to be effective. Otherwise, a correction effect failure prompt is generated, and the width of the statistical control limit is reduced by a preset ratio to improve the sensitivity of subsequent trend analysis.
[0126] The method further includes setting a quality-efficiency trade-off coefficient, and adjusting the strictness of the obtained node quality criterion and the maximum number of reprocessing based on the obtained quality-efficiency trade-off coefficient; the value range of the quality-efficiency trade-off coefficient is [0,1]; when the obtained quality-efficiency trade-off coefficient increases, the node quality criterion becomes stricter and the maximum number of reprocessing increases; when the obtained quality-efficiency trade-off coefficient decreases, the node quality criterion becomes more lenient and the maximum number of reprocessing decreases, but the node quality criterion is not lower than the baseline pass standard;
[0127] The specific adjustment method for the stringency of the node quality criterion includes: for upper limit criterion, dividing the corresponding threshold by the adjustment factor to obtain the adjusted criterion; for lower limit criterion, multiplying the corresponding threshold by the adjustment factor to obtain the adjusted criterion; for interval criterion, keeping the interval center unchanged, dividing the interval half-width by the adjustment factor to obtain the adjusted interval upper and lower limits; the adjustment factor is equal to the sum of the difference between the obtained quality and efficiency trade-off coefficient and 0.5, multiplied by the preset adjustment sensitivity coefficient, and then added to 1. In this embodiment, the adjustment sensitivity coefficient is greater than 0 and less than 2; the quality and efficiency trade-off coefficient is determined based on at least one of the current production line's computing resource load status, product quality level requirements, and historical node quality inspection pass rate; when the product quality level requirement is high precision level, or the historical node quality inspection pass rate is lower than the pass rate threshold, the quality and efficiency trade-off coefficient is increased; when the computing resource load status is higher than the load threshold, the quality and efficiency trade-off coefficient is decreased.
[0128] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 process, method, article, or apparatus.
[0129] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for quality control of the whole process of an automatic production line of remote sensing data, characterized in that, include: S1. Receive the remote sensing data to be processed and the metadata corresponding to the remote sensing data to be processed, assign a data identifier to the remote sensing data to be processed, and bind the remote sensing data to be processed, the metadata and the data identifier. S2. Perform metadata pre-inspection on the obtained metadata. The metadata pre-inspection does not read the pixel content of the remote sensing data to be processed. Extract at least two pre-inspection features from the obtained metadata to form a pre-inspection feature vector. Match the obtained pre-inspection feature vector with the pre-inspection rule table to obtain the path identifier and the pre-inspection parameter set. Generate a pre-inspection record. The pre-inspection record includes the data identifier, the path identifier, the pre-inspection parameter set, the pre-inspection feature vector, and the pre-inspection timestamp. S3. Determine the processing node sequence from the node library of the automated production line based on the obtained path identifier; write the parameters that match the processing type of each processing node in the obtained pre-inspection parameter set into the parameter register of the corresponding processing node. Make the set of upstream quality inspection records received by the first processing node include the obtained pre-inspection records; S4. For each processing node in the obtained processing node sequence, initialize the reprocessing count of the corresponding processing node to 0. Each processing node receives the input data of the corresponding processing node and the upstream quality inspection record set. The input data of the corresponding processing node is the output data of the previous processing node, and the input data of the first processing node is the remote sensing data to be processed. Based on at least one of the quality inspection records of the preceding nodes in the upstream quality inspection record set and the parameters that match the processing node in the obtained pre-inspection parameter set, determine the activation parameters of the processing node. Process the input data of the processing node according to the node processing logic and the activation parameters to obtain the output data of the processing node. The node quality gate module corresponding to each processing node calculates the node quality inspection index of the corresponding processing node based on the node quality criterion and compares the obtained node quality inspection index with the corresponding node quality criterion. If the obtained node quality inspection indicators meet the corresponding node quality criteria, a node quality inspection record for the corresponding processing node is generated. The node quality inspection record includes at least the node quality inspection indicators and a pass flag. The obtained node quality inspection record is added to the upstream quality inspection record set of the corresponding processing node to obtain the downstream quality inspection record set of the corresponding processing node. The output data and downstream quality inspection record set of the corresponding processing node are passed to the next processing node, and the downstream quality inspection record set of the corresponding processing node becomes the upstream quality inspection record set of the next processing node. If the processing node is the last processing node in the obtained processing node sequence, the corresponding output data is used as a candidate product, and the obtained downstream quality inspection record set is passed to the end quality gate verification. If the obtained node quality inspection index does not meet the corresponding node quality criteria, then the deviation type and deviation magnitude are determined based on the deviation between the obtained node quality inspection index and the corresponding node quality criteria; the obtained deviation type and deviation magnitude are used to query the parameter adjustment rule table corresponding to the processing node to obtain the correction parameter corresponding to the processing node; the parameter register of the processing node is updated according to the obtained correction parameter, the obtained correction parameter is recorded in the current failure parameter combination corresponding to the processing node, the reprocessing count of the processing node is incremented by 1, and step S4 is re-executed; if the reprocessing count of the processing node reaches the maximum reprocessing count and the obtained node quality inspection index still does not meet the corresponding node quality criteria, then a node-level non-conforming mark is generated, the output data of the processing node, the node-level non-conforming mark and the current failure parameter combination are transferred to the manual review queue, and the transmission to the next processing node is stopped; S5. Perform end-of-line quality gate verification on the obtained candidate products to obtain end-of-line quality inspection indicators and end-of-line quality inspection records. If the obtained end-of-line quality inspection indicators do not meet the end-of-line quality criteria, mark the obtained candidate products as end-of-line unqualified and transfer them to the manual review queue. If the obtained end-of-line quality inspection indicators meet the end-of-line quality criteria, construct a quality inspection evidence chain in sequence based on the obtained pre-inspection records, the node quality inspection records corresponding to each processing node, and the obtained end-of-line quality inspection records. Adjacent records in the quality inspection evidence chain are associated with each other through hash values. Output the obtained candidate products and the quality inspection evidence chain together. 2.The method according to claim 1, wherein, The pre-detection feature vector in S2 includes at least two of the following: side view angle, solar altitude angle, sensor gain status flag, data packet integrity flag, and cloud cover estimate; the pre-detection rule table contains the correspondence between pre-detection feature conditions, path identifiers, and pre-detection parameter sets; if the obtained pre-detection feature vector matches multiple rules simultaneously, the path identifier and pre-detection parameter set corresponding to the rule with the highest priority are selected according to the priority order of the rules in the obtained pre-detection rule table; if the obtained pre-detection feature vector does not match any rule, the path identifier is set to the standard product line, and the pre-detection parameter set is set to the default parameter set; the value of the path identifier includes the standard product line, the downgraded product line, the radiation correction enhancement line, and the delay processing line.
3. The method according to claim 2, wherein, The rule configuration for the pre-inspection rule table is as follows: If the obtained side view is greater than the first threshold, the path is identified as a downgraded product line, and the screening threshold for the geometric correction control point of the pre-inspection parameter set is set to a lenient threshold. If the obtained sensor gain status flag is an abnormal value, the path is identified as a radiation correction enhancement line, the initial value of the radiation correction coefficient in the pre-detection parameter set is set to the enhancement coefficient value, and a radiation correction verification node is inserted after the radiation correction node in the obtained processing node sequence. If the estimated cloud cover is greater than the second threshold, the path is identified as a delayed processing line, and the cloud mask threshold of the pre-detection parameters is set to a strict threshold. If the obtained side view is less than or equal to the first threshold, the obtained sensor gain status flag is normal, and the obtained cloud cover estimate is less than or equal to the second threshold, then the path identifier is determined to be the standard product line. The priority order of the rules in the pre-inspection rule table is as follows: radiation correction enhancement line is higher than degraded product line, degraded product line is higher than delay processing line, and delay processing line is higher than standard product line.
4. The method according to claim 1, wherein, The specific method for determining the processing node sequence from the node library of the automated production line based on the obtained path identifier in S3 includes: If the obtained path identifier is a standard product line, the processing node sequence includes, in sequence, a radiation correction node, a geometric correction node, an orthorectification node, a fusion node, and a mosaic node. If the obtained path is identified as a radiometric correction enhancement line, the processing node sequence includes, in sequence, a radiometric correction node, a radiometric correction verification node, a geometric correction node, an orthorectification node, a fusion node, and a mosaic node. If the obtained path identifier is a downgraded product line, the processing node sequence includes, in sequence, a radiation correction node, a geometric correction node, and an orthophoto correction node; If the obtained path identifier is a delayed processing line, the processing node sequence includes cloud mask node, radiation correction node, geometric correction node, orthorectification node, fusion node and mosaic node in sequence. The cloud mask node performs a cloud mask operation on the input data according to the cloud mask threshold, and uses the data after the cloud mask operation as the input data of the radiation correction node; The parameters in the pre-detection parameter set are written as follows: the initial value of the radiation correction coefficient is written to the parameter register of the radiation correction node; the screening threshold of the geometric correction control point, the DEM interpolation method, and the resampling method are written to the parameter registers of the geometric correction node and the orthorectification node; the fusion weight is written to the parameter register of the fusion node; and the cloud mask threshold is written to the parameter register of the cloud mask node.
5. The method for full-process quality control of an automated production line for remote sensing data according to claim 1, characterized in that, In step S4, the processing node selects and activates corresponding compensation parameters from the obtained pre-inspection parameter set based on the quality inspection records of the preceding nodes in the upstream quality inspection record set. The specific execution method includes: If the deviation type of the preceding node is geometric deviation and the deviation magnitude is greater than the third threshold, then the processing node will enable the more stringent control point screening threshold in the obtained pre-detection parameter set. If the deviation type of the preceding node is radiation deviation and the deviation magnitude is greater than the fourth threshold, then the processing node will enable the radiation correction coefficient fine-tuning amount in the obtained pre-detection parameter set and add the obtained radiation correction coefficient fine-tuning amount to the initial value of the radiation correction coefficient of the processing node. If the deviation type of the preceding node is spectral deviation and the deviation magnitude is greater than the fifth threshold, then the processing node will use the fusion weight adjusted from the obtained pre-detection parameters and use the adjusted fusion weight as the fusion weight of the processing node. If the deviation type of the preceding node is edge deviation and the deviation magnitude is greater than the sixth threshold, then the processing node will enable a more stringent resampling method in the obtained pre-detection parameter set, and resample the input data of the previous processing node according to the more stringent resampling method.
6. The method for full-process quality control of an automated production line for remote sensing data according to claim 1, characterized in that, The correspondence between the node quality inspection indicators obtained in S4 and the corresponding node quality criteria includes: When the processing node is a radiometric correction node, the obtained node quality inspection index is the correlation coefficient of the pixel value sequence between each band after radiometric correction. The node quality criterion is the interval in which the correlation coefficient should fall. Satisfying the criterion means that the correlation coefficient falls into the interval. When the processing node is a geometric correction node, the obtained node quality inspection index is the root mean square of the residual between the fitted coordinates and the measured coordinates of each control point after geometric correction. The node quality criterion is that the root mean square should not be greater than the first residual threshold. Satisfying the criterion means that the root mean square is not greater than the first residual threshold. When the processing node is an orthorectification node, the obtained node quality inspection index is the root mean square of the residual between the corrected coordinates and the reference coordinates of each inspection point after orthorectification. The node quality criterion is that the root mean square should not be greater than the second residual threshold. Satisfying the criterion means that the root mean square is not greater than the second residual threshold. When the processing node is a fusion node, the obtained node quality inspection index includes the spectral fidelity index and spatial sharpness index of the fused image. The node quality criterion includes a spectral threshold that the spectral fidelity index should not be greater than and a sharpness threshold that the spatial sharpness index should not be less than. Satisfying the criterion means that the spectral fidelity index is not greater than the spectral threshold and the spatial sharpness index is not less than the sharpness threshold. When the processing node is a mosaic node, the obtained node quality inspection index is the difference in grayscale mean between the two images on both sides of the mosaic edge within the edge buffer. The node quality criterion is that the grayscale mean difference should not be greater than the edge grayscale difference threshold. Satisfying the criterion means that the grayscale mean difference is not greater than the edge grayscale difference threshold.
7. The method for full-process quality control of an automated production line for remote sensing data according to claim 1, characterized in that, The method for determining the deviation magnitude in S4 based on the deviation between the obtained node quality inspection index and the corresponding node quality criterion includes: If the obtained node quality criterion is an upper limit criterion, then the obtained deviation range is the ratio of the difference between the obtained node quality inspection index and the corresponding threshold to the corresponding threshold; the upper limit criterion means that the node quality inspection index is not greater than the corresponding threshold and the criterion is satisfied. If the obtained node quality criterion is a lower limit criterion, then the obtained deviation amplitude is the quotient of the difference between the corresponding threshold and the obtained node quality inspection index divided by the corresponding threshold; the lower limit criterion means that the node quality inspection index is not less than the corresponding threshold and thus meets the criterion. If the obtained node quality criterion is an interval-type criterion, then the obtained deviation amplitude is the ratio of the deviation of the obtained node quality inspection index from the upper or lower limit of the corresponding interval to the interval width; the interval-type criterion means that the node quality inspection index falls into the corresponding interval and thus satisfies the criterion. If the obtained node quality criteria include multiple sub-criteria, then calculate the sub-item deviation magnitude corresponding to each sub-criteria, and take the maximum value among the sub-item deviation magnitudes as the obtained deviation magnitude. The parameter adjustment rule table includes the mapping relationship between deviation type, deviation amplitude range, and correction parameters; the deviation type includes radiometric deviation, geometric deviation, spectral deviation, and edge deviation; the deviation amplitude range includes at least a first amplitude range or a second amplitude range, wherein the deviation in the second amplitude range is greater than the deviation in the first amplitude range; the correction parameters include at least one of DEM interpolation method, control point screening threshold, resampling kernel function, radiometric calibration coefficient fine-tuning amount, and fusion weight adjustment amount.
8. The method for full-process quality control of an automated production line for remote sensing data according to claim 1, characterized in that, In step S5, during the process of constructing a quality inspection evidence chain in sequence based on the obtained pre-inspection records, the node quality inspection records corresponding to each processing node, and the obtained end quality inspection records, the obtained pre-inspection records contain the previous hash and the current hash. The previous hash is a preset initial hash, and the current hash is generated based on the content of the obtained pre-inspection records excluding the current hash field and the previous hash. For the i-th node quality inspection record, the node quality inspection record contains the previous hash and the current hash. The previous hash in the i-th node quality inspection record is taken as the current hash of the last record in the upstream quality inspection record set when the node quality inspection record is generated. The current hash in the i-th node quality inspection record is generated based on the content of the node quality inspection record excluding the current hash field and the current hash in the (i-1)-th node quality inspection record. For the obtained end-of-line quality inspection record, the current hash in the end-of-line quality inspection record is generated based on the content of the end-of-line quality inspection record excluding the current hash field and the previous hash. The previous hash in the end-of-line quality inspection record is taken as the current hash of the last record in the obtained downstream quality inspection record set. The obtained pre-inspection records, the node quality inspection records of each processing node in the obtained processing node sequence, and the end quality inspection records are sequentially concatenated to obtain the corresponding quality inspection evidence chain. When verifying the obtained quality inspection evidence chain, the current hash of each record is recalculated according to the same hash generation method. If the current hash of each recalculated record is consistent with the current hash stored in the obtained quality inspection evidence chain, it is confirmed that the obtained quality inspection evidence chain has not been tampered with. If the current hash of a record obtained by recalculation is inconsistent with the current hash stored in the obtained quality inspection evidence chain, then that record and all subsequent records are marked as tampered with.
9. The method for full-process quality control of an automated production line for remote sensing data according to claim 1, characterized in that, The method further includes a cross-batch quality trend analysis step, which is used to collect node quality inspection records, node quality inspection indicators, and end-of-batch quality inspection records of multiple batches of products of the same sensor; for each batch, radiation quality inspection index values and geometric quality inspection index values are extracted from the obtained node quality inspection indicators, and arranged in batch time order to form radiation index sequence and geometric index sequence; the sliding window mean method is used to perform trend analysis on the obtained radiation index sequence and geometric index sequence respectively, and calculate the mean and variance within the sliding window; When the mean offset within the obtained sliding window exceeds the preset statistical control limit, it is determined that the same sensor has radiation performance drift or geometric performance drift. When radiation performance drift is detected, a radiation calibration coefficient correction is generated based on the obtained mean offset and offset direction; when geometric performance drift is detected, an attitude control offset correction or orbit parameter correction is generated based on the obtained mean offset and offset direction. The obtained radiation calibration coefficient correction or attitude control offset correction is encapsulated as an on-orbit parameter correction suggestion and pushed to the satellite operation and control system or ground calibration system through a standard interface. After the satellite operation and control system or ground calibration system performs parameter correction, the node quality inspection indicators and end quality inspection records of subsequent batches of products are continuously tracked. If the radiation quality inspection indicator value or geometric quality inspection indicator value of subsequent batches returns to within the statistical control limit, the correction is deemed effective. Otherwise, a correction effect failure prompt is generated, and the width of the statistical control limit is reduced by a preset ratio.
10. The method for full-process quality control of an automated production line for remote sensing data according to claim 1, characterized in that, The method further includes setting a quality-efficiency trade-off coefficient, and adjusting the strictness of the obtained node quality criterion and the maximum number of reprocessing based on the obtained quality-efficiency trade-off coefficient; the value range of the quality-efficiency trade-off coefficient is [0,1]; when the obtained quality-efficiency trade-off coefficient increases, the node quality criterion becomes stricter and the maximum number of reprocessing increases; when the obtained quality-efficiency trade-off coefficient decreases, the node quality criterion becomes more lenient and the maximum number of reprocessing decreases, but the node quality criterion is not lower than the baseline pass standard; The specific method for adjusting the stringency of the node quality criterion includes: for upper limit criterion, dividing the corresponding threshold by the adjustment factor to obtain the adjusted criterion; for lower limit criterion, multiplying the corresponding threshold by the adjustment factor to obtain the adjusted criterion; for interval criterion, keeping the interval center unchanged, dividing the interval half-width by the adjustment factor to obtain the adjusted upper and lower limits of the interval; the adjustment factor is equal to the sum of the difference between the obtained quality and efficiency trade-off coefficient and 0.5, multiplied by the preset adjustment sensitivity coefficient, and then added to 1; the quality and efficiency trade-off coefficient is determined based on at least one of the current production line's computational resource load status, product quality grade requirements, and historical node quality inspection pass rate.
