Algorithm solution customization method, device, equipment and computer storage medium

Abstract

This application proposes an algorithm solution customization method, an algorithm solution customization device, an algorithm solution customization device, and a computer storage medium. The algorithm solution customization method includes: obtaining a target scene large model; combining a number of logical operators with the target scene large model to form a number of candidate algorithm solutions; training the candidate algorithm solutions through a training set to obtain a number of decision models; determining the final decision model and its final logical operator based on the output of the decision models; combining the final logical operator and the target scene large model into a final algorithm solution. Through the above-mentioned algorithm solution customization method, by combining a large model with good model training customization effect with a logical operator, customized deployment of intelligent algorithm solutions with fast deployment, high accuracy, and strong stability can be achieved.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent algorithm solution development, and in particular to an algorithm solution customization method, an algorithm solution customization apparatus, an algorithm solution customization device, and a computer storage medium. Background Art

[0002] With the rapid development of AI (Artificial Intelligence) technology, it has become a vital force driving economic and social development, with applications spanning finance, healthcare, transportation, education, and more. In particular, AI-based audio and video intelligent algorithm solutions are widely used in security, across numerous video scenarios.

[0003] The implementation of an intelligent audio and video algorithm solution often requires extensive model customization, development, and debugging by professional algorithm engineers. However, with the rapid growth of customized business, the increased investment in specialized algorithm engineers will result in higher labor costs, leading to lower customization efficiency of algorithm solutions. Summary of the Invention

[0004] In order to solve the above technical problems, the present application proposes an algorithm solution customization method, an algorithm solution customization device, an algorithm solution customization equipment and a computer storage medium.

[0005] To solve the above technical problems, this application proposes an algorithm solution customization method, which includes:

[0006] Obtain a large model of the target scene;

[0007] Combining a plurality of logical operators with the target scene large model to form a plurality of candidate algorithm solutions;

[0008] Training the candidate algorithm solutions through the training set to obtain several decision models;

[0009] Determining a final decision model and its final logical operator based on the outputs of the plurality of decision models;

[0010] The final logical operator and the target scene large model are combined into a final algorithm solution.

[0011] The step of determining the final decision model and its final logical operator based on the outputs of the plurality of decision models includes:

[0012] Get the output and weight of each decision model;

[0013] Add the weights of decision models with the same output to get the highest weight;

[0014] The decision model with the highest weight is used as the final decision model, and the logical operator of the final decision model is used as the final logical operator.

[0015] The method of combining several logical operators with the target scene large model to form several candidate algorithm solutions includes:

[0016] Obtaining the recognition categories and rule sets supported by the target scene macro model;

[0017] Several logical operators are respectively combined with the target scene large model and the recognition categories and rule sets it supports to form several candidate algorithm solutions.

[0018] After obtaining the recognition categories and rule sets supported by the target scene large model, the algorithm solution customization method further includes:

[0019] Obtaining a custom category input by a user, and converting the custom category into a custom category feature vector;

[0020] The custom category feature vector is matched with the feature vector of the recognition category supported by the target scene large model for similarity to determine the target category.

[0021] The step of performing similarity matching between the custom category feature vector and the feature vector of the recognition category supported by the target scene large model to determine the target category includes:

[0022] Performing similarity matching between the custom category feature vector and the feature vector of the recognition category supported by the target scene large model;

[0023] Generate a candidate category list based on the similarity matching results;

[0024] In response to a user selection instruction, one candidate category in the candidate category list is determined as the target category.

[0025] After obtaining the recognition categories and rule sets supported by the target scene large model, the algorithm solution customization method further includes:

[0026] Obtaining a custom rule input by a user, and converting the custom rule into a custom rule feature vector;

[0027] The custom rule feature vector is matched with the feature vector of the rule set supported by the target scene large model for similarity to determine the target rule.

[0028] After combining the final logical operator and the target scene large model into a final algorithm solution, the algorithm solution customization method further includes:

[0029] Run the final algorithm solution to obtain the output result of the algorithm solution;

[0030] Determine whether the algorithm output is consistent with the user scenario;

[0031] If yes, deploy the final algorithm solution to the user equipment;

[0032] If not, replace the scene model in the final algorithm solution until the output result of the replaced algorithm solution meets the user scenario.

[0033] In order to solve the above technical problems, the present application also proposes an algorithm solution customization device, which includes: an acquisition module, a combination module, a training module, and a decision module; wherein,

[0034] The acquisition module is used to acquire a large model of the target scene;

[0035] The combination module is used to combine a plurality of logical operators with the target scene large model to form a plurality of candidate algorithm solutions;

[0036] The training module is used to train the candidate algorithm solutions through the training set to obtain several decision models;

[0037] The decision module is used to determine a final decision model and its final logical operator based on the outputs of the multiple decision models;

[0038] The combination module is used to combine the final logical operator and the target scene large model into a final algorithm solution.

[0039] In order to solve the above technical problems, the present application also proposes an algorithm solution customization device, which includes a memory and a processor coupled to the memory; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the algorithm solution customization method as described above.

[0040] In order to solve the above technical problems, the present application also proposes a computer storage medium, which is used to store program data. When the program data is executed by a computer, it is used to implement the above algorithm solution customization method.

[0041] Compared with the existing technology, the beneficial effects of this application are as follows: the algorithm solution customization device obtains a large model of the target scene; several logical operators are respectively combined with the target scene large model to form several candidate algorithm solutions; the several candidate algorithm solutions are trained with a training set to obtain several decision models; based on the outputs of the several decision models, the final decision model and its final logical operator are determined; the final logical operator and the target scene large model are combined to form the final algorithm solution. Through the above-mentioned algorithm solution customization method, through the combination of a large model with good model training customization effect and a logical operator, a customized deployment of an intelligent algorithm solution with fast deployment, high accuracy, and strong stability is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without inventive efforts. Among them:

[0043] Figure 1 This is a flowchart of an embodiment of the algorithm solution customization method provided by this application;

[0044] Figure 2 This is a flowchart of another embodiment of the algorithm solution customization method provided by this application;

[0045] Figure 3 This is a flowchart of another embodiment of the algorithm solution customization method provided by this application;

[0046] Figure 4 This is a structural diagram of an embodiment of an algorithm solution customization device provided by this application;

[0047] Figure 5 This is a structural diagram of an embodiment of an algorithm solution customization device provided by this application;

[0048] Figure 6 It is a structural diagram of an embodiment of a computer storage medium provided by this application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0050] The terms "first," "second," "third," "fourth," and the like (if any) in the specification and claims of this application and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the application described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not necessarily limited to those steps or elements explicitly listed, but may include other steps or elements not explicitly listed or inherent to such process, method, product, or apparatus.

[0051] In order to solve the problems of the existing technology, this application provides a method for deploying a new audio and video intelligent algorithm solution with fast deployment, high accuracy and strong stability. Figure 1 , Figure 1 This is a flowchart of an embodiment of the algorithm solution customization method provided by this application.

[0052] The algorithm solution customization method of the present application is applied to an algorithm solution customization device, wherein the algorithm solution customization device of the present application can be a server, a terminal device, or a system composed of a server and a terminal device. Accordingly, the various parts of the algorithm solution customization device, such as the various units, subunits, modules, and submodules, can be all set in the server, all set in the terminal device, or separately set in the server and the terminal device.

[0053] Furthermore, the server described above may be either hardware or software. When the server is hardware, it may be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When the server is software, it may be implemented as multiple software programs or software modules, such as software or software modules for providing a distributed server, or as a single software program or software module, without further limitation.

[0054] like Figure 1 As shown, the specific steps are as follows:

[0055] Step S11: Obtain a large model of the target scene.

[0056] In the embodiment of the present application, the algorithm solution customization device pre-deploys common intelligent solutions on the audio and video intelligent algorithm platform, classifies them by scene, and deploys pre-trained large models for different scenes on the AI ​​platform, as shown below:

[0057] Smodel={m1,m2,…,mi}

[0058] Among them, Smodel represents the set of deployed large models, and mi represents the large model of the i-th scene.

[0059] The algorithm solution customization device selects one or more target scene large models from the above-mentioned large model set Smodel. The selection criteria are that the user directly selects the target scene large model, or the user provides selectable scenes through the algorithm solution customization device. The user only needs to select the scene, and the algorithm solution customization device determines one or more large models corresponding to the scene.

[0060] Step S12: Combining several logical operators with the target scene large model to form several candidate algorithm solutions.

[0061] In an embodiment of the present application, the algorithm solution customization device combines several logical operators deployed by the audio and video intelligent algorithm platform with the target scene large model selected in step S11 to form candidate algorithm solutions, wherein the logical operators in the candidate algorithm solution can be one or more.

[0062] Furthermore, the several logical operators may be all logical operators deployed in the audio and video intelligent algorithm platform, or may be some logical operators bound to the target scene among all logical operators deployed in the audio and video intelligent algorithm platform.

[0063] In summary, the algorithm solution customization device obtains Smodel with different logical operators for audio and video intelligent solutions, as shown below:

[0064] Ssolu={solu1, solu2,...,solua}

[0065] Here, solua represents the ath audio and video solution.

[0066] Furthermore, the algorithm solution customization device obtains data such as the logical operators and scene models of each audio and video intelligent solution, as shown below:

[0067] solui={modeli, logicAlgi}

[0068] Among them, solui represents the i-th solution, modeli represents the scene model used in this application, and logicAlgi represents the logical operator matched with the scene model.

[0069] Step S13: Train several candidate algorithm solutions through the training set to obtain several decision models.

[0070] In the embodiment of the present application, the algorithm solution customization device trains the logical operator decision model composed of step S12.

[0071] Specifically, the algorithm solution customization device uses the above-mentioned audio and video intelligent solution to train m different neural networks to obtain m decision models. The output of the final decision model is as follows:

[0072] out={w1*out1,w2*out2,...,wm*outm}

[0073] Among them, w is the weight and out is the output of each decision model.

[0074] It should be noted that the weight of each decision model can be understood as the output confidence, that is, the higher the output confidence, the higher the weight.

[0075] Step S14: Determine a final decision model and its final logical operator based on the outputs of several decision models.

[0076] In this embodiment of the present application, the algorithm solution customization device adds the weights of the same outputs of all decision models, ultimately taking the output with the highest weight as the final output, which is output as a logical operator. Therefore, it can be determined that the decision model corresponding to the output with the highest weight is the final decision model, and the logical operators that make up the final decision model are the final logical operators.

[0077] Step S15: Combine the final logical operator and the target scene large model into the final algorithm solution.

[0078] In an embodiment of the present application, the algorithm solution customization device combines the final logical operator determined in step S14 with the target scene large model determined in step S11 to form a customized audio and video intelligent solution required by the user.

[0079] In this application, an algorithm solution customization device obtains a target scene large model; combines several logical operators with the target scene large model to form several candidate algorithm solutions; trains the several candidate algorithm solutions through a training set to obtain several decision models; determines the final decision model and its final logical operator based on the output of the several decision models; and combines the final logical operator and the target scene large model into a final algorithm solution. Through the above-mentioned algorithm solution customization method, a large model with good model training customization effect is combined with a logical operator to achieve customized deployment of intelligent algorithm solutions with fast deployment, high accuracy, and strong stability.

[0080] Furthermore, in order to improve the pertinence and accuracy of customized deployment of intelligent algorithm solutions, this application can also add relevant identification categories and logical operator rules to the intelligent algorithm solutions. Figure 2 , Figure 2 This is a flow chart of another embodiment of the algorithm solution customization method provided by this application.

[0081] like Figure 2 As shown, the specific steps are as follows:

[0082] Step S21: Obtain the recognition categories and rule sets supported by the target scene macro model.

[0083] In the embodiment of the present application, the algorithm solution customization device further obtains the categories supported by each pre-trained scene model for detection and recognition as follows:

[0084] mi={label1, label2,..., labelj}

[0085] Among them, labelj represents the jth category supported for recognition.

[0086] Furthermore, in order to facilitate storage and matching, the algorithm solution customization device can also use a text-to-vector model to convert the category text into a feature vector, as shown below:

[0087] mi={label1{v1, v2,..., vk}, label2{v1, v2,..., vk},..., labelj{v1, v2,..., vk}}

[0088] Among them, k represents the dimension of the transformed feature vector, and vk represents the kth feature in the feature vector.

[0089] The algorithm solution customization device further obtains the rules supported by the logical operators of the current target scene large model. The logical operators can perform further logical analysis based on the large model detection results and rules to provide rule results. The specific rules are as follows:

[0090] rulei={rule1, rule2,..., rulep}

[0091] rulep indicates the number of rules supported by the current logical operator.

[0092] Furthermore, in order to facilitate storage and matching, the algorithm solution customization device can also use a text-to-vector model to convert the regular text into a feature vector, as shown below:

[0093] rulei={rule1{v1, v2,..., vk}, rule2{v1, v2,..., vk},..., rulep{v1, v2,..., vk}}

[0094] Among them, k represents the dimension of the transformed feature vector.

[0095] Step S22: Combine several logical operators with the target scene model and the recognition categories and rule sets it supports to form several candidate algorithm solutions.

[0096] In the embodiment of the present application, according to step S21, the algorithm solution customization device obtains data such as target categories, rules, and scene models supported by each audio and video intelligent solution for detection, as shown below:

[0097] solui={modeli, labeli, rulei, logicAlgi}

[0098] Among them, solui represents the i-th solution, modeli represents the scene model used in this application, labeli represents the category set supported by the scene model, logicAlgi represents the logical operator matched with the scene model, and rulei represents the rule set supported by the scene model.

[0099] The subsequent process of training the logic operator decision model has been explained in detail in the above steps S13 and S14 and will not be repeated here.

[0100] Furthermore, in order to improve the customization level of the present application, the algorithm solution customization device of the present application can also further confirm the detection category capabilities and rules of the embodiments of the present application.

[0101] Specifically, the user can select the scene model modely through the algorithm solution customization device, and the user can customize the target category labelx that the current customized solution needs to detect. Furthermore, the algorithm solution customization device uses the text-to-vector model to convert the target category labelx into a feature vector, as shown below:

[0102] labelx = {v1, v2, ..., vk}

[0103] Among them, k represents the dimension of the converted feature vector. The similarity matching is performed on the feature and the set of category feature vectors supported in the corresponding scene modely. The calculation formula is as follows:

[0104]

[0105] Among them, sim represents the similarity between label_1 and label_2, and the similarity result with the vector set in modely is obtained as follows:

[0106] Ssim={sim1, sim2,..., simj}

[0107] Among them, j represents the number of detection categories supported by the current large model.

[0108] Furthermore, the algorithm solution customization device sorts the results in the Ssim set, obtains the top n results, and feeds them back to the user for selection. If the user selects the category to be detected, the next step is continued; if not, the current step is continued.

[0109] After the user confirms the scene and confirms that the target category is detected successfully, the user can further customize the rules currently required, such as area intrusion, area retention, area counting, etc. The algorithm solution customization device uses the text-to-quantity model to convert the rule into a feature vector, as shown below:

[0110] ruley={v1,v2,...,vk}

[0111] Among them, k represents the dimension of the converted feature vector. The similarity matching is performed on the feature and the set of rule feature vectors supported in the corresponding scene modely. The calculation formula is as follows:

[0112]

[0113] Furthermore, the algorithm solution customization device obtains the similarity result with the rule vector set in modely, as follows:

[0114] Ssim={sim1, sim2,..., simp}

[0115] Here, p represents the number of rules supported by the current logical operator. Furthermore, the similarity results in the current Ssim set are sorted, the top m results are obtained, and feedback is given to the user. If the user selects one of the top m rules or does not need a rule, the next step is continued. If no rule is selected, the current step is continued.

[0116] Finally, after confirming the scene model, detection category, and issuing rule data, the user uses these data as input to the decision model in step S13, and obtains the logic operator output type by training the logic operator decision model.

[0117] Based on the above Figure 1 and / or Figure 2 After the algorithm solution customization method shown customizes the audio and video intelligent solution, the algorithm solution customization device also needs to consider the operation of the audio and video intelligent solution deployed in the user equipment, that is, to test the deployment effect of the audio and video intelligent solution.

[0118] Please continue to read for details Figure 3 , Figure 3 This is a flow chart of another embodiment of the algorithm solution customization method provided by this application.

[0119] like Figure 3 As shown, the specific steps are as follows:

[0120] Step S31: Run the final algorithm solution and obtain the output result of the algorithm solution.

[0121] In the embodiments of this application, users provide actual scene materials, with formats not limited to images or videos. The user compresses and uploads the scene materials through the algorithm solution customization device, using compression types not limited to zip, rar, or tarballs. After the user uploads the materials, the algorithm solution customization device uses the materials as input to the established intelligent algorithm solution for inference, obtaining detection and rule results.

[0122] Step S32: Determine whether the algorithm output result is consistent with the user scenario.

[0123] In the embodiment of the present application, the user evaluates the final algorithm solution based on the detection and rule results. If the output result meets the user's needs, the process proceeds to step S33; if the output result does not meet the user's needs, the process proceeds to step S34.

[0124] Step S33: deploy the final algorithm solution to the user equipment.

[0125] In an embodiment of the present application, the algorithm solution customization device packages the intelligent algorithm solution and deploys it to the user-specified device. The deployment method is to decompress and replace the corresponding model, configuration file, executable file and dependent dynamic library.

[0126] Step S34: Replace the scene model in the final algorithm solution until the output result of the replaced algorithm solution meets the user scenario.

[0127] In an embodiment of the present application, the algorithm solution customization device fails to confirm the scenario for the customer. The user needs to provide the materials of the corresponding scenario, complete the labeling work, and package and upload them. The packaging method is to compress them into a compressed package with tar, zip, or rar suffixes.

[0128] The algorithm solution customization device fine-tunes the pre-trained large model of the scenario selected in the above steps. After the fine-tuning is completed, the large model in the intelligent algorithm solution is replaced to form a new intelligent algorithm solution.

[0129] The user's labeled materials for the corresponding scenes are used as test materials and uploaded to the platform's automated testing module. Automated testing can reduce manual testing costs. By uploading the intelligent algorithm solution and the corresponding test materials, the detection accuracy, recall rate, mAP value, etc. can be output after the test. It also supports viewing the inference results frame by frame, making it convenient for users to check whether the fine-tuned model achieves the expected effect. If it achieves the user's expected effect, the algorithm solution is deployed. If it does not meet the customer's expected effect, continue with this step.

[0130] For the above deployment plan steps, users only need to download the packaged intelligent algorithm solution, and then deploy the model, configuration file, executable file and dynamic library in the packaged intelligent algorithm solution to the target device accordingly. They can then view the intelligent algorithm results in real time on site, reducing the professional requirements for users.

[0131] In this way, users can intuitively see the effect of customized solutions in corresponding demand scenarios, which greatly reduces the solution customization cycle and enables rapid customization, delivery, and deployment.

[0132] Those skilled in the art will understand that in the above-mentioned method of the specific implementation method, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0133] In order to implement the above algorithm solution customization method, this application also proposes an algorithm solution customization device, please refer to Figure 4 , Figure 4 It is a structural diagram of an embodiment of an algorithm solution customization device provided in this application.

[0134] The algorithm solution customization device 500 of this embodiment includes: an acquisition module 51 , a combination module 52 , a training module 53 , and a decision module 54 .

[0135] The acquisition module 51 is used to acquire a large model of the target scene.

[0136] The combination module 52 is used to combine a plurality of logical operators with the target scene large model to form a plurality of candidate algorithm solutions.

[0137] The training module 53 is used to train the candidate algorithm solutions through a training set to obtain a number of decision models.

[0138] The decision module 54 is used to determine a final decision model and its final logical operator according to the outputs of the multiple decision models.

[0139] The combination module 52 is used to combine the final logical operator and the target scene large model into a final algorithm solution.

[0140] In order to implement the above algorithm solution customization method, this application also proposes an algorithm solution customization device, please refer to Figure 5 , Figure 5 It is a structural diagram of an embodiment of an algorithm solution customization device provided in this application.

[0141] The algorithm solution customization device 400 of this embodiment includes a processor 41 , a memory 42 , an input / output device 43 , and a bus 44 .

[0142] The processor 41 , memory 42 , and input / output device 43 are respectively connected to a bus 44 . The memory 42 stores program data, and the processor 41 is used to execute the program data to implement the algorithm solution customization method described in the above embodiment.

[0143] In the embodiments of the present application, the processor 41 may also be referred to as a CPU (Central Processing Unit). The processor 41 may be an integrated circuit chip with signal processing capabilities. The processor 41 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. A general-purpose processor may be a microprocessor, or the processor 41 may be any conventional processor.

[0144] This application also provides a computer storage medium, please continue to refer to Figure 6 , Figure 6 It is a structural diagram of an embodiment of a computer storage medium provided in the present application. The computer storage medium 600 stores a computer program 61. When the computer program 61 is executed by the processor, it is used to implement the algorithm solution customization method of the above embodiment.

[0145] When the embodiments of the present application are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0146] The above description is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for customizing an algorithm solution, characterized in that: The algorithm solution customization method includes: Obtain a large model of the target scene; Combining a plurality of logical operators with the target scene large model to form a plurality of candidate algorithm solutions; Training the candidate algorithm solutions through the training set to obtain several decision models; Determining a final decision model and its final logical operator based on the outputs of the plurality of decision models; The final logical operator and the target scene large model are combined into an audio and video intelligent solution.

2. The algorithm solution customization method according to claim 1, characterized in that: Determining a final decision model and its final logical operator based on the outputs of the plurality of decision models includes: Obtaining the output and weight of each decision model, wherein the output of the decision model is a logical operator; Add the weights of decision models with the same output to get the highest weight; The decision model with the highest weight is used as the final decision model, and the logical operator output by the final decision model is used as the final logical operator.

3. The algorithm solution customization method according to claim 1, characterized in that: The combining of several logical operators with the target scene large model into several candidate algorithm solutions includes: Obtaining the recognition categories and rule sets supported by the target scene macro model; Several logical operators are respectively combined with the target scene large model and the recognition categories and rule sets it supports to form several candidate algorithm solutions.

4. The algorithm solution customization method according to claim 3, characterized in that: After obtaining the recognition categories and rule sets supported by the target scene large model, the algorithm solution customization method further includes: Obtaining a custom category input by a user, and converting the custom category into a custom category feature vector; The custom category feature vector is matched with the feature vector of the recognition category supported by the target scene large model for similarity to determine the target category.

5. The algorithm solution customization method according to claim 4, characterized in that: The similarity matching of the custom category feature vector with the feature vector of the recognition category supported by the target scene large model to determine the target category includes: Performing similarity matching between the custom category feature vector and the feature vector of the recognition category supported by the target scene large model; Generate a candidate category list based on the similarity matching results; In response to a user selection instruction, one candidate category in the candidate category list is determined as the target category.

6. The algorithm solution customization method according to claim 3, characterized in that: After obtaining the recognition categories and rule sets supported by the target scene large model, the algorithm solution customization method further includes: Obtaining a custom rule input by a user, and converting the custom rule into a custom rule feature vector; The custom rule feature vector is matched with the feature vector of the rule set supported by the target scene large model for similarity to determine the target rule.

7. The algorithm solution customization method according to claim 1, characterized in that: After combining the final logical operator and the target scene large model into an audio and video intelligent solution, the algorithm solution customization method further includes: Run the audio and video intelligent solution to obtain the output results of the algorithm solution; Determine whether the algorithm output is consistent with the user scenario; If so, deploy the audio and video intelligent solution to the user device; If not, replace the scene model in the audio and video intelligent solution until the output result of the replaced algorithm solution meets the user scenario.

8. An algorithm solution customization device, characterized in that: The algorithm solution customization device includes: an acquisition module, a combination module, a training module, and a decision module; wherein, The acquisition module is used to acquire a large model of the target scene; The combination module is used to combine a plurality of logical operators with the target scene large model to form a plurality of candidate algorithm solutions; The training module is used to train the candidate algorithm solutions through the training set to obtain several decision models; The decision module is used to determine a final decision model and its final logical operator based on the outputs of the multiple decision models; The combination module is used to combine the final logical operator and the target scene large model into an audio and video intelligent solution.

9. An algorithm solution customization device, characterized in that: The algorithm solution customization device includes a memory and a processor coupled to the memory; The memory is used to store program data, and the processor is used to execute the program data to implement the algorithm solution customization method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that The computer storage medium is used to store program data, and when the program data is executed by a computer, it is used to implement the algorithm solution customization method according to any one of claims 1 to 7.

Images