Algorithm scheme integration method, electronic equipment and computer readable storage medium

Abstract

The invention discloses an algorithm scheme integration method, electronic equipment and a computer readable storage medium, and the method comprises the steps: carrying out the function splitting of an obtained to-be-integrated scheme, and obtaining at least one atomic function; each atomic function is matched with an executable atomic function in an atomic function database to obtain a matching result, and the atomic function database comprises at least one executable atomic function and an executable operator required for executing each executable atomic function; and in response to the matching result, representing that the target executable atomic functions matched with the atomic functions exist in the atomic function database, and merging executable operators required for executing the target executable atomic functions to obtain a target integration scheme. Therefore, the algorithm scheme integration efficiency can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to an algorithm scheme integration method, an electronic device and a computer readable storage medium. BACKGROUND

[0002] Under the tide of intelligence, artificial intelligence algorithms gradually affect people's production and life. On this basis, due to the changing scenes of production and life, the demand for algorithms also changes. Even if many functional algorithms have been developed, it cannot be guaranteed that the demand is fully covered.

[0003] Especially in the face of a variety of fragmented needs from the demand side, it is necessary to customize a scheme package containing different functions to meet the requirements of the demand side, resulting in waste of human cost and resources.

[0004] Therefore, an algorithm scheme integration method is urgently needed to cope with a variety of needs. SUMMARY

[0005] The technical problem solved by the present application is to provide an algorithm scheme integration method, an electronic device and a computer readable storage medium, which can improve the algorithm scheme integration efficiency.

[0006] To solve the above technical problems, one technical solution adopted by the present application is to provide an algorithm scheme integration method, which includes: performing function splitting on the obtained to-be-integrated scheme to obtain at least one atomic function; performing matching processing on each atomic function and the executable atomic functions in an atomic function database to obtain a matching result, the atomic function database including at least one executable atomic function and executable operators required for executing each executable atomic function; in response to the matching result representing that there is a target executable atomic function matching each atomic function in the atomic function database, performing merging processing on the executable operators required for executing the target executable atomic function to obtain a target integrated scheme.

[0007] To solve the above technical problems, another technical solution adopted by the present application is to provide an electronic device including a memory and a processor, the memory storing program instructions, and the processor fetching the program instructions from the memory to execute the above-mentioned algorithm scheme integration method.

[0008] To solve the above technical problems, another technical solution adopted by the present application is to provide a computer readable storage medium including stored program data, the program data being executed by a processor to implement the above-mentioned algorithm scheme integration method.

[0009] Differing from the prior art, the algorithm scheme integration method of the application performs function splitting on the obtained to-be-integrated scheme to obtain at least one atomic function; performs matching processing on each atomic function and the executable atomic functions in the atomic function database to obtain a matching result, the atomic function database including at least one executable atomic function and an executable operator required for executing each executable atomic function; in response to the matching result representing that there is a target executable atomic function matching each atomic function in the atomic function database, performing merging processing on the executable operator required for executing the target executable atomic function to obtain a target integrated scheme. Thus, after obtaining the to-be-integrated scheme, the required to-be-integrated scheme can be directly integrated through the atomic function database without repeatedly constructing an algorithm flow, which can effectively reduce the workload of repeated development and improve the algorithm scheme integration efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0011] Figure 1 is a flowchart of an exemplary embodiment of the algorithm scheme integration method shown by the application;

[0012] Figure 2 is a schematic diagram of an exemplary embodiment of the integration page shown by the application;

[0013] Figure 3 is Figure 1 is a flowchart of an exemplary embodiment of step S130 in the algorithm scheme integration method shown by the application;

[0014] Figure 4 is a schematic diagram of an exemplary embodiment of the merging process shown by the application;

[0015] Figure 5 is Figure 1 is a specific flowchart of an exemplary embodiment of step S130 in the algorithm scheme integration method shown by the application;

[0016] Figure 6 is Figure 1 is a flowchart of an exemplary embodiment after step S120 in the algorithm scheme integration method shown by the application;

[0017] Figure 7 is a framework schematic diagram of an exemplary embodiment of the preset model storage module shown by the application;

[0018] Figure 8 is a framework schematic diagram of an exemplary embodiment of a preset operator storage module shown in the present application;

[0019] Figure 9 is a framework schematic diagram of an exemplary embodiment of a preset function storage module shown in the present application;

[0020] Figure 10 is a framework schematic diagram of an exemplary embodiment of an algorithm scheme integration method shown in the present application;

[0021] Figure 11 is a structural schematic diagram of an exemplary embodiment of an algorithm scheme integration device shown in the present application;

[0022] Figure 12 is a structural schematic diagram of an embodiment of an electronic device provided by the present application;

[0023] Figure 13 is a structural schematic diagram of an embodiment of a computer readable storage medium provided by the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It can be understood that the specific embodiments described herein are only used to explain the present application, rather than limit the present application. In addition, it should be noted that, for the convenience of description, only parts related to the present application are shown in the drawings, rather than all structures. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0025] First of all, it should be noted that artificial intelligence technology has played an important role in various industries, especially for behavior analysis algorithms in the field of computer vision, which are actually deployed in different scenarios, including energy, education, hospitals, banks and factories, etc. From the algorithm demand side, on the one hand, different industries or different scenarios in the same industry require different behaviors to be analyzed, resulting in differences in the algorithm schemes to be delivered; on the other hand, the same behavior analysis function is used in many different scenarios. For the algorithm providing side, due to the limitations of hardware storage, computing power and real-time performance, it is difficult to develop a set of general and complete algorithm scheme to solve all the needs of the scene, so when facing many different algorithm needs, how to achieve rapid and efficient delivery is a big problem.

[0026] Based on this, the present application proposes an algorithm scheme integration method, an electronic device and a computer readable storage medium, which realizes the rapid integration of the to-be-integrated scheme through an atomic function database. For details, please refer to Figure 1 , Figure 1is a flowchart of an exemplary embodiment of the algorithm scheme integration method shown in the present application.

[0027] The execution subject of the algorithm scheme integration method can be a terminal device or a server or other processing device, wherein the terminal device can be a user equipment (UE), a computer, a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. The execution subject of the algorithm scheme integration method can also be an algorithm scheme integration apparatus. In some possible implementation manners, the algorithm scheme integration method can be realized by a processor invoking computer readable instructions stored in a memory.

[0028] Specifically, the algorithm scheme integration method of the embodiment includes the following steps:

[0029] S110: Functionally splitting the obtained to-be-integrated scheme to obtain at least one atomic function.

[0030] The to-be-integrated scheme can refer to a demand scheme to be developed, which can include multiple data processing demands that are associated with or independent of each other. Exemplarily, the demand provided by the demand side can be directly determined as the to-be-integrated scheme, or the demand provided by the demand side can be preprocessed to obtain the to-be-integrated scheme. The preprocessing includes process combing, logical relationship combing, etc. For example, the to-be-integrated scheme can be traffic violation behavior detection, abnormal behavior detection in a factory, etc.

[0031] The atomic function is a small executable module divided from the to-be-integrated scheme. Exemplarily, the to-be-integrated scheme can be divided into several atomic functions according to its logical structure and data characteristics. For example, if the to-be-integrated scheme is traffic violation behavior detection, the split atomic functions can include but are not limited to phone call detection, seat belt detection, red light running detection, etc. If the to-be-integrated scheme is abnormal behavior detection in a factory, the split atomic functions can include but are not limited to sleeping on the job detection, dressing detection, abnormal target detection, etc. Here, only a part of the atomic functions of the to-be-integrated scheme is exemplified, and the actual application also needs to be determined according to the detection demand of the demand side.

[0032] Of course, in another application scenario, if the obtained to-be-integrated scheme is already a single atomic function, it can be understood that the demand side only requires to analyze one atomic function, and then the to-be-integrated scheme can not be functionally split, and the to-be-integrated scheme can be directly taken as an atomic function.

[0033] In some embodiments, the atomic function database includes a plurality of integrated solutions, and the to-be-integrated solution is matched with each integrated solution. If a match is found, the matched integrated solution is determined as the target integrated solution. If no match is found, the to-be-integrated solution is functionally split, and the split atomic functions are matched to obtain the final target integrated solution.

[0034] In S120, each atomic function is matched with an executable atomic function in the atomic function database to obtain a matching result. The atomic function database includes at least one executable atomic function and an executable operator required for executing each executable atomic function.

[0035] The atomic function database is configured to store executable atomic functions and executable operators. In some embodiments, each integrated solution that has been developed is split to obtain executable atomic functions and executable operators, and the executable atomic functions and the executable operators are stored in the atomic function database. In some other embodiments, developers can construct different executable operators and executable atomic functions, and the executable atomic functions and the executable operators are stored in the atomic function database. In some other embodiments, the atomic function database is further configured to store models. The models can be split from the developed integrated solutions or constructed by developers. The atomic function database can include a model storage module, an operator storage module, and a function storage module. The model storage module is configured to store each model in a classified manner, the operator storage module is configured to store each operator in a classified manner, and the function storage module is configured to store each function in a classified manner.

[0036] An executable atomic function is a relatively independent executable module configured to execute a relatively independent task. For example, the executable atomic function can be a single task such as phone call detection, seat belt detection, red light running detection, and fall detection.

[0037] An executable operator is a smaller executable unit than an executable atomic function. Generally, one executable atomic function can be composed of one or more executable operators. For example, the executable operator can be target detection, target tracking, target classification, etc., and the phone call detection can be composed of related executable operators such as target detection, target tracking, target classification, and logic processing. The executable operator includes at least one operator library and / or a relationship mapping file. The operator library can be regarded as a code module configured to implement a specific operation logic. The relationship mapping file is configured to store a mapping relationship between the operator library and the model, and exists only when the model association attribute of the executable operator is model related.

[0038] The matching result represents whether there is a target executable atomic function in the atomic function database that matches the atomic function of the to-be-integrated scheme. In some embodiments, the semantic information of each atomic function and the executable atomic function in the atomic function database can be matched to obtain the matching result between each atomic function and the executable atomic function. In some embodiments, the features of each atomic function and the executable atomic function in the atomic function database can also be calculated to obtain the matching result between each atomic function and the executable atomic function.

[0039] S130: In response to the matching result representing that there is a target executable atomic function in the atomic function database that matches each atomic function, executable operators required to execute the target executable atomic function are combined to obtain a target integration scheme.

[0040] The target executable atomic function can be a function that matches the atomic function in the to-be-integrated scheme. For example, when the matching result between the atomic function and the executable atomic function meets a preset requirement, the corresponding executable atomic function is determined as the target executable atomic function. For example, the atomic function split in the to-be-integrated scheme includes phone call detection, and if the executable atomic function in the atomic function database includes phone call detection or a function with the same meaning, the corresponding executable atomic function is determined as the target executable atomic function.

[0041] The target integration scheme can be a scheme package integrating one or more atomic functions. For example, after selecting the target executable atomic function corresponding to the to-be-integrated scheme, the executable operators in each target executable atomic function are combined to obtain the target integration scheme. Specifically, the algorithm scheme integration device includes an atomic function database and an integration page. The integration page is used to output the complete target integration scheme and the work page of the requirement delivery to the outside. In some embodiments, in response to receiving the input to-be-integrated scheme, the to-be-integrated scheme is functionally split to obtain at least one atomic function. Each atomic function is matched with the executable atomic function in the atomic function database, and the matched executable atomic function is determined as the target executable atomic function.

[0042] In some embodiments, each executable atomic function in the atomic function database is displayed in the integration page. The executable atomic function that has been input in the integration page can be checked, and the checked executable atomic function is determined as the target executable atomic function. For details, please refer to Figure 2 , Figure 2is a schematic diagram of an exemplary embodiment of the integrated page shown in the present application. In the first preset area of the integrated page, each executable atomic function is displayed, and in the second preset area of the integrated page, the to-be-integrated scheme is displayed. There can be multiple to-be-integrated schemes, and corresponding executable atomic functions are selected for different to-be-integrated schemes, and each to-be-integrated scheme corresponding executable atomic function is integrated to obtain the corresponding target integrated scheme.

[0043] It can be seen that the algorithm scheme integration method of the embodiment of the present application performs function splitting on the obtained to-be-integrated scheme to obtain at least one atomic function; each atomic function is matched with the executable atomic function in the atomic function database to obtain a matching result, the atomic function database includes at least one executable atomic function and an executable operator required to execute each executable atomic function; in response to the matching result representing that there is a target executable atomic function matched with each atomic function in the atomic function database, the executable operator required to execute the target executable atomic function is merged to obtain the target integrated scheme. Thus, after obtaining the to-be-integrated scheme, it is not necessary to repeatedly construct the algorithm flow, and the required to-be-integrated scheme can be directly integrated through the atomic function database, which can effectively reduce the workload of repeated development and improve the algorithm scheme integration efficiency.

[0044] On the basis of the above-mentioned embodiments, the embodiment of the present application adopts Figure 3 The flow chart describes in detail how to merge the executable operators required by each target executable atomic function to obtain the target integrated scheme. Please refer to Figure 3 , Figure 3 is Figure 1 a flowchart of an exemplary embodiment of step S130 in the algorithm scheme integration method shown in the present application. Specifically, the process of step S130 merging the executable operators required to execute the target executable atomic function to obtain the target integrated scheme includes the following steps:

[0045] S310: According to the execution order of each executable operator in each target executable atomic function, the executable operators required by different target executable atomic functions are combined to obtain at least one executable operator group.

[0046] The execution order refers to the processing flow of each executable operator determined for each target executable atomic function. Exemplarily, the execution order of each executable operator can be determined according to the execution logic of each target executable atomic function. For example, if the target executable atomic function is mobile phone playing detection, the executable operators required include target detection operator, target tracking operator, target classification operator and logic processing operator, and the execution order of each executable operator is determined according to the execution logic of the target executable atomic function in the order of target detection operator, target tracking operator, target classification operator and logic processing operator.

[0047] The executable operator group is a combination of executable operators in different target executable atomic functions according to different target executable atomic functions. In some embodiments, executable operators in the same execution order in different target executable atomic functions can be combined to obtain an executable operator group. For example, executable operators in the Nth execution order are combined to obtain an executable operator group. In other embodiments, executable operators in the same execution order in different target executable atomic functions can also be combined two by two to obtain an executable operator group composed of two executable operators.

[0048] The algorithm scheme integration device obtains the execution order of each executable operator in each target executable atomic function, and combines each executable operator in the same execution order to obtain an executable operator group composed of executable operators in the same execution order.

[0049] In other embodiments, each executable operator in a target executable atomic function can also be traversed according to the execution order. For the currently traversed executable operator, the currently traversed executable operator is combined with each executable operator in other target executable atomic functions to form an executable operator group. For example, the first target executable atomic function includes three executable operators, and the second target executable atomic function includes four executable operators. The executable operator in the first execution order in the first target executable atomic function is first traversed, and the executable operator in the first execution order is combined with the four executable operators in the second target executable atomic function to obtain four executable operator groups. The executable operator in the second execution order in the second target executable atomic function is then traversed, and the executable operator in the second execution order is combined with the four executable operators in the second target executable atomic function to obtain four executable operator groups. Finally, eight executable operator groups are obtained.

[0050] In other embodiments, the same executable operators in each target executable atomic function can also be obtained, and the same executable operators are combined to obtain an executable operator group. Then, whether the same executable operators can be merged is determined according to the execution order and execution logic of each executable operator. For example, the executable operator in the first execution order in the first target executable atomic function is the same as the executable operator in the second execution order in the second target executable atomic function. At this time, it is necessary to determine whether the function required to be implemented can be achieved by first processing the to-be-detected object through the executable operator in the first execution order in the second target executable atomic function and then processing the to-be-detected object through the same executable operator. If so, the executable operator in the first execution order in the first target executable atomic function and the executable operator in the second execution order in the second target executable atomic function are merged.

[0051] In some other embodiments, the user can also select executable operators in different target executable atomic functions to form an executable operator group.

[0052] S320: Merge the executable operators in each executable operator group according to the model correlation attribute of the executable operators, to obtain each merged executable operator group.

[0053] The model correlation attribute includes model-dependent and model-independent. Model-dependent means that the executable operator needs to load a model to implement the function to be implemented. Model-independent means that the executable operator does not need to load a model to implement the function to be implemented. Illustratively, the mapping relationship between the model and the executable operator can be pre-set to determine the model correlation attribute of the executable operator. Generally, the executable operator that needs to use machine learning technology can be set as model-dependent to load the corresponding model to implement the corresponding function; the executable operator that does not need to use machine learning technology can be set as model-independent, without the need to load a model to implement the corresponding function.

[0054] The merging strategy of each executable operator group is different due to the difference in the model correlation attribute. Illustratively, the model-dependent executable operator is determined as a first type of executable operator, and the model-independent executable operator is determined as a second type of executable operator; in response to each executable operator in the executable operator group being a first type of executable operator, the merging strategy of the executable operator group is determined according to the consistency of the operator information of each first type of executable operator and the corresponding model, and the first type of executable operator is merged according to the merging strategy to obtain the merged executable operator group; in response to each executable operator in the executable operator group being a second type of executable operator, the merging strategy of the executable operator is determined according to the consistency of the operator information of each second type of executable operator, and each second type of executable operator is merged according to the merging strategy to obtain the merged executable operator group. In this way, a more reasonable and adaptive merging strategy can be obtained, and the probability of merging errors of the executable operators can be reduced.

[0055] In this embodiment, the model-dependent executable operator is determined as a first type of executable operator, and the model-independent executable operator is determined as a second type of executable operator. It should be noted that this is not a limitation on the executable operator, but is only for the convenience of distinguishing and description. The model-dependent executable operator can also be determined as a second type of executable operator, and the model-independent executable operator can be determined as a first type of executable operator, and the like. The following is described by taking the model-dependent executable operator as a first type of executable operator and the model-independent executable operator as a second type of executable operator as an example.

[0056] When each executable operator in the executable operator group is a first type executable operator, a merging strategy of the executable operator group needs to be determined according to consistency of operator information corresponding to each first type executable operator and a model. The operator information includes, but is not limited to, functional attributes, operator libraries, configuration files and the like of each executable operator. The model refers to a model required to be loaded by the executable operator, which can be various end-to-end neural network models, including but not limited to basic tasks such as object detection, object tracking, object classification and object segmentation.

[0057] In some embodiments, when the operator information and the model of each executable operator in the executable operator group are consistent, each executable operator in the executable operator group is subjected to fusion processing to obtain a merged executable operator group. The fusion processing refers to fusing multiple executable operators into one operator. When the operator information and the model of each executable operator in the executable operator group are inconsistent, each executable operator in the executable operator group is subjected to parallel processing to obtain a merged executable operator group. The parallel processing refers to parallel display of multiple executable operators, each of which performs a corresponding task according to the original structure. As can be seen, the merging processing can include two processing modes, and the fusion processing is performed for those that can be fused, and the parallel display is performed for those that cannot be fused.

[0058] In other embodiments, the operator information includes functional attributes and operator libraries. The functional attributes refer to tasks to be performed by the executable operator group, which can be determined according to the functions that can be implemented by the executable operator group. For example, the functional attributes can be one of the following: object detection, object classification, object tracking, object filtering, object segmentation, image caching, logical processing and the like, which are single functions irrelevant to business. The operator library is a programming language written for constructing the executable operator, for example, a high-level programming language such as C language, C++ language, Java or the like can be used to write the operator library of the executable operator.

[0059] The functional attributes, the operator library and the model corresponding to the executable operator determine whether the executable operators can be merged. If the executable operators can be merged, the merging strategy of each executable operator is determined as fusion processing, and if the executable operators cannot be merged, the merging strategy of each executable operator is determined as parallel processing.

[0060] Specifically, in response to the function attribute, the operator library and the model corresponding to each first-type executable operator being consistent, each first-type executable operator in the executable operator group is fused to obtain a merged executable operator group; in response to the function attribute and the operator library corresponding to each first-type executable operator being consistent, and the model being inconsistent, a target model is selected from the model corresponding to each first-type executable operator, and each first-type executable operator in each pair of executable operator groups is fused based on the target model to obtain a merged executable operator group; in response to the function attribute and the model corresponding to each first-type executable operator being consistent, the operator library being inconsistent and the operator library being compatible, a target operator library is selected from the operator library corresponding to each first-type executable operator, and each first-type executable operator in each pair of executable operator groups is fused based on the target operator library to obtain a merged executable operator group. Thus, whether the multiple executable operators can be fused is determined through the function attribute, the operator library and the model, the accuracy of fusion is ensured, and the reliability of the target integration scheme is ensured.

[0061] If the function attribute, the operator library and the model corresponding to each first-type executable operator in the executable operator group are consistent, that is, the multiple first-type executable operators are the same, the multiple executable operators can be naturally fused into one, and the fusion manner can be to select one executable operator as the merged executable operator group.

[0062] If the function attributes, the operator libraries and the models corresponding to the first-type executable operators are inconsistent, compatibility needs to be further considered to determine whether the first-type executable operators can be fused. The following describes various cases: when the function attributes and the operator libraries of the first-type executable operators are the same, but the models are different, a target model can be selected from the models corresponding to the first-type executable operators by a user, the target model is taken as the model corresponding to the first-type executable operators, and the first-type executable operators are fused based on the target model to obtain the executable operator group after the fusion processing. When the function attributes and the models of the first-type executable operators are consistent, but the operator libraries are inconsistent, it can be determined whether the operator libraries are compatible, that is, the operator libraries of the first-type executable operators can be compatible with each other. If compatible, a target operator library is selected from the operator libraries corresponding to the first-type executable operators, the target operator library is taken as the operator library of the first-type executable operators in the executable operator group, and the first-type executable operators are fused based on the target operator library to obtain the executable operator group after the fusion processing. If incompatible, the first-type executable operators cannot be fused and are displayed in parallel. When the function attributes corresponding to the first-type executable operators are consistent, but the operator libraries and the models are inconsistent, it indicates that the first-type executable operators in the executable operator group cannot be fused and are displayed in parallel. When the function attributes corresponding to the first-type executable operators are inconsistent, it indicates that the first-type executable operators in the executable operator group cannot be fused and are displayed in parallel.

[0063] When the executable operators in the executable operator group are all second-type executable operators, the merging strategy of the executable operator group needs to be determined according to the consistency of the operator information corresponding to the second-type executable operators. Specifically, the merging strategy of the executable operator group can be determined according to the consistency of the function attributes and the operator libraries of the second-type executable operators in the executable operator group. When the function attributes and the operator libraries of the second-type executable operators are consistent, the second-type executable operators can be fused; when the function attributes of the second-type executable operators are consistent but the operator libraries are inconsistent, whether the second-type executable operators can be fused can be determined according to the compatibility of the operator libraries.

[0064] When the executable operators in the executable operator group include both first-type executable operators and second-type executable operators, the executable operators in the executable operator group cannot be fused but are merged in a parallel display manner.

[0065] As an example, refer to Figure 4The first executable atomic function includes a target detection operator 1 (OD_1), a target tracking operator 1 (OT_1), a target classification operator 1 (OC_1), and a logic processing operator 1 (LF_1) connected in series. The second executable atomic function includes a target detection operator 2 (OD_2), a target tracking operator 1 (OT_1), a target classification operator 2 (OC_2), and a logic processing operator 2 (LF_2) connected in series.

[0066] Among them, the association model of OD_1 is object detection model 1 (OD_model_1), the association model of OC_1 is object classification model 1 (OC_model_1), the association model of OD_2 is object detection model 1 (OD_model_1), and the association model of OC_2 is object classification model 2 (OC_model_2).

[0067] The first executable atomic function OD_1 and the second executable atomic function OD_2 are taken as the first executable operator group. The functional attributes and models of the two executable operators are consistent, and the operator libraries are compatible. At this time, the operator library of OD_2 can be selected as the target operator library. The first executable operator group is fused based on the target operator library to obtain the merged first executable operator group OD_2.

[0068] The first executable atomic function OT_1 and the second executable atomic function OT_1 are used as the second executable operator group. The functional attributes, models and operator libraries of the two executable operators are consistent. At this time, the second executable operator group can be directly fused to obtain the merged second executable operator group. For example, one of the executable operators can be selected as the merged second executable operator group OT_1.

[0069] The first executable atomic function OC_1 and the second executable atomic function OC_2 are combined into the third executable operator group. The two executable operators have the same functional attributes, but their models and operator libraries are inconsistent. At this time, OC_1 and OC_2 are displayed side by side to obtain the merged third executable operator group.

[0070] The first executable atomic function LF_1 and the second executable atomic function LF_2 are combined into the fourth executable operator group. Since the two executable operators are model-independent executable operators and their operator libraries are different, LF_1 and LF_2 are displayed side by side to obtain the merged fourth executable operator group.

[0071] To illustrate in detail the operator merging method in the embodiments of this application, Figure 5 The process shown will be further explained below. Figure 5 yes Figure 1The specific flowchart of an exemplary embodiment of step S130 in the algorithm scheme integration method is shown as follows:

[0072] First, the type of each executable operator in each executable operator group is determined.

[0073] When all executable operators in the executable operator group are of the second type, the same second type executable operators can be automatically fused, where the same refers to the operator information of the second type executable operators being consistent; if not, they are displayed side by side.

[0074] When all executable operators in the executable operator group are of the first type, first, it is determined whether the functional attributes of each first type executable operator group in the executable operator group are the same; if not, they cannot be fused and are displayed side by side; if the same, it is further determined whether the operator libraries are the same; if the operator libraries are the same, it is further determined whether the models are the same; when the models are the same, the fusion processing is performed; when the models are different, the final retained target model can be selected by the integration personnel according to the actual use scenario, and the fusion processing is performed on each first type executable operator in the executable operator group based on the target model; if the operator libraries are different, it is further determined whether the models are the same; if the models are different, the fusion processing cannot be performed and they are displayed side by side; if the models are the same, it is determined whether the operator libraries are compatible; if compatible, the retained target operator library is selected, and the fusion processing is performed on each first type executable operator in the executable operator group based on the target operator library; if not compatible, they cannot be fused and are displayed side by side. In addition to determining whether the operator libraries are compatible, it is also necessary to determine whether the operator libraries are reusable; when the operator libraries are compatible and reusable, the retained target operator library is selected, and the fusion processing is performed on each first type executable operator in the executable operator group based on the target operator library.

[0075] S330: Determine the target integration scheme based on each merged executable operator group.

[0076] Each merged executable operator group and related configuration information are packaged to obtain the target integration scheme. In some embodiments, when the executable operator group is composed of two executable operators, after obtaining the merged executable operator group, it is determined whether the merged executable operator group can be fused again to obtain a re-merged executable operator group, and so on, until the fusion is not possible, to obtain a target merged executable operator; the target integration scheme is determined based on the target merged executable operator.

[0077] On the basis of the above embodiments, the embodiments of the present application adopt Figure 6 The flowchart describes in detail how to obtain the atomic function database, please refer to Figure 6 , Figure 6 is Figure 1A flowchart of an exemplary embodiment of the algorithm scheme integration method shown after step S120. Specifically, after step S120, each atomic function is matched with an executable atomic function in the atomic function database to obtain a matching result, and the atomic function database includes at least one executable atomic function and executable operators required to execute each executable atomic function. The process further includes the following steps:

[0078] S610: Obtain an initial function.

[0079] The initial function can refer to a function to be constructed. Exemplarily, a single function required to be implemented can be established first, and then the single function is determined as the initial function. The initial function can be extracted from a large amount of user demand data, and the user demand data is information extracted to obtain the initial function. The user demand data can be obtained from online databases, research information, and the like. For example, the initial function can be a mobile phone detection, a fall detection, a sleep detection, and the like, which is only an expression of a function requirement and is different from the executable atomic function described above. The executable atomic function is a function that can be implemented by running the executable atomic function.

[0080] The algorithm scheme integration device analyzes the user demand data to obtain at least one initial function. It should be noted that the initial function of the present embodiment can cover function requirements in multiple fields to improve the comprehensiveness of the atomic function database as much as possible.

[0081] S620: According to the algorithm flow of the initial function and the function attribute of the executable operator, a target executable operator is obtained from each executable operator for combination to obtain an executable atomic function.

[0082] The algorithm flow of the initial function refers to the processing flow required to implement the initial function. The function attribute of the executable operator refers to the effect that can be achieved after running the executable operator. Exemplarily, the algorithm flow of the initial function can be determined in advance. For example, at least one initial function includes a first initial function and a second initial function arranged in a sequence of execution, the function attribute of the first initial function is set to mobile phone detection, and the function attribute of the second initial function is set to fall detection.

[0083] The executable operator is constructed from the initial operator. Specifically, the algorithm scheme integration device obtains an initial operator; in response to the model association attribute of the initial operator representing model relevance, a target model is selected from a preset model storage module according to the function attribute of the initial operator for mapping association to obtain an executable operator; and in response to the model association attribute of the initial operator representing model irrelevance, the initial operator is determined as an executable operator. Thus, the association between the operator and the model is established, and the reliability of the atomic function database is improved.

[0084] The initial operator represents an operator that has not yet established a mapping relationship with a model in the embodiment. In some embodiments, the developed scheme can be split to obtain at least one initial operator. In other embodiments, the initial operator can be written by the developer according to the functional attributes. The initial operator can include at least one operator library, which can be regarded as composed of code modules that implement specific operation logic. The code modules can include, but are not limited to, mathematical logic operation modules, scheduling modules, and the like.

[0085] When the model association attribute of the initial operator represents model relevance, that is, the initial operator needs to load a model. The algorithm scheme integration device can select a target model from the preset model storage module according to the functional attribute of the initial operator, and map and associate the target model with the initial operator to obtain an executable operator. The number of target models can be one or more. In some embodiments, the functional attribute of the initial operator can be matched with the functional attribute of each model in the preset model storage module to obtain the target model. In other embodiments, the target model can also be selected from the preset model storage module according to the functional attribute and the number of parameters of the initial operator. The preset model storage module is used to store a plurality of models according to the functional attributes of the models. For details, please refer to Figure 7 , Figure 7 is a frame diagram of an exemplary embodiment of the preset model storage module shown in the present application. The preset model storage module stores target detection models, key point detection models, target classification models, and target segmentation models, etc. separately, and models with the same functional attribute are stored together to facilitate subsequent searching. The models with the same functional attribute have certain differences in size, accuracy, target type, etc. The integration personnel can select as needed.

[0086] In other embodiments, the integration personnel can also select the target model of the initial operator from the preset model storage module. The executable operator includes an operator library and a relationship mapping file, and the relationship mapping file is used to store the mapping and association between the operator library and the target model. The mapping and association can be represented as <operator library, target model>, for example, alg_1:<lib_alg_1, <model_1, model_2, model_3>>. It indicates that in the executable operator alg_1, the operator library lib_alg_1 supports loading and running target models model_1, model_2, and model_3.

[0087] When the model association attribute of the initial operator represents model irrelevance, that is, the initial operator does not need to load a model. The algorithm scheme integration device can directly determine the initial operator as an executable operator.

[0088] For details, please refer to Figure 8 , Figure 8is a schematic diagram of a framework of an exemplary embodiment of a preset operator storage module shown in the present application. Executable operators can be stored in the preset operator storage module, in which the executable operators are divided into a first type of executable operators related to models and a second type of executable operators unrelated to models. For the first type of executable operators, an initial mapping association between the executable operators and the models can be established according to the functional attributes of the first type of executable operators and the functional attributes of the models in the preset model storage module. For example, an initial mapping association between an object detection operator and an object detection model, an initial mapping association between a key point detection operator and a key point detection model, an initial mapping association between an object classification operator and an object classification model, and an initial mapping association between an object segmentation operator and an object segmentation model, and the like are established; then for a single executable operator, a target model can also be selected from the models in the initial mapping association, and the selection method can be selected according to the parameter quantity, accuracy or target type, etc. For example, the executable operators include an object detection operator 1 and an object detection operator 2. For the object detection operator 1 (OD_1), the target models supported by the object detection operator library 1 (OD_LIB_1) of OD_1 include an object detection model 1 (OD_model_1) and an object detection model 2 (OD_model_2); for the object detection operator 2 (OD_2), the target models supported by the object detection operator library 2 (OD_LIB_2) of OD_2 include the object detection model 2 (OD_model_2) and an object detection model N (OD_model_N). When the integration scheme is selected, the target model corresponding to the operator library can be selected according to the requirements. The second type of executable operators can also be classified and stored according to the functional attributes, for example, the operators for object tracking are stored together, and the operators for logical processing are stored together.

[0089] The target executable operator is an executable operator matched with the functional attribute of the initial function. Exemplarily, the functional attribute of the initial function and the functional attribute of the executable operator can be matched to obtain the target executable operator corresponding to the initial function. In some embodiments, the target executable operator required by the initial function can be selected by the integration personnel, and the target executable operator can be sequentially connected by the integration personnel according to the processing flow of the initial function to obtain the executable atomic function.

[0090] In some embodiments, when the initial function is split, the initial function can be split by a large model, and the executable operator is used as a prompt word of the large model to prompt the large model to split the initial function into at least one to-be-matched operator based on the executable operator. In other embodiments, the initial function can be input into a pre-trained task disassembly model to obtain at least one to-be-matched operator of the initial function. In other embodiments, the initial function can also be pre-split by an integration personnel to obtain at least one to-be-matched operator.

[0091] In some embodiments, when the initial function is split, the initial function can be split by a large model, and the executable operator is used as a prompt word of the large model to prompt the large model to split the initial function into at least one to-be-matched operator based on the executable operator. In other embodiments, the initial function can be input into a pre-trained task disassembly model to obtain at least one to-be-matched operator of the initial function. In other embodiments, the initial function can also be pre-split by an integration personnel to obtain at least one to-be-matched operator.

[0092] The to-be-matched operator is a small execution unit extracted from the initial function. The to-be-matched operator also has a functional attribute. As an example, for the initial function of detecting a phone call, the initial function is split to obtain a target detection of an execution object and a phone, a tracking detection of an execution object and a phone, an action classification detection of an execution object, and related logical processing, etc. The functional attribute of the first to-be-matched operator is to detect an execution object and a phone in a picture, the functional attribute of the second to-be-matched operator is to track an execution object and a phone in a picture, and the functional attribute of the third to-be-matched operator is to determine whether the detected execution object in the picture is making a phone call, etc.

[0093] The functional attribute of the to-be-matched operator is matched with the functional attribute of the executable operator to obtain a target executable operator corresponding to each to-be-matched operator. For example, for the to-be-matched operator of the target detection of the execution object and the phone, the OD_1 (such as an operator for detecting the head and shoulder) is matched as the corresponding target executable operator, etc.

[0094] After the algorithm scheme integration device matches the target executable operator corresponding to each to-be-matched operator, the target executable operator is sorted according to the algorithm flow of each to-be-matched operator to obtain an executable atomic function. When the target executable operator is mapped to multiple models, an integration personnel can select one of the multiple models as a model for actually running the executable atomic function. The executable atomic function can be stored in a preset function storage module, which can be referred to in Figure 9 , Figure 9is a schematic diagram of an exemplary embodiment of a preset function storage module shown in the present application. The preset function storage module stores at least one executable atomic function, and each executable atomic function is composed of at least one executable operator. For example Figure 9 In the preset function storage module, executable atomic functions such as mobile phone playing detection, phone call detection, fall detection, running detection, and sleep state detection are included. Taking mobile phone playing detection as an example, the required executable operators include target detection operator 2 (OD_2), target tracking operator 1 (OT_1), target classification operator 2 (OC_1), and logic processing operator 2 (LF_2) connected in sequence. Among them, the target detection operator 2 and the target classification operator 2 are model-related first-type executable operators, so the target detection operator 2 can select the target detection model 1, and the target classification operator 2 can select the target classification model 2 as the model actually running the mobile phone playing detection.

[0095] S630: Classify and store the executable atomic function and the executable operator to obtain an atomic function database.

[0096] The atomic function database can include a preset function storage module and a preset operator storage module. The executable atomic function is classified and stored in the preset function storage module, and the executable operator is classified and stored in the preset operator storage module. In other embodiments, the atomic function database can also include a preset model storage module, and the preset model storage module is used to classify and store a plurality of models. The model can be a trained model.

[0097] It can be seen that the algorithm scheme integration method of the embodiment of the present application obtains an initial function; target executable operators are obtained from each executable operator according to the algorithm flow of the initial function and the function attribute of the executable operator for combination to obtain an executable atomic function; and the executable atomic function and the executable operator are classified and stored to obtain an atomic function database. Thus, the executable atomic function and the executable operator are stored in layers, so as to facilitate subsequent delivery and use.

[0098] In constructing the atomic function database, although the atomic function database has as comprehensively as possible included atomic functions in different fields, there can still be omissions, at which time the required atomic function needs to be reconstructed. Specifically, in response to the matching result representing that there is at least one atomic function in each atomic function that does not match each executable atomic function in the atomic function database, a function addition prompt is generated according to the unmatched atomic function; the atomic function database is updated according to the function addition prompt to obtain an updated atomic function database; the target executable atomic function corresponding to the to-be-integrated scheme is matched from the updated atomic function database, and the executable operators required for executing the target executable atomic function are merged to obtain a target integrated scheme. In this way, it can be ensured that each to-be-integrated scheme can be delivered, the user experience is improved, and the atomic function database can be continuously enriched to improve the comprehensiveness of the atomic function database.

[0099] The unmatched atomic function is a function that cannot match the target executable atomic function from the atomic function database. When there is an unmatched atomic function, a function addition prompt needs to be fed back to the client. The integration personnel are prompted to add the corresponding atomic function.

[0100] The function addition prompt includes related information of the atomic function that needs to be added, such as function description, accuracy requirement, etc. The algorithm scheme integration device can create the corresponding atomic function according to the function addition prompt, and add the newly created atomic function to the atomic function database to obtain an updated atomic function database; the updated atomic function database already includes all atomic functions required in the to-be-integrated scheme, so the target executable atomic function corresponding to the to-be-integrated scheme is matched from the updated atomic function database, and the target executable atomic function is merged to obtain a target integrated scheme. The merging manner can refer to the above embodiments, which will not be described here. The updated atomic function database is used for subsequent algorithm scheme integration.

[0101] In some application scenarios, the atomic function required to be newly created in the function addition prompt can be constructed according to the executable operators already in the atomic function database, at which time the newly created atomic function can be obtained by combining the executable operators. In other application scenarios, the atomic function required to be newly created in the function addition prompt cannot be constructed according to the executable operators already in the atomic function database, at which time the code module needs to be re-written to obtain an initial operator, and if a new model is required, the model also needs to be re-trained to obtain the newly created atomic function.

[0102] In order to elaborate the algorithm scheme integration method of the present application, it is further described as follows with reference to the framework diagram shown in Figure 10

[0103] ​The algorithm scheme integration device can include an atomic function database and an integration page. The atomic function database includes a preset function storage module, a preset operator storage module, and a preset model storage module. The preset model storage module is used to store various end-to-end network models. The preset operator storage module is used to manage different categories of executable operators, which are the basic units of a scheme. Each executable operator includes an operator library and an operator configuration file. Model-related executable operators also include a relationship mapping file. The implementation of the operator library includes, but is not limited to, a static library formed by compiling and packaging using C, C++, or Java, etc. The preset function storage module is used to manage executable atomic functions and processing flow files of the executable atomic functions. The processing flow files describe the topology of the initial function in the form of a directed graph. Different executable atomic functions can be constructed by combining the executable operators in the preset operator storage module. The integration page is used to interact with integration personnel. The integration personnel can check executable atomic functions on the integration page as needed. After the checked executable atomic functions are packaged and processed, a complete target integration scheme is obtained. The atomic function database constructed in this way has high reusability, so that rapid delivery can be achieved without the need to understand and familiarize with the underlying operator library and model, thereby reducing the communication cost of different levels of management.

[0104] After obtaining the scheme to be integrated, the scheme to be integrated is functionally split to obtain at least one atomic function. Each atomic function is matched with an executable atomic function in the atomic function database. If a match is found, the corresponding target executable atomic function is directly selected, and the executable operators required by each target executable atomic function are merged and processed to obtain a target integration scheme. If a match is not found, the corresponding atomic function that does not match needs to be split to obtain the operators required by the atomic function that does not match. It is determined whether the operators required by the atomic function that does not match exist in the preset operator storage module. If the operators required by the atomic function that does not match exist in the preset operator storage module, the corresponding executable operators are obtained to constitute the atomic function that does not match, and the constituted atomic function is added to the atomic function database to obtain an updated atomic function database. If the operators required by the atomic function that does not match do not exist in the preset operator storage module, the operators required by the atomic function that does not match need to be re-written, it is determined whether the operator needs to load a model and whether the required model exists in the preset model storage module. If the model exists, it is directly loaded. If the model does not exist, it needs to be re-trained to obtain new executable operators. The atomic function that does not match is constituted according to the new executable operators, and the constituted atomic function is added to the atomic function database to obtain an updated atomic function database. Then, the corresponding target executable atomic function is selected from the updated atomic function database, and the executable operators required by each target executable atomic function are merged and processed to obtain a target integration scheme.

[0105] Referring to Figure 11 , Figure 11 is a structural schematic diagram of an exemplary embodiment of an algorithm scheme integration apparatus shown in the present application. The algorithm scheme integration apparatus 1100 includes a splitting module 1110, a matching module 1120, and a merging module 1130. The splitting module 1110 is configured to perform functional splitting on an acquired scheme to be integrated to obtain at least one atomic function. The matching module 1120 is configured to perform matching processing on each atomic function and executable atomic functions in an atomic function database to obtain a matching result. The atomic function database includes at least one executable atomic function and executable operators required for executing each executable atomic function. The merging module 1130 is configured to, in response to the matching result representing that there is a target executable atomic function matching each atomic function in the atomic function database, perform merging processing on executable operators required for executing the target executable atomic function to obtain a target integrated scheme.

[0106] The above scheme, the algorithm scheme integration apparatus performs functional splitting on an acquired scheme to be integrated to obtain at least one atomic function. The matching module 1120 is configured to perform matching processing on each atomic function and executable atomic functions in an atomic function database to obtain a matching result. The atomic function database includes at least one executable atomic function and executable operators required for executing each executable atomic function. The merging module 1130 is configured to, in response to the matching result representing that there is a target executable atomic function matching each atomic function in the atomic function database, perform merging processing on executable operators required for executing the target executable atomic function to obtain a target integrated scheme. Thus, after obtaining the scheme to be integrated, the algorithm flow does not need to be repeatedly constructed, and the atomic function database can be directly used to integrate the required scheme to be integrated, which can effectively reduce the workload of repeated development and improve the algorithm scheme integration efficiency.

[0107] The functions of each module can be referred to the algorithm scheme integration method embodiments, which will not be described here.

[0108] To implement the algorithm scheme integration method of the above embodiments, the present application provides another electronic device, which is specifically described in Figure 12 , Figure 12 is a structural schematic diagram of an embodiment of the electronic device provided by the present application.

[0109] The electronic device 1200 includes a memory 1210 and a processor 1220, wherein the memory 1210 and the processor 1220 are coupled.

[0110] The memory 1210 is configured to store program data, and the processor 1220 is configured to execute the program data to implement the algorithm scheme integration method of the above embodiments.

[0111] In this embodiment, the processor 1220 can also be referred to as a CPU (Central Processing Unit). The processor 1220 can be an integrated circuit chip having a processing capability of signals. The processor 1220 can also be a general processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general processor can be a microprocessor or the processor 1220 can also be any conventional processor or the like.

[0112] The present application also provides a computer readable storage medium, such as Figure 13 As shown, the computer readable storage medium 1300 is used to store program data 1310, which when executed by a processor, is used to implement the algorithm scheme integration method in the method embodiments of the present application.

[0113] The method involved in the algorithm scheme integration method embodiments of the present application, when implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a device, such as a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the whole or part of the technical solutions that make essential contributions to the prior art can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0114] The above only describes the embodiments of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation using the content of the specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. An algorithmic solution integration method, characterized by, The algorithm scheme integration method comprises: performing function decomposition on the obtained to-be-integrated scheme to obtain at least one atomic function; matching each atomic function with an executable atomic function in an atomic function database to obtain a matching result, the atomic function database comprising at least one executable atomic function and an executable operator required for executing each executable atomic function; in response to the matching result indicating that there is a target executable atomic function matching each atomic function in the atomic function database, performing merging processing on the executable operator required for executing the target executable atomic function to obtain a target integrated scheme.

2. The algorithmic solution integration method of claim 1, wherein, The step of performing merging processing on the executable operator required for executing the target executable atomic function to obtain a target integrated scheme comprises: combining the executable operators required for different target executable atomic functions according to the execution order of each executable operator in each target executable atomic function to obtain at least one executable operator group; performing merging processing on the executable operators in each executable operator group according to the model correlation attribute of the executable operators to obtain each merged executable operator group; determining the target integrated scheme based on each merged executable operator group.

3. The algorithmic solution integration method of claim 2, wherein, The model correlation attribute of the executable operators comprises model-dependent and model-independent, and the step of performing merging processing on the executable operators in each executable operator group according to the model correlation attribute of the executable operators to obtain each merged executable operator group comprises: determining the model-dependent executable operators as first-type executable operators and the model-independent executable operators as second-type executable operators; in response to each executable operator in the executable operator group being the first-type executable operator, determining a merging strategy of the executable operator group according to the consistency of the operator information corresponding to each first-type executable operator and the corresponding model, and merging the first-type executable operators according to the merging strategy to obtain the merged executable operator group; in response to each executable operator in the executable operator group being the second-type executable operator, determining a merging strategy of the executable operator according to the consistency of the operator information of each second-type executable operator, and merging each second-type executable operator according to the merging strategy to obtain the merged executable operator group.

4. The algorithmic scheme integration method of claim 3, wherein, The operator information comprises a function attribute and an operator library, and the merging strategy comprises fusion processing, and the step of determining a merging strategy of the executable operator group according to the consistency of the operator information corresponding to each first-type executable operator and the corresponding model, and merging the first-type executable operators according to the merging strategy to obtain the merged executable operator group comprises: in response to the function attribute, the operator library and the model corresponding to each first-type executable operator being consistent, performing fusion processing on each first-type executable operator in the executable operator group to obtain the merged executable operator group; In response to the function attribute corresponding to each first-type executable operator being consistent with the model and the operator library being inconsistent, a target model is selected from the model corresponding to each first-type executable operator, and each first-type executable operator in the executable operator group is fused based on the target model to obtain the merged executable operator group. In response to the function attribute corresponding to each first-type executable operator being consistent with the model and the operator library being inconsistent, a target model is selected from the model corresponding to each first-type executable operator, and each first-type executable operator in the executable operator group is fused based on the target model to obtain the merged executable operator group.

5. The algorithmic solution integration method of claim 1, wherein, Before the step of matching each atomic function with an executable atomic function in an atomic function database to obtain a matching result, the atomic function database including at least one executable atomic function and an executable operator required to execute each executable atomic function, the method further includes: obtaining an initial function; combining target executable operators obtained from each executable operator according to the algorithm flow of the initial function and the function attribute of the executable operator to obtain the executable atomic function; storing the executable atomic function and the executable operator in a classified manner to obtain the atomic function database.

6. The algorithmic scheme integration method of claim 5, wherein, Before the step of obtaining an initial function, the method further includes: obtaining an initial operator; in response to the model association attribute of the initial operator representing model relevance, a target model is selected from a preset model storage module according to the function attribute of the initial operator to perform mapping association, and the executable operator is obtained; in response to the model association attribute of the initial operator representing model irrelevance, the initial operator is determined as an executable operator.

7. The algorithmic scheme integration method of claim 5, wherein, The step of combining target executable operators obtained from each executable operator according to the algorithm flow of the initial function and the function attribute of the executable operator to obtain the executable atomic function includes: splitting the initial function according to the algorithm flow of the initial function to obtain at least one to-be-matched operator; matching the function attribute of the to-be-matched operator with the function attribute of the executable operator to obtain a target executable operator corresponding to each to-be-matched operator; sorting and combining the corresponding target executable operator according to the algorithm flow of each to-be-matched operator to obtain the executable atomic function.

8. The algorithmic solution integration method of claim 1, wherein, After the step of matching each atomic function with an executable atomic function in an atomic function database to obtain a matching result, the atomic function database including at least one executable atomic function and an executable operator required to execute each executable atomic function, the method further includes: in response to the matching result representing that there is at least one atomic function in each atomic function that does not match each executable atomic function in the atomic function database, generating a function addition prompt according to the unmatched atomic function; updating the atomic function database according to the function addition prompt to obtain an updated atomic function database; Matching the target executable atomic function corresponding to the to-be-integrated scheme from the updated atomic function database, and performing a merging process on executable operators required for executing the target executable atomic function to obtain the target integrated scheme.

9. An electronic device, comprising: Comprising: A memory and a processor, wherein the memory stores program instructions, and the processor fetches the program instructions from the memory to execute the method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, Comprising: Program data stored in a memory, wherein the program data is executed by a processor to implement the method according to any one of claims 1-8.

Images