Intelligent community task report generation method and device and related components

Abstract

The embodiment of the invention provides an intelligent community task report generation method and device and a related component, and the method comprises the steps: obtaining a monitoring picture through the shooting of a camera, then detecting the monitoring picture through a trained target detection model, obtaining a task key frame, transmitting the task key frame to a multi-mode large model, and carrying out the analysis, therefore, transaction and suggested action data of the community task are obtained, a preset response strategy is triggered for matching, and a task report is generated. According to the embodiment of the invention, a worker does not need to carry out real-time monitoring, the community affair needing to be processed can be obtained only according to the task report, and the efficiency and accuracy of community management are improved.

Description

Technical Field

[0001] The present invention relates to the field of community management technology, and in particular to a method and device for generating a smart community task report and related components. Background Art

[0002] In the past, community security prevention and control and management work mainly relied on real-time video monitoring by staff to respond to and handle various matters. However, this method has many problems. For example, during the monitoring process, staff need to stare at the monitoring screen for a long time, which not only affects the staff's eyesight, but also easily leads to fatigue and distraction, thus affecting the monitoring effect.

[0003] Moreover, as the size of the community expands and the number of surveillance cameras increases, the workload of monitoring also increases, requiring more staff to monitor, which increases labor costs. In addition, manual monitoring is inefficient because staff need to manually handle various matters, such as receiving, verifying and processing alarm information, which makes the entire process cumbersome and time-consuming, which not only reduces the response speed of security work, but also increases the risk of errors. Summary of the invention

[0004] The embodiments of the present invention provide a method for generating a smart community task report, a decision-making device and related components, aiming to solve the problem that the existing community management increases labor costs and is inefficient through manual monitoring.

[0005] In a first aspect, an embodiment of the present invention provides a method for generating a smart community task report, comprising: Pre-build an AI-Agent system and use it to manage community tasks; Selecting a target camera according to the type of the community task, and obtaining a surveillance image captured by the target camera; Use the trained target detection model to detect the corresponding monitoring screen and obtain the task key frame; Analyze the task keyframes using a multimodal large model to obtain transaction and recommended action data for the community task; The transaction and recommended action data are matched with the preset response strategy to generate a task report.

[0006] In a second aspect, an embodiment of the present invention provides a smart community task report generation device, including: Construction unit, used to pre-build AI-Agent system and use AI-Agent system to manage community tasks; A picture acquisition unit, used to select a target camera according to the type of the community task, and acquire the monitoring picture taken by the target camera; A detection unit, used to detect the corresponding monitoring picture using the trained target detection model to obtain a task key frame; An analysis unit, configured to analyze the task key frames using a multimodal large model to obtain transaction and recommended action data for the community task; The matching unit is used to match the transaction and recommended action data with a preset response strategy to generate a task report.

[0007] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the smart community task report generation method as described above when executing the computer program.

[0008] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for generating a smart community task report as described above is implemented.

[0009] The embodiment of the present invention provides a method, device and related components for generating a smart community task report. The method obtains a monitoring screen by shooting with a camera, then uses a trained target detection model to detect the monitoring screen to obtain a task key frame, and then transmits the task key frame to a multimodal large model for analysis, thereby obtaining the affairs and recommended action data of the community task, and then triggering a preset response strategy for matching and generating a task report. In this embodiment, the staff does not need to monitor in real time, but can obtain the community affairs to be processed based on the task report, thereby improving the efficiency and accuracy of community management. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0011] Figure 1 A flowchart of a method for generating a smart community task report provided by an embodiment of the present invention; Figure 2 A schematic diagram of a sub-process of a method for generating a smart community task report provided by an embodiment of the present invention; Figure 3 A schematic diagram of another sub-process of a method for generating a smart community task report provided by an embodiment of the present invention; Figure 4A schematic diagram of another sub-process of a method for generating a smart community task report provided by an embodiment of the present invention; Figure 5 A schematic block diagram of a smart community task report generating device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0014] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0015] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0016] See also Figure 1 , an embodiment of the present invention provides a method for generating a smart community task report, comprising steps S10-S50: S10. Pre-build an AI-Agent system and use it to manage community tasks; In this step, the pre-built AI-Agent system is an artificial intelligence system that can perceive the environment, make decisions and take actions based on the decisions. In the AI-Agent system, community tasks need to be classified and managed. Specifically, community tasks can be divided into community management, road management, elevator management, fire management and underground garage management. In community management, it is necessary to pay attention to the passage of people, such as whether there is congestion, etc. In road management, it is necessary to pay attention to whether there are vehicles on the road that are illegally parked or in other situations. In elevator management, it is necessary to pay attention to whether there are electric vehicles entering the elevator. In fire management, it is necessary to pay attention to whether there are fire extinguishers in the corresponding location. In underground garage management, it is necessary to pay attention to whether the vehicle is parked in the specified area.

[0017] In the specific implementation, the AI-Agent system is constructed as follows: the AI-Agent system is based on a multimodal large model and is equipped with a planning component, a memory component and a tool component, wherein the planning component is used for target decomposition and reflection and improvement, the memory component is divided into short-term memory and long-term memory, the multimodal large model uses short-term memory for learning, and uses long-term memory to achieve long-term retention of information, usually using external vectors for storage and retrieval, and the tool component is used to call external plug-ins to obtain additional information of the multimodal large model.

[0018] S20, selecting a target camera according to the type of the community task, and obtaining a monitoring image captured by the target camera; In step S10, the community tasks have been classified, and cameras are installed at corresponding locations according to the events that need to be paid attention to. Multiple cameras are monitored in real time, and the positions of each camera are managed by the AI-Agent system. In this step, when one type of community task is selected, it is only necessary to select the target camera according to the type of community task to view and obtain the monitoring images taken by the target camera for subsequent processing. Among them, the camera includes a ball camera and a gun camera. The ball camera has the functions of 360-degree rotation and up and down tilt. The gun camera is set in a fixed position and its viewing angle is fixed. In the specific implementation, the appropriate camera can be selected for installation according to the events that need to be paid attention to in the community task so that the area to be monitored can be covered, thereby obtaining an accurate monitoring image.

[0019] S30, using the trained target detection model to detect the corresponding monitoring picture to obtain the task key frame; In this step, the detection target is customized in advance according to the community task, and then the corresponding monitoring screen is detected according to the detection target through the trained target detection model, so as to obtain the task key frame. For example, when paying attention to the traffic of people, the detection target is pedestrians, and the frame image in which pedestrians are detected is the task key frame. When paying attention to the parking situation of road vehicles, the detection target is vehicles, and the frame image in which vehicles are detected is the task key frame. When paying attention to the situation of elevators, the detection target is electric vehicles, and the frame image in which electric vehicles are detected is the task key frame. When conducting fire inspections, the detection target is fire extinguishers, and the frame image in which fire extinguishers are detected is the task key frame. When paying attention to underground garages, the detection target is vehicles, and the frame image in which vehicles are detected is the task key frame. It should be noted that when paying attention to the parking situation of road vehicles and underground garages, the detection targets are all vehicles, but because of different scenes, the positions of the detected vehicles are also different, and then the task key frames of the predetermined number of frames are collected.

[0020] The training process of the target detection model is as follows: Figure 2 As shown, steps S31-S32 are included: S31, using the MobileNetv4 network as the backbone network of the original target detection model, using the PAN network as the neck network of the original target detection model, adding a detection head as the second detection head on the basis of the original detection head, and introducing an inverted bottleneck search module into the MobileNetv4 network to obtain an adjusted target detection model; S32, pre-setting initial training parameters of the adjusted target detection model, importing the pre-collected data set into the adjusted target detection model for training, and obtaining a trained target detection model.

[0021] In this embodiment, in order to enable the target detection model to accurately detect in practical applications and obtain the required task key frames, it is necessary to first train the original target detection model to improve the detection accuracy of the target detection model. Specifically, the original target detection model is a YOLOV10 model, which includes a backbone network, a neck network and a detection head. The backbone network adopts a MobileNetv4 network to extract the main features of the target image. An inverted bottleneck search module is introduced in the MobileNetv4 network. The inverted bottleneck search module is composed of an inverted bottleneck, a convolutional model (ConvNext), a feedforward network and an additional deep convolution variant fusion. The inverted bottleneck is used to perform spatial mixing on extended feature activation to provide a larger model capacity at a higher cost; ConvNext is used to perform spatial mixing before extending the features to allow cheaper spatial mixing using a larger kernel size; the feedforward network is used for stacking two point-by-point convolutions, which is conducive to acceleration; the additional deep convolution is used to increase the network depth and receptive field. By introducing the inverted bottleneck search module, the search effectiveness of the MobileNetv4 network is improved.

[0022] The neck network adopts PAN network to accurately preserve spatial information, which helps to correctly locate the pixels of the target image and form a mask.

[0023] On the basis of the detection head (i.e., the original detection head), another detection head (i.e., the second detection head) is added, wherein the original detection head is a one-to-many detection head, and the second detection head is a one-to-one detection head. The structure of the second detection head is the same as that of the original detection head, and the same optimization goal is adopted, but the second detection head adopts a one-to-one matching strategy, and the original detection head adopts a one-to-many matching strategy. During the training process, the two detection heads are jointly optimized with the target detection model, so that the backbone network and the neck network can enjoy the rich supervision signals provided by the one-to-many allocation. During the reasoning process, the one-to-many detection head (i.e., the original detection head) needs to be discarded, and only the one-to-one detection head (i.e., the second detection head) needs to be used for prediction, thereby avoiding non-maximum suppression, speeding up the calculation speed while maintaining the performance, reducing the computing resource overhead, and improving the reasoning efficiency.

[0024] By adjusting the backbone network, neck network and detection head of the original target detection model, an adjusted target detection model is obtained. During the training process, the adjusted target detection model is mainly trained through the data set. The data set needs to be collected in advance. A predetermined number of images, such as ten thousand, are collected according to the type of community task, and the ratio of positive and negative samples is guaranteed to be 1:1.

[0025] In one embodiment, if Figure 3 As shown, S32 includes: S321, using the label assignment method of the original detection head and the second detection head to assign labels to the data set to obtain classification labels and regression labels, and using the second detection head to predict the data set; S322, evaluating the original detection head and the second detection head using a matching metric to obtain a degree of consistency between the prediction and the instance; S323, calculating the classification loss and the regression loss according to the loss function, and performing back propagation according to the classification loss and the regression loss; S324: After a predetermined number of training rounds, a target detection model corresponding to the model weights is obtained.

[0026] In this embodiment, during the training process, two detection heads (i.e., the original detection head and the second detection head) are required to assign labels to the data set to obtain classification labels and regression labels. During the inference process, only the second detection head is required to predict the data set. During the label assignment process, both detection heads need to use matching metrics to evaluate the consistency between the prediction and the instance, which can be evaluated as follows: .

[0027] in, is the matching metric, and is a hyperparameter, s represents the spatial prior, p is the classification score, and b represent the bounding boxes of the prediction and instance respectively, and IoU is the original IoU value, which represents the area intersection-over-union ratio of the instance box A and the prediction box B: .

[0028] During the training process, it is also necessary to calculate the loss of the target detection model in order to obtain a model with better weights for saving. Specifically, the regression loss and classification loss are calculated through the loss function. The regression loss is a positioning loss function. By calculating the offset from the center point of the anchor point to the upper left corner and the lower right corner, the backbone network and the neck network can quickly focus on learning. The probability of values ​​near the continuous position of the bounding box (i.e., the prediction box) is defined by the following formula: : ; .

[0029] Among them, yi and yi+1 represent the probabilities of the i-th and i+1-th regression labels, yi+1 represents the probability of the i+1-th regression label, y is the regression label, Si is the probability of the value of yi near the continuous position of the bounding box, and Si+1 is the probability of the value of yi+1 near the continuous position of the bounding box.

[0030] In the prior art, the positioning loss function is CIOU loss, which only calculates the positioning loss of positive samples. Therefore, in order to more comprehensively calculate the loss of positive and negative samples, the positioning loss function is improved to Focaler-IoU, so that different regression samples can be focused in different detection tasks, so that the bounding box regression effect is improved. The formula is: .

[0031] in, represents Focaler-IoU, , u and d are adjustable parameters, and IoU refers to the above formula. Among them, u and d are set by default, d=0.00, u=0.95, and then by adjusting the values ​​of d and u, Focus on different regression samples and finally regression loss Defined as: .

[0032] Next, calculate the classification loss BCE Loss. The classification loss only calculates the classification loss of positive samples and is calculated according to the following formula: .

[0033] Among them, N is the training set (the training set is obtained by dividing the data set according to a predetermined ratio), and each training set has n classification labels. is an activation function that can map x to the interval (0, 1).

[0034] In addition, when training the adjusted target detection model, it is necessary to pre-set the training parameters, which include: learning rate, weight decay parameter, and data enhancement parameter. The specific settings are as follows: learning rate is 0.01, weight decay parameter is 0.0005, and data enhancement parameter (i.e., mosaic) is 1.0. Then, the pre-collected data set is sent to the adjusted target detection model for training. After a predetermined number of training rounds (such as 300 rounds), the best model weights can be obtained, and the target detection model corresponding to the model weights is saved. This model is the trained target detection model.

[0035] S40, analyzing the task key frames using a multimodal large model to obtain transaction and recommended action data for the community task; In this step, the task keyframe obtained in step S30 is sent to the multimodal large model for analysis, so as to obtain the transaction (i.e., what happened) and recommended action data (i.e., the processing method generated based on what happened) of the community task. The staff can quickly locate the target area and process it according to the recommended action data, thereby improving the efficiency and accuracy of community management.

[0036] In a specific implementation, the multimodal large model is a llava-onevision model.

[0037] The llava-onevision model can process single, multiple images and videos, and allows migration across community tasks in different scenarios. The llava-onevision model includes a large language model, a visual decoder and a projector. The large language model is the Qwen-2 model, which is used to provide different model weights and language capabilities. The visual encoder uses the SigLIP encoder with a high level of performance. The projector uses 2 multi-layer perceptrons to project image features into the word vector space and generate visual tokens.

[0038] In the specific implementation, the llava-onevision model also needs fine-tuning training, such as Figure 4 As shown, fine-tune the training as follows: S41, screening target images and videos from a pre-collected data set, and dividing the screened videos according to a predetermined number of frames to obtain a plurality of video frames, taking the plurality of video frames and target images as a total sample and dividing them into a training set, a test set, and a validation set; S42, presetting the training rounds, learning rate and learning rate decay of the llava-onevision model, and training the llava-onevision model using the training set, and testing and verifying the model using the test set and the verification set, respectively; S43, saving the llava-onevision model after the predetermined rounds of training, and deploying the saved llava-onevision model.

[0039] In step S41, the predetermined number of frames can be 16 frames, and then the multiple video frames and target images are converted into the format of image-text pairs, and then the multiple video frames and target images are used as total samples to divide the training set, test set and verification set, and the total samples are divided into positive and negative samples, and the ratio of positive and negative samples is 1:1.

[0040] In step S42, the training parameters of the llava-onevision model are pre-set, wherein the training parameters include training rounds, learning rate, and learning rate decay parameter. The training rounds can be specifically set to 3, the learning rate to 0.0001, and the learning rate decay parameter to 0. The training set is sent to the llava-onevision model for training, and the test set and the verification set are sent to the llava-onevision model for testing and verification.

[0041] In step S43, after training to the predetermined training rounds, the training is stopped, and then the trained llava-onevision model is saved and deployed. The llava-onevision model uses the API interface to receive requests and returns analysis results based on the prompt results set for each community task. If a keyword is triggered, the AI-agent system calls the tool component to execute.

[0042] S50: Match the transaction and recommended action data with a preset response strategy to generate a task report.

[0043] In this step, different response strategies are set according to the type of community task. Therefore, it is necessary to match the transaction and recommended action data with the preset response strategies in the corresponding community task to generate a task report. The staff can handle the events that need to be paid attention to in the community task based on the task report.

[0044] After S50, including: Receive the task items added by users to the community tasks and the adjusted task execution strategies through the UI interface; The task report is displayed through the UI interface.

[0045] In this step, the user can design the task parameters and view the task report in the UI interface. The task parameter design function can flexibly add task items and adjust the task execution strategy. When viewing the task report, the task report includes the task items, event descriptions, and recommended action guidelines. The staff can take timely actions based on the task report to improve the operational efficiency of the entire community work. In addition, during the specific implementation, the AI-Agen system is continuously updated and optimized based on historical task data and task reports.

[0046] The embodiment of the present invention also provides a smart community task report generation device. Figure 5 , Figure 5 600 is a schematic block diagram of a smart community task report generation device provided by an embodiment of the present invention. The smart community task report generation device 600 includes: A construction unit 610 is used to pre-construct an AI-Agent system and use the AI-Agent system to manage community tasks; The picture acquisition unit 620 is used to select a target camera according to the type of the community task and acquire the monitoring picture taken by the target camera; A detection unit 630 is used to detect the corresponding monitoring picture using the trained target detection model to obtain a task key frame; An analysis unit 640 is used to analyze the task keyframes using a multimodal large model to obtain transaction and recommended action data for the community task; The matching unit 650 is used to match the transaction and recommended action data with a preset response strategy to generate a task report.

[0047] In one embodiment, the construction unit 610 includes: The AI-Agent system is based on a multimodal large model and includes a planning component, a memory component and a tool component. The planning component is used for target decomposition and reflection and improvement, the memory component is divided into short-term memory and long-term memory, and the tool component is used to call an external plug-in to obtain additional information of the multimodal large model.

[0048] In one embodiment, the detection unit 630 includes: The training process of the target detection model is: The MobileNetv4 network is used as the backbone network of the original target detection model, the PAN network is used as the neck network of the original target detection model, and a detection head is added as the second detection head on the basis of the original detection head, and an inverted bottleneck search module is introduced into the MobileNetv4 network to obtain an adjusted target detection model; Initial training parameters of the adjusted target detection model are preset, and the pre-collected data set is imported into the adjusted target detection model for training to obtain a trained target detection model.

[0049] In one embodiment, the importing of the pre-collected data set into the adjusted object detection model for training includes: Using the label assignment method of the original detection head and the second detection head to assign labels to the data set, obtaining classification labels and regression labels, and using the second detection head to predict the data set; evaluating the original detection head and the second detection head using a matching metric to obtain a degree of consistency between the prediction and the instance; Calculate the classification loss and the regression loss according to the loss function, and perform back propagation according to the classification loss and the regression loss; After training for a predetermined number of rounds, an object detection model corresponding to the model weights is obtained.

[0050] In one embodiment, the analysis unit 640 includes: The multimodal large model is a llava-onevision model.

[0051] In one embodiment, the llava-onevision model is fine-tuned and trained as follows: Filter target images and videos from a pre-collected data set, and divide the filtered videos according to a predetermined number of frames to obtain multiple video frames, and use the multiple video frames and target images as a total sample and divide them into a training set, a test set, and a validation set; Presetting the training rounds, learning rate and learning rate decay of the llava-onevision model, training the llava-onevision model using the training set, and testing and verifying the model using the test set and verification set, respectively; Save the llava-onevision model after training for a predetermined number of rounds, and deploy the saved llava-onevision model.

[0052] In one embodiment, the matching unit 650 then includes: Receive the task items added by users to the community tasks and the adjusted task execution strategies through the UI interface; The task report is displayed through the UI interface.

[0053] An embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for generating a smart community task report as described in the above embodiment is implemented.

[0054] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for generating a smart community task report as described in the above embodiment is implemented.

[0055] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for generating a smart community task report, characterized in that: include: Pre-build an AI-Agent system and use it to manage community tasks; Selecting a target camera according to the type of the community task, and obtaining a surveillance image captured by the target camera; Use the trained target detection model to detect the corresponding monitoring screen and obtain the task key frame; Analyze the task keyframes using a multimodal large model to obtain transaction and recommended action data for the community task; The transaction and recommended action data are matched with the preset response strategy to generate a task report.

2. The method for generating a smart community task report according to claim 1, characterized in that: The multimodal large model is a llava-onevision model.

3. The method for generating a smart community task report according to claim 2, characterized in that: The llava-onevision model is fine-tuned and trained as follows: Filter target images and videos from a pre-collected data set, and divide the filtered videos according to a predetermined number of frames to obtain multiple video frames, and use the multiple video frames and target images as a total sample and divide them into a training set, a test set, and a validation set; Presetting the training rounds, learning rate and learning rate decay of the llava-onevision model, training the llava-onevision model using the training set, and testing and verifying the model using the test set and verification set, respectively; Save the llava-onevision model after training for a predetermined number of rounds, and deploy the saved llava-onevision model.

4. The method for generating a smart community task report according to claim 1, characterized in that: The AI-Agent system is based on a multimodal large model and includes a planning component, a memory component and a tool component. The planning component is used for target decomposition and reflection and improvement, the memory component is divided into short-term memory and long-term memory, and the tool component is used to call an external plug-in to obtain additional information of the multimodal large model.

5. The method for generating a smart community task report according to claim 1, characterized in that: The training process of the target detection model includes: The MobileNetv4 network is used as the backbone network of the original target detection model, the PAN network is used as the neck network of the original target detection model, and a detection head is added as the second detection head on the basis of the original detection head, and an inverted bottleneck search module is introduced into the MobileNetv4 network to obtain an adjusted target detection model; Initial training parameters of the adjusted target detection model are preset, and the pre-collected data set is imported into the adjusted target detection model for training to obtain a trained target detection model.

6. The method for generating a smart community task report according to claim 5, characterized in that: The step of importing the pre-collected data set into the adjusted target detection model for training includes: Using the label assignment method of the original detection head and the second detection head to assign labels to the data set, obtaining classification labels and regression labels, and using the second detection head to predict the data set; evaluating the original detection head and the second detection head using a matching metric to obtain a degree of consistency between the prediction and the instance; Calculate the classification loss and the regression loss according to the loss function, and perform back propagation according to the classification loss and the regression loss; After training for a predetermined number of rounds, an object detection model corresponding to the model weights is obtained.

7. The method for generating a smart community task report according to claim 1, characterized in that: After the step of matching the transaction and recommended action data with the preset response strategy and generating a task report, the following steps are included: Receive the task items added by users to the community tasks and the adjusted task execution strategies through the UI interface; The task report is displayed through the UI interface.

8. A smart community task report generation device, used to implement the smart community task report generation method according to any one of claims 1 to 7, characterized in that: include: Construction unit, used to pre-build AI-Agent system and use AI-Agent system to manage community tasks; A picture acquisition unit, used to select a target camera according to the type of the community task, and acquire the monitoring picture taken by the target camera; A detection unit, used to detect the corresponding monitoring picture using the trained target detection model to obtain a task key frame; An analysis unit, configured to analyze the task key frames using a multimodal large model to obtain transaction and recommended action data for the community task; The matching unit is used to match the transaction and recommended action data with a preset response strategy to generate a task report.

9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for generating a smart community task report as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the method for generating a smart community task report according to any one of claims 1 to 7 is implemented.

Images