Target identification method and system

Abstract

The invention provides a target recognition method and system, and belongs to the technical field of image processing, and the method comprises the steps: analyzing an input instruction of a user, calling at least one first model according to an analysis result, and carrying out the target recognition of a target of a corresponding type in a to-be-recognized image, determining an identification target with the highest confidence output by each first model; generating prompt point information according to the position information of the recognition target with the highest confidence coefficient, and matching a preset second model according to a numerical value interval in which the highest confidence coefficient is located; based on the prompt point information and a second model, performing target identification on the to-be-identified image to obtain an expanded identification target; and calculating the similarity between the image of any expanded identification target of which the confidence coefficient is less than a first threshold value and the sample image of the same type in the task type library, and keeping the any expanded identification target under the condition that the similarity is greater than a second threshold value. According to the invention, the precision and robustness of target identification are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to an object recognition method and system. Background Art

[0002] With the continuous development of remote sensing technology, remote sensing images are increasingly widely used in the fields of environmental monitoring, river and lake shoreline management, disaster warning, etc. As one of the key technologies for remote sensing data processing, remote sensing intelligent recognition has become an important means to achieve efficient and accurate monitoring by extracting various information from a large number of remote sensing images. Traditional remote sensing recognition methods rely on a large number of samples for training and extract information from remote sensing images through deep learning and pattern recognition technologies. Although these methods have improved the recognition efficiency and accuracy to a certain extent, they still face some challenges in practical applications, mainly reflected in the problems of low recognition accuracy and weak robustness.

[0003] Currently, the types in remote sensing images are complex and variable, and factors such as the resolution, illumination change, and weather conditions of the images have a significant impact on the recognition effect. Traditional methods often rely on a single model for classification and recognition, lacking the comprehensive adaptability to complex scenarios and diversification, resulting in a decrease in recognition accuracy in some complex scenarios. To solve these problems, how to improve the accuracy and robustness of object recognition has become the core problem to be solved urgently in the field of remote sensing recognition. Summary of the Invention

[0004] The present invention provides an object recognition method and system to solve at least one defect in object recognition in the prior art.

[0005] In a first aspect, the present invention provides an object recognition method, including: parsing an input instruction of a user, calling at least one first model according to the parsing result to perform object recognition on corresponding types of objects in an image to be recognized, and determining the recognition object with the highest confidence output by each first model; generating prompt point information according to the position information of the recognition object with the highest confidence, and matching a preset second model according to the numerical interval where the highest confidence is located; based on the prompt point information and the second model, performing object recognition on the image to be recognized to obtain an extended recognition object having the same object type as the recognition object with the highest confidence; calculating the similarity between the image of any extended recognition object with a confidence less than a first threshold and a sample image of the same type in a task type library, and retaining the any extended recognition object when the similarity is greater than a second threshold; wherein, the sample images in the task type library include the images corresponding to the recognition object with the highest confidence output by the first model.

[0006] A target recognition method provided by the present invention parses the input instruction of the user and calls at least one first model according to the parsing result, including: performing text parsing on the input instruction based on a large language model to obtain at least one type keyword; based on the type keyword, using semantic similarity to match the type name of the first model in the recognition type library; the recognition type library includes the type names of multiple first models; calling the first model based on the matched type name of the first model.

[0007] A target recognition method provided by the present invention, based on the type keyword, using semantic similarity to match the type name of the first model in the recognition type library, includes: using a CLIP multi-modal pre-trained model to encode the type keyword and the type name in the recognition type library respectively, and converting them into vector forms; in the same vector space, calculating the similarity between the vector of the type keyword and the vector of the type name of each first model in the recognition type library; determining the most matched type name of the first model according to the similarity.

[0008] A target recognition method provided by the present invention generates prompt point information according to the position information of the recognition target with the highest confidence, including: extracting the outer contour according to the position information of the recognition target with the highest confidence; fitting the outer contour with at least one rectangle, and setting multiple generalized prompt points inside each rectangle; screening all the generalized prompt points, and retaining the generalized prompt points located inside the outer contour to generate prompt point information.

[0009] A target recognition method provided by the present invention sets multiple generalized prompt points inside each rectangle, including: setting a generalized prompt point at the center point of the rectangle; setting at least one generalized prompt point on each side of the center point on the central axis in the long side direction of the rectangle.

[0010] A target recognition method provided by the present invention calculates the similarity between the image of any extended recognition target with a confidence level less than the first threshold and the sample image of the same type in the task type library, including: determining a feature extraction model according to the target type of the any extended recognition target; using the feature extraction model to extract the first image feature of the image of the any extended recognition target and the second image feature of the sample image of the same type respectively; calculating the similarity between the first image feature and the second image feature.

[0011] The image to be recognized in a target recognition method provided by the present invention is a remote sensing image.

[0012] In a second aspect, the present invention also provides a target recognition system, including: A type of intelligent agent is used to parse the input instructions of the user, call at least one first model according to the parsing result to perform target recognition on the corresponding type of target in the image to be recognized, and determine the recognition target with the highest confidence output by each first model; A generalization intelligent agent is used to generate prompt point information according to the position information of the recognition target with the highest confidence, and match a preset second model according to the numerical interval where the highest confidence is located; It is also used to perform target recognition on the image to be recognized based on the prompt point information and the second model to obtain an extended recognition target with the same target type as the recognition target with the highest confidence; An induction intelligent agent is used to calculate the similarity between the image of any extended recognition target with a confidence less than the first threshold and the sample image of the same type in the task type library, and retain the any extended recognition target when the similarity is greater than the second threshold; Wherein, the sample images in the task type library include the images corresponding to the recognition targets with the highest confidence output by the first model.

[0013] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the target recognition method as described in any one of the above are implemented.

[0014] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the target recognition method as described in any one of the above are implemented.

[0015] The target recognition method and system provided by the present invention have the following beneficial effects compared with the prior art: (1) The target recognition method and system provided by the present invention improve the accuracy and robustness of target recognition by introducing the first model, the second model, etc. to cooperate in recognizing the target.

[0016] (2) The target recognition method and system provided by the present invention call a specific recognition model according to the user's instructions, quickly recognize the target in the image and output the result with the highest confidence.

[0017] (3) The target recognition method and system provided by the present invention analyze the highest confidence information, call different second models, and extract all ground objects of the same type in the image, thereby expanding the recognition range.

[0018] (4) When facing low-confidence targets, the present invention matches the low-confidence targets with the samples in the task type library based on the feature extraction model, and eliminates the results with strong interference, ensuring the accuracy and integrity of the final recognition result. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 is a schematic framework diagram of the target recognition system provided by the present invention; Figure 2 is a schematic diagram of the generalization hint points provided by the present invention; Figure 3 is a schematic structural diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0022] It should be noted that in the description of the embodiments of the present invention, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the phrase "including a..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0023] The terms "first", "second", etc. in the present application are used to distinguish similar objects and are not used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same category and do not limit the number of objects. For example, the first object can be one or multiple. In addition, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates an "or" relationship between the associated objects before and after.

[0024] The following will combine with Figures 1 - 3 to describe the target recognition method and system provided by the embodiments of the present invention.

[0025] The target recognition method provided by the present invention includes, but is not limited to, the following steps: Step 101: Parse the input instruction of the user, and according to the parsing result, call at least one first model (recognition model) to perform target recognition on the corresponding type of target in the image to be recognized, and determine the recognition target with the highest confidence level output by each first model.

[0026] Optionally, the image to be recognized in the present invention is a remote sensing image, and the recognized targets are different types of ground object targets.

[0027] The present invention can design a type agent to implement the function of step 101.

[0028] Among them, parsing the input instruction of the user and calling at least one first model according to the parsing result, the specific implementation method includes: (1) Perform text parsing on the input instruction based on a large language model to obtain at least one type keyword.

[0029] For example, for the input instruction: Please identify cage aquaculture from the picture; then the type keyword parsed by the text is: cage aquaculture.

[0030] (2) Based on the type keyword, use semantic similarity to match the type name of the first model in the recognition type library; the recognition type library includes the type names of multiple first models.

[0031] For example, the recognition type library includes cage aquaculture, and the semantic similarity between cage aquaculture and net enclosure aquaculture is the highest, then the type name of the first model can be determined as cage aquaculture.

[0032] Optionally, the method of using semantic similarity to match the type name of the first model in the recognition type library includes: 1) Use the CLIP multi-modal pre-training model to encode the type keyword and the type name in the recognition type library respectively, and convert them into vector forms; Use the CLIP model to encode the type keyword and the type name in the recognition type library. CLIP (Contrastive Language–Image Pre–training) is a multi-modal model that can understand both images and texts. In this step, both the type keyword and the type name in the recognition type library will be converted into vector forms.

[0033] 2) In the same vector space, calculate the similarity between the vector of the type keyword and the vector of the type name of each first model in the recognition type library.

[0034] In the same vector space, calculate the similarity between the vector of the type keyword and the type name vector of each first model in the recognition type library. For example, it can be achieved by calculating the cosine similarity between vectors. The cosine similarity can well measure the closeness of two vectors in direction and is less affected by their magnitudes. This similarity score can reflect the semantic similarity between the type keyword and the type name in the recognition type library.

[0035] 3) Determine the type name of the first model with the best match according to the similarity.

[0036] According to the similarity score obtained in the previous step, determine the type name of the first model with the best match. Usually, the type name with the highest similarity score is selected as the best match result. This method not only considers the literal match but also the underlying semantic information. Therefore, even if the type keyword is not exactly the same as some names in the recognition type library, as long as they are semantically similar, they can be correctly matched.

[0037] (3) Based on the type name of the first model that is matched, call the first model.

[0038] It should be noted that the first model in the present invention is an object recognition model, and the first model has different types and is used for the recognition of different objects. For example, the first model with the type name of cage aquaculture is particularly suitable for the object recognition of cage aquaculture.

[0039] It can be understood that the recognition object of each first model, including but not limited to the category, confidence level, and location information (such as bounding box information or mask information, etc.) of the recognition object, etc., then the recognition object with the highest confidence level can be determined as the final output of each first model.

[0040] Step 102: Generate hint point information according to the location information of the recognition object with the highest confidence level, and match a preset second model (generalization model) according to the numerical interval where the highest confidence level is located; Based on the hint point information and the second model, perform object recognition on the image to be recognized to obtain an extended recognition object with the same object type as the recognition object with the highest confidence level.

[0041] The present invention can design a generalization agent to implement the function of step 102.

[0042] First, obtain the location information of the ground object recognition object with the highest confidence level from the first model. If the location information is provided in the form of a mask, the mask indicates the specific location and contour of the object in the image.

[0043] Based on the above mask information, key points can be extracted therefrom as hint point information. These hint points may include the center point of the target, the corner points of the bounding box, or a series of points selected along the contour of the target. The role of the hint points is to provide a positioning reference for the subsequent model to help it more accurately identify targets of similar types.

[0044] Among them, the hint point information is generated according to the position information of the recognized target with the highest confidence, including: (1) According to the position information of the recognized target with the highest confidence, extract the outer contour; Among them, the position information can be given in the form of a mask, and the mask marks the area of the recognized target in the image to be recognized. From this mask, the outer contour of the recognized target can be extracted.

[0045] (2) Fit the outer contour with at least one rectangle, and set multiple generalized hint points inside each rectangle.

[0046] Once the outer contour is obtained, at least one rectangle can be used to fit this outer contour. One or more rectangles can be selected according to actual needs to cover the target area as accurately as possible. It should be noted that before fitting the outer contour, the outer contour can be smoothed.

[0047] Among them, setting multiple generalized hint points inside each rectangle can be: Set a generalized hint point at the center point of the rectangle; Set at least one generalized hint point on each side of the center point on the central axis in the long side direction of the rectangle.

[0048] (3) Screen all the generalized hint points, and retain the generalized hint points located inside the outer contour to generate hint point information.

[0049] After screening, the remaining hint points are the final hint point information. These hint points will be used to guide further image recognition tasks.

[0050] The present invention can pre-divide the confidence into several predefined intervals, and each interval corresponds to a second model with different accuracies or characteristics. Select the most suitable second model according to the interval where the confidence is located. For example, a higher confidence may mean that the target is already quite clear, so a model focusing on detailed feature analysis can be selected; while a lower confidence may require a model that can handle a larger range of uncertainties.

[0051] After selecting the appropriate second model, use the hint point information generated in the first step to guide the target recognition process of the model in the image to be recognized, and further improve the recognition range.

[0052] Step 103: Calculate the similarity between the image of any extended recognition target with a confidence level less than the first threshold and the sample images of the same type in the task type library. If the similarity is greater than the second threshold, retain the any extended recognition target; otherwise, eliminate the extended recognition target. Among them, the sample images in the task type library include the images corresponding to the recognition targets with the highest confidence levels output by the first model.

[0053] Similarly, the present invention can design a generalization agent to implement the function of Step 103.

[0054] The present invention first screens out the extended recognition targets with confidence levels less than the first threshold from the extended recognition targets output in Step 102; Then, for any extended recognition target with a confidence level less than the first threshold, calculate the similarity with the sample images of the same type in the task type library; Finally, screen the extended recognition targets with confidence levels less than the first threshold based on the magnitude of the similarity.

[0055] Among them, calculating the similarity between the image of any extended recognition target with a confidence level less than the first threshold and the sample images of the same type in the task type library includes: (1) Determine the feature extraction model according to the target type of the any extended recognition target; According to the target type of the extended recognition target, select a suitable feature extraction model. This feature extraction model should be able to effectively extract the features of a specific target type from the image.

[0056] (2) Use the feature extraction model to extract the first image features of the image of the any extended recognition target and the second image features of the sample images of the same type respectively; (3) Calculate the similarity between the first image features and the second image features.

[0057] The present invention can use the finally retained extended recognition targets and the extended recognition targets with confidence levels greater than or equal to the first threshold as the final output results.

[0058] Figure 1 is a schematic framework diagram of the target recognition system provided by the present invention. The following combines Figure 1 to introduce the target recognition system. The system includes: A type agent, which is used to parse the input instructions of the user, call at least one first model according to the parsing result to perform target recognition on the corresponding type of targets in the image to be recognized, and determine the recognition targets with the highest confidence levels output by each first model; A generalization agent is used to generate prompt point information based on the position information of the recognition target with the highest confidence, and match a preset second model according to the numerical interval where the highest confidence is located; It is also used to perform target recognition on the image to be recognized based on the prompt point information and the second model, so as to obtain an extended recognition target with the same target type as the recognition target with the highest confidence; An induction agent is used to calculate the similarity between the image of any extended recognition target with a confidence level less than the first threshold and the sample image of the same type in the task type library, and retain the any extended recognition target when the similarity is greater than the second threshold; Among them, the sample images in the task type library include the images corresponding to the recognition targets with the highest confidence output by the first model.

[0059] The following Figure 1 further describes the target recognition system provided by the present invention: The first part (type agent): The goal of the type agent is to "find the right one", mainly to extract the target of the specified ground object type with the highest accuracy from the remote sensing image. It mainly consists of three parts, namely, a semantic understanding module, a recognition model call module, and a result screening module.

[0060] The semantic understanding module refers to accurately understanding the user's intention from the user's instruction, and the user's instructions are as follows: "Please identify the dock from the picture" "Please find all buildings and bridges" "Extract storage tanks, farmland, cage aquaculture, water body targets".

[0061] This module is mainly implemented based on the understanding ability of the large language model. The conventional idea is to directly input the user's instruction into the large language model and let it automatically select the model to be called. In order to improve the understanding ability of the existing large language model in the context of intelligent recognition, the present invention proposes an optimization method of "prompt word template guidance + large model generation", which is specifically as follows: (1) Define an initial prompt word template, such as "Please select which model to call according to the user's instruction. Input instruction: {text}.".

[0062] (2) "{text}" is the user's input instruction, and the output result of the large language model is obtained and analyzed.

[0063] (3) Identify type deviation correction, that is, whether there is a deviation in the expression between the type keywords output by the large language model and the type names in the recognition type library. For example, for cage aquaculture, there is a deviation in the expression from the cage aquaculture in the recognition type library but the semantics are similar, and the type name in the recognition type library needs to be used for replacement.

[0064] (4)The recognition model calling module refers to calling the corresponding recognition model (the first model) for recognition according to the results output by the large language model. This process requires calling a single type or multiple types of recognition models.

[0065] (5)The result screening module refers to screening the results output by the large language model. For the input image, for the results output by each recognition model, the target with the highest confidence value is taken as the screening result of this type, and the screening results of all types are summarized, which is the output result of the first part.

[0066] Assume the input image is , and the recognition model to be called is , then the screening results can be expressed as follows:

[0067]

[0068] Among them, represents the recognition target object with the highest confidence value of category c , represents the output result of the first part, represents the output of the confidence value of the th model prediction target, represents the recognition model of category .

[0069] The second part (generalized agent): The goal of the generalized agent is to "find all", and identify all target type objects from the image. Its main work process is as follows: (1)According to the level of confidence value, the generalized models to be called by the generalized agent are divided into multiple ones, which are used to process the results of different recognition degrees of the type agents.

[0070] (2)First, generate generalized prompt points according to the target position of the recognition target with the highest confidence in the first part (as shown in Figure 2 , Figure 2 is the schematic diagram of the generalized prompt points provided by the present invention), and then input the generalized prompt points into the generalized model (the second model) to assist in generating more ground object targets of the same type. The present invention proposes a new strategy for generating generalized prompt points, and the method is as follows: 1) Extract its outer contour according to the target object contour mask; 2) Smooth the outer contour, and fit each part of the outer contour with a rectangle. Generate three points inside each rectangle, namely the centroid and the center points of the two short sides. The center point of the short side needs to be at a certain distance from the short side, and the specific value is determined during method debugging.

[0071] 3) Screen generalization hint points. The screening rule is whether the hint point is included in the contour mask of the target object. If not, it is excluded.

[0072] (3) Aggregate the results of various types of ground object targets to form an extended result. Each ground object target result includes a category, a mask, and a confidence value.

[0073] The third part (inductive agent): The goal of the inductive agent is to "fill in the gaps", screen the low-confidence results of the generalization agent, and further improve the recognition effect of this method. The work process of the inductive agent is as follows: (1) Collect typical sample data of different ground object types from the publicly available remote sensing sample set, including the corresponding pictures and labels, to jointly form a general ground object type library.

[0074] (2) Aggregate the general ground object type library and the output results of the type agent to jointly form a task ground object type library, so as to improve the diversity of different types of ground objects and avoid the lack of representativeness of the task type library caused by insufficient richness of ground object forms, backgrounds, etc., which affects the overall performance of this method.

[0075] (3) Screen out the results with a confidence level of the generalization agent less than the first threshold. Use the picture corresponding to this result as the input picture. By calling the feature extraction model of the corresponding type, calculate the similarity between the ground object picture corresponding to the input picture mask and the sample pictures in the task ground object type library. If the similarity is higher than the threshold and the matching ground object type is the same as the ground object type of the input picture, keep this result; otherwise, exclude this result. The formula description is as follows:

[0076]

[0077] Among them, represents the output result of the generalization agent with type c and serial number m ; represents the output result of the generalization agent with a confidence value less than thres 1 selected; represents The ground object picture intercepted by the mask corresponding to the element with serial number j and type p in; represents the feature extraction model of type p ; represents the similarity calculation function; represents the picture of the ground object type library sample with serial number k and type p ; represents the calculated similarity score. representation elements in which the similarity result calculated with the sample pictures in the task ground object type library is greater than the threshold thres 2

[0078] (4)Merge the result in (3) (i.e., the retained result) with the high-confidence results of the generalization agent (results with confidence values higher than the preset threshold) as the final output result.

[0079] To better highlight the technical advantages of the present invention, in the experiment, this method was compared with the latest remote sensing ground object recognition methods TransUnet, CM-UNet, and a single agent. Among them, the single agent mainly uses the type agent in the present invention without screening the highest confidence value. Table 1 shows the comparison results between the present invention and the single agent method.

[0080] Table 1 is a schematic table of the comparison results between the present invention and the single agent method

[0081] It can be seen from the table that this method achieves the best results in all indicators.

[0082] The present invention has the following beneficial effects compared with the prior art: (1)The present invention provides a target recognition method, which can be applied to the remote sensing intelligent recognition of ground object targets. By introducing multi-agent collaborative work, it effectively overcomes the problem of inaccurate recognition of traditional single agents or single models.

[0083] (2)The present invention proposes a method for constructing a type agent, which can understand user instructions and accurately extract high-precision ground object targets.

[0084] (3)The present invention proposes a method for constructing a generalization agent, which can make up for the problem of missed detection of traditional models.

[0085] (4)The present invention proposes a method for constructing an induction agent, which can effectively screen misrecognition results, thereby improving the precision rate of this technical method.

[0086] (5)The type agent, generalization agent, and induction agent in the present invention have strong flexibility and can be combined with the prior art methods as modules.

[0087] It should be noted that the target recognition system provided by the embodiments of the present invention can execute the target recognition method described in any of the above embodiments during specific operation, and this embodiment will not be elaborated herein.

[0088] Figure 3 is a schematic structural diagram of the electronic device provided by the present invention, as Figure 3As shown in the figure, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340. Among them, the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logical instructions in the memory 330 to execute the target recognition method.

[0089] In addition, when the logical instructions in the above-mentioned memory 330 are implemented in the form of software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, may be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0090] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the target recognition method provided in each of the above embodiments.

[0091] On yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the target recognition method provided in each of the above embodiments.

[0092] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the technical solution, in essence, or the part that contributes to the prior art, may be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disks, optical discs, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A target recognition method, characterized in that, It includes: Parse the input instruction of the user, call at least one first model according to the parsing result to perform target recognition on the corresponding type of target in the image to be recognized, and determine the recognition target with the highest confidence output by each first model; Generate prompt point information according to the position information of the recognition target with the highest confidence, and match a preset second model according to the numerical interval where the highest confidence is located; Based on the prompt point information and the second model, perform target recognition on the image to be recognized to obtain an extended recognition target with the same target type as the recognition target with the highest confidence; Calculate the similarity between the image of any extended recognition target with a confidence level less than the first threshold and the sample image of the same type in the task type library, and retain the any extended recognition target if the similarity is greater than the second threshold; Among them, the sample images in the task type library include the images corresponding to the recognition targets with the highest confidence output by the first model.

2. The target recognition method according to claim 1, characterized in that, Parse the input instruction of the user, and call at least one first model according to the parsing result, including: Perform text parsing on the input instruction based on a large language model to obtain at least one type keyword; Based on the type keyword, use semantic similarity to match the type name of the first model in the recognition type library; the recognition type library includes the type names of multiple first models; Call the first model based on the matched type name of the first model.

3. The target recognition method according to claim 1, characterized in that, Based on the type keyword, use semantic similarity to match the type name of the first model in the recognition type library, including: Use the CLIP multi-modal pre-training model to encode the type keyword and the type name in the recognition type library respectively into vector form; In the same vector space, calculate the similarity between the vector of the type keyword and the vector of the type name of each first model in the recognition type library; Determine the most matching type name of the first model according to the similarity.

4. The target recognition method according to claim 1, characterized in that, Generate prompt point information according to the position information of the recognition target with the highest confidence, including: Extract the outer contour according to the position information of the recognition target with the highest confidence; Use at least one rectangle to fit the outer contour, and set multiple generalization prompt points inside each rectangle; Screen all the generalization prompt points, and retain the generalization prompt points located inside the outer contour to generate prompt point information.

5. The target recognition method according to claim 1, characterized in that, Set multiple generalization prompt points inside each rectangle, including: Set a generalization prompt point at the center point of the rectangle; Set at least one generalization prompt point on each side of the center point on the central axis in the long side direction of the rectangle.

6. The target recognition method according to claim 1, characterized in that, Calculate the similarity between the image of any extended recognition target with a confidence level less than the first threshold and the sample image of the same type in the task type library, including: Determine the feature extraction model according to the target type of the any extended recognition target; Use the feature extraction model to extract the first image feature of the image of the any extended recognition target and the second image feature of the sample image of the same type respectively; Calculate the similarity between the first image feature and the second image feature.

7. The target recognition method according to claim 1, characterized in that, The image to be recognized is a remote sensing image.

8. A target recognition system, characterized in that, It includes: A type of intelligent agent is used to parse the input instructions of the user, call at least one first model according to the parsing result to perform object recognition on the corresponding type of object in the image to be recognized, and determine the recognized object with the highest confidence output by each first model; A generalization intelligent agent is used to generate prompt point information based on the position information of the recognized object with the highest confidence, and match a preset second model according to the numerical interval where the highest confidence is located; It is also used to perform object recognition on the image to be recognized based on the prompt point information and the second model to obtain an extended recognized object with the same object type as the recognized object with the highest confidence; An induction intelligent agent is used to calculate the similarity between the image of any extended recognized object with a confidence level less than the first threshold and the sample image of the same type in the task type library, and retain the any extended recognized object when the similarity is greater than the second threshold; Among them, the sample images in the task type library include the images corresponding to the recognized objects with the highest confidence output by the first model.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the object recognition method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the object recognition method according to any one of claims 1 to 7.

Images