Change detection pattern spot optimization method, electronic equipment and computer product

Abstract

The invention discloses a change detection pattern spot optimization method, electronic equipment and a computer product, and the method comprises the steps: firstly obtaining the slice data of a change detection pattern spot result, including cutting out the position region of a pattern spot from a first time phase image, a second time phase image and a mask image; then, selectively labeling the slice data, selecting partial slice data labeling based on an active learning technology, and obtaining labeling information of the partial slice data; and inputting the slice data with the labeling information into the refined model, calculating the performance data of the slice data, judging whether the model reaches the standard or not, if not, training the model based on the slice data and the labeling information, and repeating the labeling selection and judgment steps until the model reaches the standard or labeling of all the slice data is completed. And finally, obtaining an optimized change detection pattern spot result based on the refined model, the slice data and the annotation information. According to the method, the false alarm rate can be reduced on the basis of high recall ratio, the checking workload is reduced, and specific services are optimized.

Description

Technical Field

[0001] The present invention relates to the fields of natural resource change monitoring business applications, artificial intelligence neural network models, transfer learning, active learning and other related technical fields, and in particular to a method, electronic equipment and computer product for optimizing existing change detection patterns to improve accuracy and reduce the workload of subsequent verification. Background Art

[0002] With the ever-changing demand for remote sensing applications, the rapid development of high-resolution remote sensing data, the accumulation of geographical knowledge, and the development of artificial intelligence technology, automated and intelligent natural resource monitoring has become possible. One of the key technical capabilities is intelligent natural resource change pattern detection.

[0003] In the prior art, some of the following patent documents may relate to this issue, such as CN118968338A, CN116051437A, CN117056856A, CN112990109A, etc.

[0004] Although many studies have shown that the most effective natural resource change detection method is based on deep neural networks, its unstable accuracy and weak generalization have greatly restricted its widespread application. The main reasons for this result are the diversity of ground features and the incompleteness of remote sensing data used for training.

[0005] The diversity of land features is reflected in the fact that the land features have complex morphological characteristics (for example, due to differences in the geographical environment of different agricultural regions, the shapes and sizes of cultivated land plots show great differences) and phenological characteristics (for example, the same plot carries a variety of crop types on the timeline, with complex growth forms and farming measures).

[0006] The remote sensing data used for training is not comprehensive enough. Due to the diverse scales and complex modalities of remote sensing data, which are easily affected by factors such as clouds, rain, fog, and sensors, the data used to train deep neural network models cannot cover all time periods, geographical regions, and all ranges of changes in the above influencing factors. Remote sensing data that only covers limited time, space, and spectral bands cannot represent the dynamic changes of natural objects with complex morphology and phenological characteristics, resulting in huge challenges in the full-scale, full-cycle, and full-dimensional cognition of natural resource objects.

[0007] In the existing intelligent change pattern detection technology, the deep neural network model is usually trained based on historical open source / proprietary data sets. The amount of data in the historical data set is always limited, covering only some regions, phenology, data characteristics, application scenarios / demands, etc., and cannot cover the various personalized situations that occur in the actual business process.

[0008] Therefore, the actual business performance of the existing intelligent change spot detection technology is poor, which is particularly reflected in the high false alarm rate while maintaining a high recall rate, and cannot effectively reduce the workload of subsequent spot verification. It is necessary to propose a solution that can optimize the change spot detection results output by the existing change detection system by using the actual data and verification process of the current specific business. Summary of the invention

[0009] The main purpose of the present invention is to provide a dual-phase change detection pattern optimization method, electronic equipment and computer product to solve the problems existing in the existing natural resource change detection technology, such as unstable accuracy, weak generalization, high false alarm rate and inability to effectively deal with personalized situations in actual business.

[0010] Based on the first main aspect of the present invention, a dual-phase change detection pattern optimization method is provided, which is characterized by comprising the following steps:

[0011] Step 1: Obtain slice data of the change detection spot result, including locating each spot determined to have changed, and cutting out a specific size area where the spot is located from the first phase image, the second phase image and the mask image to form slice data;

[0012] Step 2: selectively label the slice data to obtain labeling information of part of the slice data; including selecting a part of the slice data from all the unlabeled slice data for labeling, and obtaining information on whether the image spot area in the selected slice data has changed between two time phases;

[0013] Step 3: Based on the slice data with the annotated information, obtaining the performance data of the refined model, including taking the slice data as the input of the refined model, the model outputting the prediction result of whether the pattern area in the slice data changes, and calculating the performance data of the refined model based on the prediction result of the model on the slice data and the annotated information of the slice data;

[0014] Step 4: judging whether the refined model meets the requirements based on the performance data, if so, proceed to step 7, otherwise, proceed to step 5;

[0015] Step 5: training the refined model based on the slice data and the annotation information, including performing unsupervised training on the refined model using the slice data, and performing supervised training on the refined model using the slice data with the annotation information;

[0016] Step 6: Determine whether all slice data have been labeled. If all slice data have been labeled, go to step 7; otherwise, go to step 2.

[0017] Step 7: Based on the refined model, slice data and annotation information, an optimized change detection pattern result is obtained.

[0018] In some embodiments, as a further preferred scheme, the change detection spot result in step 1 includes the aligned first phase image, the second phase image and the mask image, and the mask image uses different values ​​to distinguish changed pixels and unchanged pixels, and each independent connected area formed by the changed pixels corresponds to a change spot.

[0019] In some embodiments, as a further preferred solution, when the spots are cut out to form slice data in step 1, each spot corresponds to a slice data, and each slice data includes a slice of the first phase image, a slice of the second phase image and a slice of the mask image.

[0020] In some embodiments, as a further preferred solution, the refined model in step 3 is a pre-trained deep neural network model, whose input is slice data, and whose output is a prediction of whether the pattern area in the input slice data changes.

[0021] In some embodiments, as a further preferred solution, the optimized change detection pattern result in step 7 includes:

[0022] 1) Based on the slice data with the annotation information, output all the change spot results marked as changes;

[0023] 2) Based on the refined model, all slice data without labeled information are predicted, and all change patch results predicted to be changes are output. The two parts of the output patches constitute the optimized change detection patch results.

[0024] In some embodiments, as a further preferred solution, in step 2, a portion of slice data is selected from all unlabeled slice data using active learning technology for labeling, including:

[0025] 1) All unlabeled slice data are sent to the refined model to obtain the predicted probability of whether the image area in all unlabeled slice data has changed;

[0026] 2) Obtaining the performance data of the current refined model, and in the first active learning iteration, the refined model has no performance data; in subsequent iterations, the performance data calculated in step 3 is taken;

[0027] 3) Using active learning methods, select a portion of slice data from all unlabeled slice data based on the information obtained in 1) and 2).

[0028] In some embodiments, as a further preferred solution, in the first active learning iteration, some unlabeled slice data are randomly selected; in subsequent iterations, let the number of unlabeled slice data be M, and the number of slice data to be selected be 3*N, then the method of selecting slice data is as follows:

[0029] 1) All unlabeled slice data are sorted in ascending order according to the entropy value of the predicted probability obtained by the refined model and the probability value of being judged as the positive class;

[0030] 2) Randomly select N slice data from the unlabeled slice data whose entropy value sorting position is in the interval [start,M], and determine the start value as follows: Let Fb represent a normalized performance index value of the refined model. First, let start be equal to the integer value of the product of Fb and M. Then, if M-start is less than the integer value B of the product of parameter a and N, then let start be equal to MB. Otherwise, keep the original value of start and set parameter a to 2.

[0031] 3) Select n1 slice data from the front end of the slice data sequence sorted by the positive class probability value, and select 2N-n1 slice data from the end of the sequence, where n1 is equal to the integer value of the product of R and 2N, and R is the proportion coefficient of the positive class slice data in the slice data annotated in the last active learning iteration, or in the accumulated annotated slice data.

[0032] Based on the second main aspect of the present invention, the present invention provides a method for monitoring changes in natural resources, including: obtaining change information of natural resources using change spots obtained by the aforementioned change detection spot optimization method.

[0033] Based on the third main aspect of the present invention, the present invention provides a method for providing a patch optimization service based on existing change detection patch results, including: using the aforementioned change detection patch optimization method to provide patch optimization services for natural resource monitoring business parties, the service including providing optimization results of existing patch results, providing a refined model that can be used to optimize existing patch results, embedding the aforementioned method of the present invention into the business party's patch verification workflow, etc.

[0034] Based on the fourth main aspect of the present invention, the present invention provides a device for optimizing a change detection pattern, comprising at least:

[0035] The slice acquisition module is configured to acquire slice data of the change detection spot result based on the existing change detection spot result.

[0036] The slice annotation module is configured to selectively annotate the slice data to obtain annotation information of part of the slice data.

[0037] The performance calculation module is configured to obtain performance data of the refined model based on the slice data with annotation information.

[0038] The model training module is configured to train a refined model based on the slice data and annotation information.

[0039] The result output module is configured to obtain an optimized change detection pattern result based on the refined model, the slice data and the annotation information.

[0040] In the above device, the functions of each module can be implemented by hardware, or by hardware executing corresponding software. The aforementioned hardware or software includes one or more modules corresponding to the above functions. In a possible design, the structure of the above device includes a memory and a processor, wherein the memory is used to store one or more computer instructions that support the above device to execute the above corresponding method, and the processor is configured to execute the computer instructions stored in the memory. The above device may also include a communication interface for the above device to communicate with other devices or communication networks.

[0041] Based on the fifth main aspect of the present invention, the present invention provides an electronic device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described in any one of the above aspects.

[0042] Based on the sixth main aspect of the present invention, the present invention provides a computer-readable storage medium for storing computer instructions used by any of the above-mentioned devices, and when the computer instructions are executed by a processor, they are used to implement the method described in any of the above-mentioned aspects.

[0043] According to a seventh main aspect of the present invention, the present invention provides a computer program product, which comprises computer instructions, and when the computer instructions are executed by a processor, they are used to implement the method described in any one of the above aspects.

[0044] Compared with the prior art, the present invention has the following advantages and effects:

[0045] (1) Compared with the prior art, the present invention can reduce the false alarm rate of the change spot results on the basis of a high recall rate. In order to ensure a high recall rate, the false alarm rate of the change spot results output by the prior art is very high. The present invention is optimized on the basis of the change spot results of the prior art, and the optimized output change spot results are composed of two parts: one is the spots that have been marked and confirmed to have changed. Because some erroneous spots are filtered out during the marking process, the false alarm rate is effectively reduced; the other is the spots that have not been marked and confirmed but are judged to be changed by the refined model. During the iterative training process, the refined model is required to meet performance requirements in terms of recall rate and false alarm rate. If the performance requirements are not met, it is not allowed to participate in the output of the final spot results. Therefore, the false alarm rate of some spots judged to have changed is also lower than that of the existing spot results.

[0046] (2) Compared with the prior art, the present invention can significantly reduce the workload of the business party in spot verification. The present invention filters out some erroneous spots in the prior art results and reduces the false alarm rate. Therefore, the number of clue spots that need to be verified by the business party will be significantly reduced, thereby significantly reducing the workload of verification.

[0047] (3) When training models, existing methods mainly use open source data and historical business data, and do not perform performance tuning for each specific business; while when training models, the present invention mainly uses the specific business data to be processed and performs performance tuning for the current specific business. This difference enables the present invention to optimize the output results of existing methods when applied to specific businesses, thereby adapting to various personalized change detection business tasks.

[0048] (4) The greater the number of identical / similar businesses, the smaller the average labeling workload required to implement the present invention. The refined model obtained by implementing the present invention in one business can be used as the initial refined model when implementing the present invention in subsequent identical / similar businesses. If the initial model at this time has met the performance requirements, there is no need to continue labeling the slice data; if the initial model at this time has not yet met the performance requirements, it is usually only necessary to label less data than the amount of data labeled in the previous business to meet the performance requirements. In addition, as the number of identical / similar businesses increases, the accumulated slice data and marking information also increase, and the performance of the initial refined model trained using these data also improves, thereby reducing the need to label new data.

[0049] (5) The model training process of the existing method is separated from the specific spot verification business process, and its model tuning process is not suitable to be placed in the specific spot verification business process, but the model training and tuning process of the present invention is suitable for combination with the specific spot verification business process. One reason is that the present invention is optimized for personalized services and needs to be combined with the specific spot verification business process. The second reason is that the labeling process of slice data in the present invention overlaps with the functions of the specific spot verification business process, that is, the specific spot verification business can be directly used to complete the labeling of slice data in the present invention; the third reason is that the present invention adopts an active learning iterative method to complete the model training, which minimizes the required labeling workload, which is consistent with the goal of the specific spot verification business to keep the verification workload as low as possible.

[0050] (6) The present invention can achieve win-win cooperation with existing change detection methods / systems. Business parties / users do not need to abandon or stop implementing their existing change detection methods / systems. The present invention optimizes the output image patch results of existing change detection methods / systems.

[0051] (7) In the implementation of the present invention, if a specific image spot verification business process is used to complete slice data annotation, some of the image spots that need to be verified are selected by the method of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of a change detection pattern optimization method in one embodiment of the present invention is shown;

[0053] Figure 2 A schematic diagram of an application scenario of a change detection spot optimization service in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0054] Hereinafter, the exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.

[0055] In the present disclosure, it should be understood that terms such as "include" or "have" are intended to indicate the presence of features, numbers, steps, behaviors, components, parts, or a combination thereof disclosed in the specification, and do not exclude the possibility that one or more other features, numbers, steps, behaviors, components, parts, or a combination thereof exist or are added.

[0056] It should also be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0057] The details of the embodiments of the present disclosure are described in detail below through specific examples.

[0058] Figure 1 FIG. 2 is a flow chart showing a change detection pattern optimization method according to an embodiment of the present disclosure. Figure 1 As shown, the change detection pattern optimization method includes the following steps:

[0059] Step 1: Obtain slice data of the change detection spot result. This includes locating each spot that is determined to have changed, and cutting out a specific size area where the spot is located from the first phase image, the second phase image, and the mask image to form slice data. Each spot corresponds to a slice data, and each slice data includes a slice of the first phase image, a slice of the second phase image, and a slice of the mask image.

[0060] In this embodiment, the change detection pattern result includes the registered first phase image, the second phase image and the mask image. In the mask image, different values ​​are used to distinguish changed pixels from unchanged pixels, and each independent connected area formed by the changed pixels corresponds to a change pattern. The change detection pattern result can be a change detection pattern result obtained by using an existing method and system for the current business data (first phase image data and second phase image data), or it can be a historical change detection pattern result.

[0061] Step 2: selectively label the slice data to obtain labeling information of some slice data, including selecting a portion of slice data from all unlabeled slice data for labeling, and obtaining information on whether the image spot area in the selected slice data has changed between two time phases.

[0062] In this embodiment, the process of selecting a portion of slice data from all unlabeled slice data using active learning technology includes:

[0063] (1) All unlabeled slice data are sent to the refined model to obtain the predicted probability of whether the image area in all unlabeled slice data has changed;

[0064] (2) Obtain the performance data of the current refined model (in the first active learning iteration, the refined model has no performance data; in subsequent iterations, the performance data calculated in step 3 is used).

[0065] (3) Using active learning methods, select a portion of slice data from all unlabeled slice data based on the information obtained in (1) and (2). Active learning methods can use all / part of the information in (1) and (2) or none of the information.

[0066] In this embodiment, the refined model is a pre-trained deep neural network model, whose input is slice data (including first phase image slices, second phase image slices and mask image slices), and whose output is a prediction of whether the pattern area in the input slice data changes.

[0067] In this embodiment, in the first active learning iteration, some unlabeled slice data are randomly selected; in subsequent iterations, let the number of unlabeled slice data be M, and the number of slice data to be selected be 3*N, then the method of selecting slice data is as follows:

[0068] (1) All unlabeled slice data are sorted in ascending order according to the entropy value of the predicted probability obtained by the refined model and the probability value of being judged as the positive class (i.e., changed);

[0069] (2) Randomly select N slice data from the unlabeled slice data whose entropy value sorting position is in the interval [start,M].

[0070] In this embodiment, the value of start is determined as follows: Fb represents a normalized performance index value of the refined model, and start is first set equal to the integer value of the product of Fb and M; then, if M-start is less than the integer value B of the product of parameter a and N, start is set equal to MB, otherwise the original value of start is maintained. In this embodiment, parameter a is set to 2, and in other embodiments a can be set to other values ​​within the range of [1, M / N];

[0071] (3) Select n1 slices from the front of the slice data sequence sorted by the positive probability value, and select 2*N-n1 slices from the end of the sequence. Where n1 is equal to the integer value of the product of R and 2*N, and R is the proportion coefficient of the positive slice data in the slice data annotated in the last active learning iteration or in the accumulated annotated slice data.

[0072] It should be pointed out that the slice data selection method in this embodiment tends to maintain a balance in the number of positive slice data and negative slice data to avoid the adverse consequences that an unbalanced data set may have on model training; at the same time, the better the performance of the refined model, the more the slice data selection method tends to select slice data that are difficult to distinguish (high entropy) to further develop the performance of the refined model; in addition, the specific implementation of the slice data selection method is highly adaptive because it is dynamically adjusted according to the proportion of positive categories in the labeled slice data / accumulated labeled slice data in the previous active learning iteration and the performance of the refined model.

[0073] Step 3: Based on the slice data with annotation information, the performance data of the refined model is obtained, which includes taking the slice data as the input of the refined model, the model outputting the prediction result of whether the image spot area in the slice data has changed, and calculating the performance data of the refined model based on the prediction result of the model on the slice data and the annotation information of the slice data.

[0074] In this embodiment, the performance data of the refined model is calculated, including calculating the training performance parameters of the refined model based on the slice data with labeled information that participates in the training of the refined model; and calculating the test performance parameters of the refined model based on the slice data with labeled information that does not participate in the training of the refined model.

[0075] In this embodiment, the performance parameters used are recall rate, precision rate and Fβ coefficient. Other performance parameters such as AUC etc. may also be used.

[0076] Step 4: Determine whether the refined model meets the requirements based on the performance data. If the requirements are met, go to step 7, otherwise, go to step 5.

[0077] Step 5: Based on the slice data and the annotation information, train the refined model. This includes using the slice data to perform unsupervised training on the refined model, and using the slice data with annotation information to perform supervised training on the refined model. Using data without annotation information to perform unsupervised training on the model and using data with annotation information to perform supervised training on the model are well-known technologies in the industry and will not be repeated here.

[0078] Step 6: Determine whether all slice data have been labeled. If all slice data have been labeled, go to step 7, otherwise, go to step 2.

[0079] Step 7: Based on the refined model, slice data and annotation information, an optimized change detection pattern result is obtained. This includes:

[0080] (1) Based on the slice data with the annotation information, output all the change spot results marked as changes;

[0081] (2) Based on the refined model, all slice data without labeled information are predicted, and all change pattern results predicted to be changes are output. The two parts of output patterns constitute the optimized change detection pattern results.

[0082] In some embodiments, the spot results in (1) can be used as the final output together with the spot results in other unlabeled slice data; or based on the refined model, all slice data are predicted, and then all change spots predicted to be changes are output as the final result.

[0083] In some embodiments, existing active learning methods (such as entropy-based sample selection methods) can be used to select a portion of samples from all unlabeled slice data for labeling. It is understandable that the embodiments of the present disclosure are not limited to the use of entropy-based sample selection methods, and any other sample selection methods can be used.

[0084] Based on this embodiment, a change detection image result for a specific business can be obtained. Compared with the prior art, the change detection image obtained by the method of this embodiment maintains a high recall rate and a lower false alarm rate. The image result can be used directly or after business verification to monitor and analyze natural resource changes.

[0085] Based on this embodiment, a neural network model (refined model) can be obtained that optimizes the existing change detection patch data for a specific business. This model can provide patch optimization services for natural resource change detection businesses.

[0086] The method and refined model training process of this embodiment can also be embedded in the natural resource change monitoring business process to achieve iterative mutual promotion between refined model-assisted manual verification and manual verification-assisted refined model optimization training.

[0087] A device for optimizing a change detection pattern based on an implementation of the present disclosure includes:

[0088] (1) A slice acquisition module is configured to acquire slice data of the change detection pattern result based on the existing change detection pattern data.

[0089] (2) A slice annotation module is configured to selectively annotate the slice data to obtain annotation information of part of the slice data.

[0090] (3) A performance calculation module configured to obtain performance data of a refined model based on the slice data with annotation information.

[0091] (4) A model training module, configured to train a refined model based on the slice data and annotation information.

[0092] (5) A result output module is configured to obtain an optimized change detection pattern result based on the refined model, the slice data and the annotation information.

[0093] The above functions are implemented by a server. The memory in the server is used to store one or more computer instructions that support the above-mentioned device to execute the above-mentioned corresponding method, and the processor of the server is configured to execute the computer instructions stored in the memory. The server is configured with a communication interface for communicating with other devices and communication networks. In other embodiments, the corresponding software implementation can also be executed by other hardware.

[0094] An electronic device based on an implementation of the present disclosure is a server, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the various steps of the above-mentioned change detection spot optimization method, including operations such as obtaining slice data, selective labeling, performance calculation, model training, and result output, thereby optimizing the change detection spot results. The server exchanges data with other related devices through its communication interface, such as receiving the original spot result data from the change detection server, and transmitting the optimized results to the spot business server / computer, etc.

[0095] A computer-readable storage medium based on an implementation method of the present disclosure is a hard disk, which is used to store computer instructions used in any aspect of the present invention. When the processor reads and executes these computer instructions in the hard disk, it can implement the above-mentioned change detection pattern optimization method and complete the optimization processing of the change detection pattern results, including operations such as slice data acquisition, labeling, model training and result generation, thereby improving the accuracy of change detection patterns, reducing false alarm rates, reducing subsequent verification workload, and adapting to personalized business needs.

[0096] A computer program product based on an implementation method of the present disclosure is a program product written in Python, which includes computer instructions. When the program product is run on a server or other device, and its computer instructions are executed by a processor, each step of the above-mentioned change detection pattern optimization method can be implemented. For example, slice data is obtained according to a set process, slice data is selected for annotation using active learning technology, model performance data is calculated and refined, the model is trained, and finally the optimized change detection pattern results are output, so as to achieve collaborative work with the entire change detection pattern optimization system and improve the efficiency and accuracy of natural resource change monitoring services.

[0097] Figure 2 A schematic diagram of an application scenario of a change detection spot optimization service according to an embodiment of the present disclosure is shown. Figure 2 As shown, the change detection server sends the change detection patch result data obtained based on the existing method to the refinement server, and the refinement server optimizes the existing change detection patch result data and transmits the optimization result to the patch business server / computer.

[0098] In the process of optimizing the existing change detection patch results, the refinement server transmits the slice data to the annotation server, which distributes the slice data to general Internet terminals and patch verification business terminals, obtains the annotation information of the slice data from each terminal, and then transmits the summarized annotation information back to the refinement server. The terminal device for annotating slices can be a desktop computer, laptop, tablet computer or mobile phone.

[0099] The matters not described in detail in the present invention are all known technologies to those skilled in the art.

[0100] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A dual-phase change detection pattern optimization method, characterized in that: The following steps are involved: Step 1: Obtain slice data of the change detection spot result, including locating each spot determined to have changed, and cutting out a specific size area where the spot is located from the first phase image, the second phase image and the mask image to form slice data; Step 2: selectively label the slice data to obtain labeling information of part of the slice data; The method includes selecting a portion of slice data from all unlabeled slice data for labeling, and obtaining information on whether a patch area in the selected slice data has changed between two time phases; Step 3: Based on the slice data with the annotated information, obtaining the performance data of the refined model, including taking the slice data as the input of the refined model, the model outputting the prediction result of whether the pattern area in the slice data changes, and calculating the performance data of the refined model based on the prediction result of the model on the slice data and the annotated information of the slice data; Step 4: judging whether the refined model meets the requirements based on the performance data, if so, proceed to step 7, otherwise, proceed to step 5; Step 5: training the refined model based on the slice data and the annotation information, including performing unsupervised training on the refined model using the slice data, and performing supervised training on the refined model using the slice data with the annotation information; Step 6: Determine whether all slice data have been labeled. If all slice data have been labeled, go to step 7; otherwise, go to step 2. Step 7: Based on the refined model, slice data and annotation information, an optimized change detection pattern result is obtained.

2. The dual-phase change detection pattern optimization method according to claim 1, characterized in that: The change detection spot result in step 1 includes the aligned first phase image, second phase image and mask image. The mask image uses different values ​​to distinguish changed pixels from unchanged pixels. Each independent connected area formed by changed pixels corresponds to a change spot.

3. The dual-phase change detection pattern optimization method according to claim 1, characterized in that: When the spots are cut out to form slice data in step 1, each spot corresponds to a slice data, and each slice data includes a slice of the first phase image, a slice of the second phase image and a slice of the mask image.

4. The dual-phase change detection pattern optimization method according to claim 1, characterized in that: The refined model described in step 3 is a pre-trained deep neural network model, whose input is slice data and whose output is a prediction of whether the image patch area in the input slice data changes.

5. The dual-phase change detection pattern optimization method according to claim 1, characterized in that: The optimized change detection patch results described in step 7 include: 1) Based on the slice data with the annotation information, output all the change spot results marked as changes; 2) Based on the refined model, all slice data without labeled information are predicted, and all change patch results predicted to be changes are output. The two parts of the output patches constitute the optimized change detection patch results.

6. The dual-phase change detection pattern optimization method according to claim 1, characterized in that: In step 2, a portion of slice data is selected from all unlabeled slice data using active learning technology for labeling, including: 1) All unlabeled slice data are sent to the refined model to obtain the predicted probability of whether the image area in all unlabeled slice data has changed; 2) Obtaining the performance data of the current refined model, and in the first active learning iteration, the refined model has no performance data; in subsequent iterations, the performance data calculated in step 3 is taken; 3) Using active learning methods, select a portion of slice data from all unlabeled slice data based on the information obtained in 1) and 2).

7. The dual-phase change detection pattern optimization method according to claim 6, characterized in that: In the first active learning iteration, some unlabeled slice data are randomly selected; in subsequent iterations, let the number of unlabeled slice data be M, and the number of slice data to be selected be 3*N, then the method of selecting slice data is as follows: 1) All unlabeled slice data are sorted in ascending order according to the entropy value of the predicted probability obtained by the refined model and the probability value of being judged as the positive class; 2) Randomly select N slice data from the unlabeled slice data whose entropy value sorting position is in the interval [start,M], and determine the start value as follows: Let Fb represent a normalized performance index value of the refined model. First, let start be equal to the integer value of the product of Fb and M. Then, if M-start is less than the integer value B of the product of parameter a and N, then let start be equal to MB. Otherwise, keep the original value of start and set parameter a to 2. 3) Select n1 slice data from the front end of the slice data sequence sorted by the positive class probability value, and select 2N-n1 slice data from the end of the sequence, where n1 is equal to the integer value of the product of R and 2N, and R is the proportion coefficient of the positive class slice data in the slice data annotated in the last active learning iteration, or in the accumulated annotated slice data.

8. A device for optimizing change detection patterns, characterized in that: Includes one or a combination of the following modules: A slice acquisition module is configured to acquire slice data of the change detection pattern result based on the existing change detection pattern result; A slice annotation module is configured to selectively annotate the slice data to obtain annotation information of part of the slice data; A performance calculation module is configured to obtain performance data of a refined model based on the slice data with the annotation information; A model training module is configured to train a refined model based on the slice data and the annotation information; The result output module is configured to obtain an optimized change detection pattern result based on the refined model, slice data and annotation information.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the dual-phase change detection pattern optimization method described in any one of claims 1-7.

10. A computer program product, comprising computer instructions, which are used to implement the dual-phase change detection pattern optimization method described in any one of claims 1 to 7 when executed by a processor.

Images