CN117726588A - Image analysis method, device, electronic equipment and storage medium - Google Patents

Image analysis method, device, electronic equipment and storage medium Download PDF

Info

Publication number
CN117726588A
CN117726588A CN202311727625.3A CN202311727625A CN117726588A CN 117726588 A CN117726588 A CN 117726588A CN 202311727625 A CN202311727625 A CN 202311727625A CN 117726588 A CN117726588 A CN 117726588A
Authority
CN
China
Prior art keywords
image
computing node
analyzed
structural analysis
analysis result
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311727625.3A
Other languages
Chinese (zh)
Inventor
吴克虎
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Intellifusion Technologies Co Ltd
Original Assignee
Shenzhen Intellifusion Technologies Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Intellifusion Technologies Co Ltd filed Critical Shenzhen Intellifusion Technologies Co Ltd
Priority to CN202311727625.3A priority Critical patent/CN117726588A/en
Publication of CN117726588A publication Critical patent/CN117726588A/en
Pending legal-status Critical Current

Links

Classifications

    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Image Analysis (AREA)

Abstract

The embodiment of the invention provides an image analysis method, which comprises the following steps: acquiring a first image to be analyzed sent by a request terminal and presetting analysis requirements; performing first structural analysis processing on a first image to be analyzed through a first computing node in a first computing node layer to obtain a first structural analysis result; determining an optimal scheduling strategy of a second computing node layer based on a preset analysis requirement and a first structural analysis result, and generating a second image to be analyzed, a type to be analyzed and a target second computing node according to the optimal scheduling strategy of the second computing node layer; performing second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result; and determining a final structural analysis result based on the first structural analysis result and the second structural analysis result. By the method for analyzing the image, the efficiency of image analysis can be improved under the condition of ensuring the image analysis effect.

Description

Image analysis method, device, electronic equipment and storage medium
Technical Field
The present invention relates to the field of image processing, and in particular, to an image analysis method, an image analysis device, an electronic device, and a storage medium.
Background
With the continuous development of image analysis technology, the requirements on the accuracy and the efficiency of image analysis are higher and higher. In the existing image analysis technology, the full-scale structural analysis of one picture at a single computing node cannot be completed, and the full-scale structural analysis of one picture by using the single computing node cannot be completed according to the performance of the computing node and the difference of algorithm precision. Therefore, the conventional image analysis method has problems that the image is analyzed under a large data volume, the analysis strategy cannot be dynamically adjusted, and the analysis efficiency is low.
Disclosure of Invention
The embodiment of the invention provides an image analysis method, which aims to solve the problems that the existing image analysis method can analyze pictures under a large data volume, cannot dynamically adjust analysis strategies and has low analysis efficiency. Acquiring a first analysis request sent by a request terminal; performing first structural analysis processing on a first image to be analyzed through a first computing node in a first computing node layer to obtain a first structural analysis result; determining an optimal scheduling strategy of the second computing node layer based on a preset analysis requirement and a first structural analysis result, and generating an analysis request according to the optimal scheduling strategy of the second computing node layer; a second structural analysis processing is carried out on the analysis request through a target second computing node, so that a second structural analysis result is obtained; the final structural analysis result is determined based on the first structural analysis result and the second structural analysis result, and the final structural analysis result is returned to the request terminal, so that dynamic structural analysis adjustment can be performed on the diversified image data according to the method, and the image analysis efficiency is improved under the condition of ensuring the image analysis effect.
In a first aspect, an embodiment of the present invention provides an image parsing method, where the image parsing method is applied to an image parsing system, where the image parsing system includes a first computing node layer and a second computing node layer, the first computing node layer includes at least one first computing node, the second computing node layer includes a plurality of second computing nodes, and the first computing node layer is communicatively connected with the second computing node layer, and the method includes the following steps:
acquiring a first analysis request sent by a request terminal, wherein the first analysis request comprises a first image to be analyzed and a preset analysis requirement;
performing first structural analysis processing on a first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result;
determining an optimal scheduling policy of the second computing node layer based on the preset analysis requirement and the first structural analysis result, and generating an analysis request according to the optimal scheduling policy of the second computing node layer, wherein the analysis request comprises a second image to be analyzed, a type to be analyzed and a target second computing node, the second image to be analyzed is determined according to the optimal scheduling policy of the second computing node layer, and the plurality of second computing nodes comprise the target second computing node;
Performing second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result;
and determining a final structural analysis result based on the first structural analysis result and the second structural analysis result, and returning the final structural analysis result to the request terminal.
Optionally, performing, by a first computing node in the first computing node layer, a first structural analysis process on a first image to be resolved to obtain a first structural analysis result, where the first structural analysis result includes:
performing target detection on the first image to be analyzed through the first computing node to obtain target detection data;
extracting image parameters of the first image to be analyzed through the first computing node to obtain image parameters of the first image to be analyzed;
and determining the first structural analysis result based on the target detection data and the image parameters of the first image to be analyzed.
Optionally, the determining, based on the preset parsing requirement and the first structural parsing result, the optimal scheduling policy of the second computing node layer includes:
Determining the area of a second image to be resolved and the number of the second image to be resolved in the first image to be resolved according to the preset resolving requirement and the target detection data;
determining the scheduling cost of the second computing node layer based on the area of the second image to be analyzed and the number of the second images to be analyzed;
an optimal scheduling policy for the second layer of computing nodes is determined based on the scheduling cost.
Optionally, before determining the scheduling cost of the second computing node layer based on the area of the second image to be resolved and the number of the second images to be resolved, the method further includes:
determining an encoding cost factor for the first computing node;
determining a decoding cost factor for each of the second computing nodes in the second layer of computing nodes and an inference cost factor for the second computing node;
a transmission cost factor between the first computing node and each of the second computing nodes is determined.
Optionally, the determining the scheduling cost of the second computing node layer based on the area of the second image to be resolved and the number of the second images to be resolved includes:
Determining a scheduling cost of the second image to be resolved based on the area of the second image to be resolved, the number of the second images to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor and the transmission cost factor;
determining a scheduling cost of the first image to be resolved based on the image parameters of the first image to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor, and the transmission cost factor;
and determining the scheduling cost of the second computing node layer based on the scheduling cost of the second image to be analyzed and the scheduling cost of the first image to be analyzed.
Optionally, the obtaining, by the target second computing node, a second structured analysis result by performing a second structured analysis process on the analysis request includes:
performing target attribute detection on the second image to be analyzed through the target second computing node to obtain target attribute data;
and determining the second structural analysis result based on the target attribute data.
Optionally, the determining a final structural analysis result based on the first structural analysis result and the second structural analysis result, and returning the final structural analysis result to the request terminal includes:
Performing analysis polymerization treatment on the first structural analysis result and the second structural analysis result to obtain a final structural analysis result;
and carrying out data verification on the final structured analysis result, and returning the final structured analysis result passing the data verification to the request terminal.
In a second aspect, an embodiment of the present invention further provides an image analysis apparatus, including:
the first acquisition module is used for acquiring a first analysis request sent by the request terminal, wherein the first analysis request comprises a first image to be analyzed and a preset analysis requirement;
the first analysis module is used for carrying out first structural analysis processing on the first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result;
the first determining module is configured to determine an optimal scheduling policy of the second computing node layer based on the preset analysis requirement and the first structured analysis result, and generate an analysis request according to the optimal scheduling policy of the second computing node layer, where the analysis request includes a second image to be analyzed, a type to be analyzed, and a target second computing node, the second image to be analyzed is determined according to the optimal scheduling policy of the second computing node layer, and the plurality of second computing nodes includes the target second computing node;
The second analysis module is used for carrying out second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result;
and the second determining module is used for determining a final structural analysis result based on the first structural analysis result and the second structural analysis result and returning the final structural analysis result to the request terminal.
In a third aspect, an embodiment of the present invention provides an electronic device, including: the image analysis method comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps in the image analysis method provided by the embodiment of the invention when executing the computer program.
In a fourth aspect, embodiments of the present invention provide a computer readable storage medium having a computer program stored thereon, the computer program when executed by a processor implementing the steps in the image analysis method provided by the embodiments of the present invention.
In the embodiment of the invention, a first analysis request sent by a request terminal is obtained, wherein the first analysis request comprises a first image to be analyzed and a preset analysis requirement; performing first structural analysis processing on a first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result; determining an optimal scheduling policy of the second computing node layer based on the preset analysis requirement and the first structural analysis result, and generating an analysis request according to the optimal scheduling policy of the second computing node layer, wherein the analysis request comprises a second image to be analyzed, a type to be analyzed and a target second computing node, the second image to be analyzed is determined according to the optimal scheduling policy of the second computing node layer, and the plurality of second computing nodes comprise the target second computing node; performing second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result; and determining a final structural analysis result based on the first structural analysis result and the second structural analysis result, and returning the final structural analysis result to the request terminal. Analyzing a plurality of first images to be analyzed through a first computing node to obtain a first structural analysis result, determining an optimal scheduling strategy for scheduling to a target second computing node according to the first structural analysis result and a preset analysis requirement, generating an analysis request for scheduling to the second computing node according to the optimal scheduling strategy, performing second structural analysis processing on the second images to be analyzed through the target second computing node according to the analysis request to obtain a second structural analysis result, determining a final structural analysis result according to the first structural analysis result and the second structural analysis result, determining an optimal scheduling strategy according to the preset analysis requirement and the first structural analysis result, further analyzing the first images to be analyzed according to the scheduling strategy, improving the analysis precision, and simultaneously analyzing a plurality of image data through the first computing node, so that the image analysis efficiency can be improved.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a system architecture diagram of an image analysis system according to an embodiment of the present invention;
fig. 2 is a flowchart of an image parsing method according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an image analysis device according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, an embodiment of the present invention provides an image analysis system architecture, which includes: task scheduler, first level computing node, second level computing node.
The image analysis system may include, but is not limited to, a first computing node layer and a second computing node layer, where the task scheduler is configured to obtain a first analysis request sent by a request terminal and schedule the first analysis request to a first computing node, where the first computing node is configured to perform primary structural analysis on a first image to be analyzed in the first analysis request according to a preset analysis requirement, send the request to a second computing node, aggregate an analysis result of the second computing node with an analysis result of the first computing node, return the processed result to the task scheduler, and return the processed result to a display interface of the image analysis system.
The first computing node layer may include at least one first computing node, the second computing node layer may include a plurality of second computing nodes, and the first computing node layer and the second computing node layer are in communication connection. The first-stage computing node can analyze a plurality of images at the same time, when the analysis result obtained by the first-stage computing node can meet the preset analysis requirement, the request is not required to be made to the second-stage computing node, one second-stage computing node comprises a plurality of attribute clusters, one attribute cluster also comprises a plurality of child nodes, and the plurality of first-stage computing nodes can initiate the request to one second-stage computing node.
As shown in fig. 2, fig. 2 is a flowchart of an image analysis method according to an embodiment of the present invention, where the image analysis method includes the steps of:
201. and acquiring a first analysis request sent by the request terminal.
In an embodiment of the present invention, the image analysis method may be applied to the image analysis system, where the image analysis system may include, but is not limited to, a first computing node layer and a second computing node layer, where the first computing node layer includes at least one first computing node, and the second computing node layer includes a plurality of second computing nodes, and the first computing node layer and the second computing node layer are communicatively connected.
The first analysis request includes a first image to be analyzed and a preset analysis requirement, specifically, the first image to be analyzed may be an image to be analyzed input by the request terminal on the image analysis system, and the preset analysis requirement may be a degree of target type extraction or target type extraction of the first image to be analyzed by the request terminal.
The request terminal may be an electronic device connected to the image analysis system, for example, a smart phone, a smart camera, or the like.
In one possible embodiment, the request terminal may send an analysis request to the image analysis system through the request terminal, where the analysis request may include, but is not limited to, an analysis requirement of the request terminal on the image to be analyzed.
202. And carrying out first structural analysis processing on the first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result.
In the embodiment of the present invention, the first computing node layer includes at least one first computing node, and the first computing node may be a computing unit having an image computing capability in the image analysis system or an image computing node capable of computing and analyzing an image. The target first computing node may be a first computing node in an idle state, and in general, the image analysis system may select a first computing node in which both computing power and load are in a relatively idle state to perform a first structural analysis process on a first image to be analyzed, and one first computing node may perform a first structural analysis process on a plurality of first images to be analyzed.
Specifically, in one possible embodiment, the first computing node obtains the target type of the first image to be resolved by performing a structural analysis on the first image to be resolved. The structured analysis processing may refer to a process of extracting a type of an image to be analyzed by the image calculation node.
The target type may refer to an attribute feature of a specific object in the first image to be resolved, for example, a person, a motor vehicle, a non-motor vehicle, etc. in the first image to be resolved. Specifically, in one possible embodiment, by performing structural analysis on a first image to be analyzed to obtain an image of a person and an image of a car of a specific object in the first image to be analyzed, setting the person and the car as the target types, extracting the target types to obtain position information of the specific object, and marking the first image to be analyzed with a position frame according to the position information of the specific object to obtain a plurality of position frames of the specific object.
The first structural analysis result may include, but is not limited to, a position frame of the specific object in the first image to be analyzed, a target type, and an image parameter of the first image to be analyzed. The image parameters of the first to-be-resolved image may include, but are not limited to, a resolution of the first to-be-resolved image and a size of the first to-be-resolved image.
203. And determining an optimal scheduling strategy of the second computing node layer based on the preset analysis requirement and the first structural analysis result, and generating an analysis request according to the optimal scheduling strategy of the second computing node layer.
In the embodiment of the present invention, the second computing node layer includes at least one second computing node, where the second computing node may be a computing unit with an image computing capability in the image analysis system or an image computing node capable of computing and analyzing an image, and the second computing node may perform finer type feature extraction on the specific object, so as to improve analysis accuracy, for example, if the type feature extracted by the first computing node is a vehicle, the second computing node may extract feature attributes such as a license plate, a door, and a head of the vehicle.
The scheduling strategies comprise a large-image scheduling strategy and a small-image scheduling strategy, the large-image scheduling strategy is to transmit a first image to be resolved to a target second node for processing, the small-image scheduling strategy is to transmit a second image to be resolved to the target second node for processing, the second image to be resolved is at least one image obtained after feature extraction is carried out on the first image to be resolved, and the scheduling strategy with the minimum scheduling cost is determined to be the optimal scheduling strategy. That is, if the cost of the large map scheduling policy is less than the cost of the small map scheduling policy, the optimal scheduling policy is the large map scheduling policy, and if the cost of the large map scheduling policy is greater than the cost of the small map scheduling policy, the optimal scheduling policy is the small map scheduling policy.
The optimal scheduling policy may refer to a method of scheduling the result obtained by analyzing the first structural analysis result according to the preset analysis requirement to the second computing node layer and requiring the minimum scheduling resource. It should be noted that, the optimal scheduling policy is determined according to the image analysis calculation cost and the image transmission cost.
In the case that the optimal scheduling policy is a small-scale scheduling policy, the parsing request includes, but is not limited to, a second image to be parsed, a type to be parsed, and a target second computing node, where the type to be parsed may refer to an attribute feature of the specific object that is smaller in a range of fineness than the target type and is more relevant to the specific object, the one second computing node corresponds to the type to be parsed, the second image to be parsed may be determined according to an optimal scheduling policy of a second computing node layer, and the second image to be parsed may refer to an image obtained by the first computing node after performing a first structural parsing process on the first image to be parsed, where the first structural parsing result includes a plurality of second images to be parsed. The plurality of second computing nodes includes a target second computing node.
Specifically, the second target computing node may be determined by scheduling the second image to be resolved to the second computing node, and when the scheduling cost for scheduling the second image to be resolved to the second computing node is minimum, the second computing node is used as the target second computing node, and the type of the target second computing node is used as the scheduling type in the resolving request. The second image to be resolved may be the first structural analysis result that can be structurally resolved by the second computing node, that is, the first structural analysis result may include a plurality of second images to be resolved, and the second image to be resolved may be an image obtained after the first structural analysis result is extracted by the target feature.
In a possible embodiment, the second computing node layer performs analysis processing according to the preset analysis requirement and the first structural analysis result to obtain a second computing node capable of meeting the preset analysis requirement, determines a target second computing node according to parameters such as performance of the second computing node, and obtains an optimal scheduling policy according to minimum resource consumption of scheduling the second image to be analyzed to the target second computing node, and generates an analysis request according to the optimal scheduling policy. The performance parameters of the second computing node may include, but are not limited to, a storage speed, a processing speed, and a memory capacity of the second computing node, and generally, when the performance parameters of the second computing node are higher, that is, the higher the storage speed, the faster the processing speed, and the larger the memory capacity, the higher the processing capability of the second computing node on the image is. In one possible embodiment, a second computing node with higher image processing efficiency may be obtained as the target second computing node by performing balance computation on the storage speed, the processing speed, the memory capacity and the like of the second computing node.
In the case that the optimal scheduling policy is a large-graph scheduling policy, the resolution request includes a first image to be resolved, a type to be resolved and a target second computing node, and the determination mode of the type to be resolved and the target second computing stage is the same as the determination method in the case that the optimal scheduling policy is a small-graph scheduling policy.
In addition, whether the optimal scheduling policy is a large-graph scheduling policy or a small-graph scheduling policy, the method for determining the final structural analysis result is the same.
204. And carrying out second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result.
In the embodiment of the present invention, the second structural analysis result may include, but is not limited to, location information of the specific object in the second image to be analyzed, a type to be analyzed of the specific object, and a type value corresponding to the type to be analyzed of the specific object.
The type value corresponding to the type to be analyzed of the specific object may refer to a numerical description of the type to be analyzed of the specific object, for example, the type to be analyzed of the specific object is a color of clothes, and the type value corresponding to the type to be analyzed of the specific object may be a color feature such as red, yellow, etc.
In a possible embodiment, the target second computing node analyzes the analysis request to obtain a type to be analyzed, performs structural analysis on a second image to be analyzed according to the type to be analyzed to obtain position information of the specific object in the second image to be analyzed, and marks the type to be analyzed of the specific object and a type value corresponding to the type to be analyzed of the specific object in a corresponding second image to be analyzed.
205. And determining a final structured analysis result based on the first structured analysis result and the second structured analysis result, and returning the final structured analysis result to the request terminal.
In the embodiment of the present invention, the final structural analysis result may be a result image that the request terminal wants to analyze through the image analysis system. Specifically, the image to be analyzed may be input into the image analysis system according to an analysis requirement preset by the request terminal, and the image is finally output by the image analysis system. The image output by the image analysis system may be an image with a second image position frame to be analyzed, and the frame side of the second image position frame to be analyzed is provided with a target type and an identifier of the type to be analyzed, and a corresponding type value is attached behind the identifier of the corresponding target type and the identifier of the type to be analyzed.
In a possible embodiment, the image analysis system performs data verification and comparison on the obtained first structural analysis result and the obtained second structural analysis result, and performs aggregation processing on the first structural analysis result and the second structural analysis result which are successfully compared to obtain a final structural analysis result.
In the embodiment of the invention, a first analysis request sent by a request terminal is obtained, wherein the first analysis request comprises a first image to be analyzed and a preset analysis requirement; performing first structural analysis processing on a first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result; determining an optimal scheduling policy of the second computing node layer based on the preset analysis requirement and the first structural analysis result, and generating an analysis request according to the optimal scheduling policy of the second computing node layer, wherein the analysis request comprises a second image to be analyzed, a type to be analyzed and a target second computing node, the second image to be analyzed is determined according to the optimal scheduling policy of the second computing node layer, and the plurality of second computing nodes comprise the target second computing node; performing second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result; and determining a final structural analysis result based on the first structural analysis result and the second structural analysis result, and returning the final structural analysis result to the request terminal. The method comprises the steps of analyzing a plurality of first images to be analyzed through a first computing node to obtain a first structural analysis result, determining an optimal scheduling strategy for scheduling to a target second computing node according to the first structural analysis result and a preset analysis requirement, generating an analysis request for scheduling to the second computing node according to the optimal scheduling strategy, performing second structural analysis processing on the target second computing node according to the analysis request to obtain a second structural analysis result, determining a final structural analysis result according to the first structural analysis result and the second structural analysis result, determining the optimal scheduling strategy according to the preset analysis requirement and the first structural analysis result, further analyzing the first images to be analyzed according to the scheduling strategy, improving analysis accuracy, and simultaneously analyzing a plurality of image data through the first computing node, so that image analysis efficiency can be improved.
Optionally, in the step of performing a first structural analysis process on the first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result, the first computing node may perform target detection on the first image to be analyzed to obtain target detection data, then perform image parameter extraction on the first image to be analyzed through the first computing node to obtain image parameters of the first image to be analyzed, and finally determine the first structural analysis result based on the target detection data and the image parameters of the first image to be analyzed.
In the embodiment of the present invention, the target detection may be a process of extracting image features from the first image to be resolved according to the preset resolution requirement of the first computing node. The target detection data may refer to data including, but not limited to, the second image to be resolved, a position frame of the second image to be resolved in the first image to be resolved, and a target type of the second image to be resolved, which are obtained after the first computing node performs target detection on the first image to be resolved.
The image parameter extraction may refer to a process in which the first computing node performs image parameter calculation on the first image to be resolved through an image parameter calculation unit to obtain an image parameter. The image parameters may include, but are not limited to, resolution, size, etc. of the first image to be resolved.
In one possible embodiment, when the request terminal inputs the first image to be resolved through the image resolving system, the first computing unit of the image resolving system performs the object detection and the image parameter extraction operation on the first image to be resolved, so as to obtain the object detection data and the image parameter of the first image to be resolved, and takes the object detection data and the image parameter of the first image to be resolved as the first structural resolving result.
Optionally, in the step of determining the optimal scheduling policy of the second computing node layer based on the preset analysis requirement and the first structural analysis result, an area of the second image to be analyzed and a number of the second images to be analyzed may be determined in the first image to be analyzed according to the preset analysis requirement and the target detection data, then a scheduling cost of the second computing node layer may be determined based on the area of the second image to be analyzed and the number of the second images to be analyzed, and finally the optimal scheduling policy of the second computing node layer may be determined based on the scheduling cost.
In the embodiment of the present invention, the area of the second image to be resolved and the number of the second images to be resolved may be obtained after performing target detection on the first image to be resolved according to the first computing node. Specifically, after the target detection data is obtained, the size and the number of the second image position frames to be analyzed are obtained by analyzing and calculating the second image position frames to be analyzed in the target detection data.
The scheduling cost may refer to an amount of resources required for scheduling the second image to be parsed to the target second computing node and returning data responded by the target second computing node to the first computing node. In particular, the scheduling costs may include, but are not limited to, encoding costs, decoding costs, reasoning costs, transmission costs, and the like. The encoding cost may be determined according to the degree of influence of the encoding cost factor, the decoding cost may be determined according to the degree of influence of the decoding cost factor, the inference cost may be determined according to the degree of influence of the inference cost factor, and the transmission cost may be determined according to the degree of influence of the transmission cost factor.
In this embodiment, the optimal scheduling policy may further determine the total area of the second image to be resolved and the area of the first image to be resolved (i.e. B 0 ) The ratio and the number of the second images to be analyzed are determined. Specifically, the optimal scheduling policy may be determined by comparing the ratio of the areas with an area threshold, or by comparing the number of the second images to be parsed with a number threshold.
The total area of the second image to be analyzed may refer to an area for encoding the second image to be analyzed, and may be obtained by the following equation:
wherein Sum (B) is the total area of the second image to be analyzed, B n Is the area of one of the plurality of second images to be resolved. In one possible embodiment, when Sum (B)/B0 is greater than the area threshold or n is greater than the number threshold, the first image to be resolved is directly subjected to original image scheduling (i.e. large image scheduling mode) to be used as an optimal scheduling policy, and when Sum (B)/B0 is less than the area threshold and n is less than the number threshold, each second image to be resolved is scheduled (i.e. small image scheduling mode) to be used as the most optimal scheduling policy.
In a possible embodiment, the image analysis system determines, by using a scheduling method of scheduling the second image to be analyzed for each second computing node, calculates a scheduling cost of a large graph scheduling method and a scheduling cost of a small graph scheduling method, uses a scheduling method corresponding to a minimum scheduling cost as an optimal scheduling policy, uses a second computing node corresponding to the minimum scheduling cost as a target second computing node, and uses a type corresponding to the target second computing node as a scheduling type.
In one possible embodiment, the large graph scheduling cost and the small graph scheduling cost of each second computing node are calculated to obtain the resource amount for scheduling the image to the second computing node according to the large graph scheduling policy and the small graph scheduling policy, and the corresponding small graph scheduling policy or the large graph scheduling policy is used as the optimal scheduling policy and the second computing node is used as the target second computing node according to the minimum resource amount.
Optionally, in the step before determining the scheduling cost of the second computing node layer based on the area of the second image to be resolved and the number of the second images to be resolved, the encoding cost factor of the first computing node may also be determined; determining a decoding cost factor for each second computing node in the second layer of computing nodes and an inference cost factor for each second computing node; a transmission cost factor between the first computing node and the second computing node is determined.
In an embodiment of the present invention, the encoding cost factor may be determined by the computing performance of the first computing node, and in general, the stronger the computing performance of the first computing node, the smaller the encoding cost factor, which indicates that the encoding cost is lower, and when the more images need to be encoded and the greater the resolution, the greater the encoding cost factor, which indicates that the encoding cost is higher.
The decoding cost factor and the inference cost factor may be determined according to the computing performance and the deployment of the second computing node. Specifically, when the second computing node is more sensitive to the extraction of the attribute type, the second computing node is more powerful in reasoning capability, so that the influence of the corresponding reasoning cost factor is smaller, namely, the reasoning cost is lower. It should be noted that, when each second computing node performs the transmission cost calculation, the corresponding decoding cost factor and the inference cost factor need to be determined, and since the decoding cost factor and the inference cost factor are different according to the performance parameters of the second computing node, the decoding cost factor and the inference cost factor corresponding to each second computing node are also different.
The transmission cost factor may be determined according to the optimal scheduling policy, specifically, the scheduling method selected according to the optimal scheduling policy is different, and the selected image coding object is different, so that the parameters of the other party needing to perform transmission are different, that is, the sizes of the other party needing to perform transmission are different. In this embodiment, since the deployment and setting of the above-described image analysis system are determined, and thus the bandwidth in the transmission process is also fixed, the magnitude of the influence of the transmission cost factor depends on the size of the transmission object, that is, the larger the transmission object, the larger the influence of the transmission cost factor, that is, the higher the transmission cost.
Furthermore, in the process of cost calculation, the cost factors can be quantized, so that the purpose of facilitating cost calculation is achieved. In general, quantization units of different cost factors may be determined according to different parameters, specifically, different decoding cost factor units may be set according to different resolution levels, for example, the quantization unit of the decoding cost factor is set to 4 at a resolution level of 1920×1080.
In a possible embodiment, the transmission cost of the first computing node and the transmission cost of each second computing node in the second computing node layer may be obtained by calculating the encoding cost factor of the first computing node and the decoding cost factor and the inference cost factor of each second computing node in the second computing node layer.
Optionally, in the step of determining the scheduling cost of the second computing node layer based on the area of the second image to be resolved and the number of the second images to be resolved, the scheduling cost of the second image to be resolved may be determined based on the area of the second image to be resolved, the number of the second images to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor and the transmission cost factor, then the scheduling cost of the first image to be resolved is determined based on the image parameter, the encoding cost factor, the decoding cost factor, the reasoning cost factor and the transmission cost factor of the first image to be resolved, and finally the scheduling cost of the second computing node layer is determined based on the scheduling cost of the second image to be resolved and the scheduling cost of the first image to be resolved.
In the embodiment of the present invention, the scheduling cost of the second image to be resolved may be obtained according to the scheduling cost calculation method of the small image scheduling mode, and the scheduling cost of the first image to be resolved may be obtained according to the scheduling cost calculation method of the large image scheduling mode.
Specifically, the large-graph scheduling cost can be obtained according to the following formula:
cost_total1＝cost_transport1+cost_decode1+cost_infer1
cost_transport1＝transoprt_factor×S 1
cost_decode1＝decode_factor×A 1
cost_infer1＝infer_factor×A 1
wherein cost_total1 is the scheduling cost of the large-scale scheduling scheme, cost_transport1 is the transmission cost of the large-scale scheduling scheme, cost_decoding 1 is the decoding cost of the large-scale scheduling scheme, cost_infer1 is the reasoning cost of the large-scale scheduling scheme, S 1 A is the size of the area of the first image to be analyzed 1 The resolution of the first image to be resolved. According to the above formula, the scheduling cost of the second image to be analyzed to the target second computing node is scheduled in a large-graph scheduling mode.
The above-described graph scheduling cost may be obtained according to the following equation:
cost_total2＝cost_encode_total+cost_transport_total+cost_decode_total+cost_infer_total
cost_encode_total＝encode_factor×N
cost_transport1＝transoprt_factor×N
cost_decode1＝decode_factor×N
cost_infer1＝infer_factor×N
the cost_total2 is a scheduling cost of the small-image scheduling method, the cost_encode_total is a coding cost of the small-image scheduling method, the cost_transport1 is a transmission cost of the small-image scheduling method, the cost_decode1 is a decoding cost of the small-image scheduling method, and the cost_refer 1 is an inference cost of the small-image scheduling method. N is the number of second images to be resolved to be scheduled to the target second computing node. According to the above formula, the scheduling cost of the second image to be analyzed to the target second computing node is scheduled by a small graph scheduling mode.
In one possible embodiment, each of the second computing nodes needs to calculate the cost_total1 and the cost_total2 according to the above formula, so as to obtain the scheduling cost of the corresponding scheduling mode, and the second computing node with the minimum scheduling cost is taken as the target second computing node.
In a possible embodiment, in the large-scale scheduling manner, when the first computing node performs structural analysis on the first image to be analyzed to obtain a plurality of second images to be analyzed, position information corresponding to each second image to be analyzed is recorded, and ID of each second image to be analyzed is allocated and cached in a preset cache structure. When a request is initiated to a second computing node according to a target type corresponding to a certain second image to be analyzed, searching is carried out in the preset cache structure. If the cache data of the target type corresponding to the certain second image to be analyzed is found, the cache data is used as the target type of the current request image; if the cache data of the target type corresponding to the certain second to-be-resolved image is not found, searching the target type in a second structural resolving result of structural resolving of the second computing node, and performing coincidence degree calculation on a plurality of second to-be-resolved images obtained by resolving the second computing node in the large image and a plurality of second to-be-resolved images obtained by resolving the first computing node in the large image, wherein in general, the coincidence degree calculation can be obtained by calculating the coincidence degree of a position frame of a plurality of second to-be-resolved images obtained by resolving the second computing node in the large image and a position frame of a plurality of second to-be-resolved images obtained by resolving the first computing node in the large image, when the coincidence degree is higher than a coincidence degree threshold, judging that a plurality of second to-be-resolved images obtained by resolving the second computing node in the large image and a plurality of second to-be-resolved images obtained by resolving the first computing node in the large image are the same target, and discarding the plurality of second to-be-resolved images obtained by the second computing node in the large image.
In another possible embodiment, in the process of performing the scheduling cost calculation by using the small-image scheduling manner, in order to ensure that an error caused by matting cannot occur in the calculation of the second image to be resolved by the second computing node, so as to ensure that the calculated scheduling cost is close to the actual situation, it is generally necessary to perform a pixel expansion operation on the second image to be resolved, so as to ensure that the resolving error cannot be caused by information loss when the second computing node performs the structural resolving on the second image to be resolved. Specifically, the position frame of the second image to be analyzed can be converted through a preset offset vector, so that the position frame is outwards expanded by pixels with a certain area on the basis of the position frame of the second image to be analyzed, a small picture after the pixels are outwards expanded is obtained, and finally, the scheduling cost calculation is carried out on the small picture after the pixels are outwards expanded.
Optionally, in the step of obtaining the second structural analysis result by performing the second structural analysis processing on the analysis request through the target second computing node, the target attribute detection may be performed on the second image to be analyzed through the target second computing node to obtain target attribute data, and then the second structural analysis result is determined based on the target attribute data.
In the embodiment of the present invention, the target attribute detection may refer to a process of extracting an attribute of the second image to be resolved by the target second computing node, and specifically, the target second computing node extracts an attribute of the second image to be resolved by an attribute extracting unit, so as to obtain the target attribute data. The target attribute data may refer to data obtained by performing a second structural analysis on the second image to be analyzed by the target second computing node, including, but not limited to, attribute values of a plurality of the second images to be analyzed, position information of the attribute values of the plurality of the second images to be analyzed in the second images to be analyzed, a corresponding position frame obtained based on position information of the attribute values of the plurality of the second images to be analyzed in the second images to be analyzed, and confidence degrees of the obtained attribute values of the plurality of the second images to be analyzed.
In a possible embodiment, the image analysis system performs, through the target second computing node, a coincidence degree calculation on the image in the second structural analysis result and the image in the first structural analysis result, discards the image in the second structural analysis result with a coincidence degree higher than a coincidence degree threshold, specifically, performs a one-by-one calculation on a position frame of the image in the second structural analysis result and a position frame of the image in the first structural analysis result, discards the image in the second structural analysis result with a coincidence degree higher than the coincidence degree threshold if the calculated coincidence degree is higher than the coincidence degree threshold, and uses the target attribute data subjected to the coincidence degree calculation as the second structural analysis result. The above-described overlap threshold may be adjusted according to particular embodiments.
Optionally, in the step of determining a final structured analysis result based on the first structured analysis result and the second structured analysis result and returning the final structured analysis result to the request terminal, the first structured analysis result and the second structured analysis result may be further subjected to analysis and aggregation processing to obtain a final structured analysis result, then the final structured analysis result is subjected to data verification, and the final structured analysis result passing the data verification is returned to the request terminal.
In the embodiment of the present invention, the parsing and aggregation process may refer to a process of reflecting the target attribute data obtained by the second computing node on the second image to be parsed in the first structural parsing result.
The data verification may refer to a process of obtaining whether the final structural analysis result can meet the requirement of human eye recognition by performing image parameter processing on the obtained final structural analysis result.
In one possible embodiment, the image analysis system performs analysis aggregation processing on the first structural analysis result and the second structural analysis result, so as to obtain a plurality of final structural analysis results, performs image parameter analysis on each final structural analysis result, packages the final structural analysis results which can be identified by human eyes, and sends the packaged final structural analysis results to a display interface of the image analysis system for a request terminal to query.
As shown in fig. 3, an embodiment of the present invention further provides an image analysis device, which is characterized by including:
the first obtaining module 301 is configured to obtain a first resolution request sent by a request terminal, where the first resolution request includes a first image to be resolved and a preset resolution requirement;
the first parsing module 302 is configured to perform a first structural parsing process on a first image to be parsed through a first computing node in the first computing node layer, so as to obtain a first structural parsing result;
a first determining module 303, configured to determine an optimal scheduling policy of the second computing node layer based on the preset parsing requirement and the first structured parsing result, and generate a parsing request according to the optimal scheduling policy of the second computing node layer, where the parsing request includes a second image to be parsed, a type to be parsed, and a target second computing node, the second image to be parsed is determined according to the optimal scheduling policy of the second computing node layer, and the plurality of second computing nodes includes the target second computing node;
the second parsing module 304 is configured to perform a second structural parsing process on the second image to be parsed based on the type to be parsed and the target second computing node, so as to obtain a second structural parsing result;
And the second determining module 305 is configured to determine a final structural analysis result based on the first structural analysis result and the second structural analysis result, and return the final structural analysis result to the request terminal.
Optionally, the first parsing module 302 includes:
the first detection submodule is used for carrying out target detection on the first image to be analyzed through the first computing node to obtain target detection data;
the first extraction submodule is used for extracting image parameters of the first image to be analyzed through the first computing node to obtain image parameters of the first image to be analyzed;
and the first determination submodule is used for determining the first structural analysis result based on the target detection data and the image parameters of the first image to be analyzed.
Optionally, the first determining module 303 includes:
the second determining submodule is used for determining the area of a second image to be analyzed and the number of the second image to be analyzed in the first image to be analyzed according to the preset analysis requirement and the target detection data;
a third determining submodule, configured to determine a scheduling cost of the second computing node layer based on an area of the second image to be resolved and a number of the second image to be resolved;
And a fourth determination submodule, configured to determine an optimal scheduling policy of the second computing node layer based on the scheduling cost.
Optionally, the apparatus further includes:
a fourth determining module configured to determine an encoding cost factor for the first computing node;
a fifth determining module configured to determine a decoding cost factor of each of the second computing nodes and an inference cost factor of the second computing node in the second computing node layer;
a sixth determination module is configured to determine a transmission cost factor between the first computing node and each of the second computing nodes.
Optionally, the sixth determining module further includes:
a fifth determining submodule, configured to determine a scheduling cost of the second image to be resolved based on an area of the second image to be resolved, a number of the second image to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor, and the transmission cost factor;
a sixth determining submodule, configured to determine a scheduling cost of the first image to be resolved based on the image parameter of the first image to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor, and the transmission cost factor;
And a seventh determining submodule, configured to determine a scheduling cost of the second computing node layer based on the scheduling cost of the second image to be resolved and the scheduling cost of the first image to be resolved.
Optionally, the second parsing module 305 includes:
the first acquisition submodule is used for carrying out target attribute detection on the second image to be analyzed through the target second computing node to obtain target attribute data;
an eighth determination submodule is used for determining the second structural analysis result based on the target attribute data.
Optionally, the second determining module 305 further includes:
the second acquisition submodule is used for carrying out analysis and aggregation treatment on the first structural analysis result and the second structural analysis result to obtain a final structural analysis result;
and the verification sub-module is used for carrying out data verification on the final structural analysis result and returning the final structural analysis result which passes the data verification to the request terminal.
As shown in fig. 4, an embodiment of the present invention further provides an electronic device, including a processor, where the processor may execute any one of the image analysis methods described above.
Specifically, the image analysis method includes a processor 401, a memory 402, and a computer program stored in the memory 402 and capable of executing the image analysis method on the processor 401, wherein:
The processor 401 runs a computer program of the image analysis method stored in the memory 402, and performs the steps of:
acquiring a first analysis request sent by a request terminal, wherein the first analysis request comprises a first image to be analyzed and a preset analysis requirement;
performing first structural analysis processing on a first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result;
determining an optimal scheduling policy of the second computing node layer based on the preset analysis requirement and the first structural analysis result, and generating an analysis request according to the optimal scheduling policy of the second computing node layer, wherein the analysis request comprises a second image to be analyzed, a type to be analyzed and a target second computing node, the second image to be analyzed is determined according to the optimal scheduling policy of the second computing node layer, and the plurality of second computing nodes comprise the target second computing node;
performing second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result;
and determining a final structural analysis result based on the first structural analysis result and the second structural analysis result, and returning the final structural analysis result to the request terminal.
Optionally, the executing, by the processor 401, the first structural analysis processing on the first image to be resolved by using a first computing node in the first computing node layer to obtain a first structural analysis result includes:
performing target detection on the first image to be analyzed through the first computing node to obtain target detection data;
extracting image parameters of the first image to be analyzed through the first computing node to obtain image parameters of the first image to be analyzed;
and determining the first structural analysis result based on the target detection data and the image parameters of the first image to be analyzed.
Optionally, the processor 401 executes the determining, based on the preset parsing requirement and the first structural parsing result, an optimal scheduling policy of the second computing node layer, including:
determining the area of a second image to be resolved and the number of the second image to be resolved in the first image to be resolved according to the preset resolving requirement and the target detection data;
determining the scheduling cost of the second computing node layer based on the area of the second image to be analyzed and the number of the second images to be analyzed;
An optimal scheduling policy for the second layer of computing nodes is determined based on the scheduling cost.
Optionally, before the processor 401 executes the determining the scheduling cost of the second computing node layer based on the area of the second image to be resolved and the number of the second images to be resolved, the method further includes:
determining an encoding cost factor for the first computing node;
determining a decoding cost factor for each of the second computing nodes in the second layer of computing nodes and an inference cost factor for the second computing node;
a transmission cost factor between the first computing node and each of the second computing nodes is determined.
Optionally, the determining, by the processor 401, the scheduling cost of the second computing node layer based on the area of the second image to be resolved and the number of the second images to be resolved includes:
determining a scheduling cost of the second image to be resolved based on the area of the second image to be resolved, the number of the second images to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor and the transmission cost factor;
determining a scheduling cost of the first image to be resolved based on the image parameters of the first image to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor, and the transmission cost factor;
And determining the scheduling cost of the second computing node layer based on the scheduling cost of the second image to be analyzed and the scheduling cost of the first image to be analyzed.
Optionally, the processor 401 executes the second structural analysis processing performed on the analysis request by the target second computing node to obtain a second structural analysis result, including:
performing target attribute detection on the second image to be analyzed through the target second computing node to obtain target attribute data;
and determining the second structural analysis result based on the target attribute data.
Optionally, the processor 401 executes the determining a final structural analysis result based on the first structural analysis result and the second structural analysis result, and returns the final structural analysis result to the request terminal, including:
performing analysis polymerization treatment on the first structural analysis result and the second structural analysis result to obtain a final structural analysis result;
and carrying out data verification on the final structured analysis result, and returning the final structured analysis result passing the data verification to the request terminal.
The embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements each process of the image analysis method or the application-side image analysis method provided by the embodiment of the invention, and can achieve the same technical effect, so that repetition is avoided and no further description is given here.
Those skilled in the art will appreciate that the processes implementing all or part of the methods of the above embodiments may be implemented by a computer program for instructing relevant hardware, and the program may be stored in a computer readable storage medium, and the program may include the processes of the embodiments of the methods as above when executed. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM) or the like.
The foregoing disclosure is illustrative of the present invention and is not to be construed as limiting the scope of the invention, which is defined by the appended claims.

Claims (10)

1. An image parsing method, wherein the image parsing method is applied to an image parsing system, the image parsing system includes a first computing node layer and a second computing node layer, the first computing node layer includes at least one first computing node, the second computing node layer includes a plurality of second computing nodes, the first computing node layer and the second computing node layer are in communication connection, the method includes the following steps:
Acquiring a first analysis request sent by a request terminal, wherein the first analysis request comprises a first image to be analyzed and a preset analysis requirement;
performing first structural analysis processing on a first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result;
determining an optimal scheduling policy of the second computing node layer based on the preset analysis requirement and the first structural analysis result, and generating an analysis request according to the optimal scheduling policy of the second computing node layer, wherein the analysis request comprises a second image to be analyzed, a type to be analyzed and a target second computing node, the second image to be analyzed is determined according to the optimal scheduling policy of the second computing node layer, and the plurality of second computing nodes comprise the target second computing node;
performing second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result;
and determining a final structural analysis result based on the first structural analysis result and the second structural analysis result, and returning the final structural analysis result to the request terminal.
2. The method of image analysis according to claim 1, wherein the performing, by the first computing node in the first computing node layer, a first structural analysis process on the first image to be analyzed to obtain a first structural analysis result includes:
performing target detection on the first image to be analyzed through the first computing node to obtain target detection data;
extracting image parameters of the first image to be analyzed through the first computing node to obtain image parameters of the first image to be analyzed;
and determining the first structural analysis result based on the target detection data and the image parameters of the first image to be analyzed.
3. The image parsing method of claim 2, wherein determining an optimal scheduling policy for the second computing node layer based on the preset parsing requirement and the first structured parsing result comprises:
determining the area of a second image to be resolved and the number of the second images to be resolved in the first image to be resolved according to the preset resolving requirement and target detection data in the first structural resolving result;
determining the scheduling cost of the second computing node layer based on the area of the second image to be analyzed and the number of the second images to be analyzed;
An optimal scheduling policy for the second layer of computing nodes is determined based on the scheduling cost.
4. The image parsing method according to claim 3, wherein before determining the scheduling cost of the second computing node layer based on the area of the second image to be parsed and the number of the second images to be parsed, the method further comprises:
determining an encoding cost factor for the first computing node;
determining a decoding cost factor for each of the second computing nodes in the second layer of computing nodes and an inference cost factor for the second computing node;
a transmission cost factor between the first computing node and each of the second computing nodes is determined.
5. The image parsing method according to claim 4, wherein the determining the scheduling cost of the second computing node layer based on the area of the second image to be parsed and the number of the second images to be parsed includes:
determining a scheduling cost of the second image to be resolved based on the area of the second image to be resolved, the number of the second images to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor and the transmission cost factor;
Determining a scheduling cost of the first image to be resolved based on the image parameters of the first image to be resolved, the encoding cost factor, the decoding cost factor, the reasoning cost factor, and the transmission cost factor;
and determining the scheduling cost of the second computing node layer based on the scheduling cost of the second image to be analyzed and the scheduling cost of the first image to be analyzed.
6. The image parsing method according to claim 1, wherein the obtaining, by the target second computing node, a second structured parsing result by performing a second structured parsing process on the parsing request includes:
performing target attribute detection on the second image to be analyzed through the target second computing node to obtain target attribute data;
and determining the second structural analysis result based on the target attribute data.
7. The image parsing method according to claim 1, wherein determining a final structured parsing result based on the first structured parsing result and the second structured parsing result, and returning the final structured parsing result to the requesting terminal, includes:
Performing analysis polymerization treatment on the first structural analysis result and the second structural analysis result to obtain a final structural analysis result;
and carrying out data verification on the final structured analysis result, and returning the final structured analysis result passing the data verification to the request terminal.
8. An image analysis device, characterized in that the image analysis device comprises:
the first acquisition module is used for acquiring a first analysis request sent by the request terminal, wherein the first analysis request comprises a first image to be analyzed and a preset analysis requirement;
the first analysis module is used for carrying out first structural analysis processing on the first image to be analyzed through a first computing node in the first computing node layer to obtain a first structural analysis result;
the first determining module is configured to determine an optimal scheduling policy of the second computing node layer based on the preset analysis requirement and the first structured analysis result, and generate an analysis request according to the optimal scheduling policy of the second computing node layer, where the analysis request includes a second image to be analyzed, a type to be analyzed, and a target second computing node, the second image to be analyzed is determined according to the optimal scheduling policy of the second computing node layer, and the plurality of second computing nodes includes the target second computing node;
The second analysis module is used for carrying out second structural analysis processing on the second image to be analyzed based on the type to be analyzed and the target second computing node to obtain a second structural analysis result;
and the second determining module is used for determining a final structural analysis result based on the first structural analysis result and the second structural analysis result and returning the final structural analysis result to the request terminal.
9. An electronic device, comprising: memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in the image analysis method according to any one of claims 1 to 7 when the computer program is executed.
10. A computer-readable storage medium, on which a computer program is stored, which when being executed by a processor implements the steps in the image parsing method according to any one of claims 1 to 7.
CN202311727625.3A 2023-12-14 2023-12-14 Image analysis method, device, electronic equipment and storage medium Pending CN117726588A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311727625.3A CN117726588A (en) 2023-12-14 2023-12-14 Image analysis method, device, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311727625.3A CN117726588A (en) 2023-12-14 2023-12-14 Image analysis method, device, electronic equipment and storage medium

Publications (1)

Publication Number Publication Date
CN117726588A true CN117726588A (en) 2024-03-19

Family

ID=90206541

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311727625.3A Pending CN117726588A (en) 2023-12-14 2023-12-14 Image analysis method, device, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN117726588A (en)

Similar Documents

Publication Publication Date Title
CN111798400A (en) Non-reference low-illumination image enhancement method and system based on generation countermeasure network
CN115914649B (en) Data transmission method and system for medical video
Zuo et al. Screen content image quality assessment via convolutional neural network
CN111491167B (en) Image encoding method, transcoding method, device, equipment and storage medium
CN113837942A (en) Super-resolution image generation method, device, equipment and storage medium based on SRGAN
CN110225342B (en) Video coding bit distribution system and method based on semantic distortion measurement
CN112448827A (en) Information processing method, device, equipment and computer readable storage medium
CN117726588A (en) Image analysis method, device, electronic equipment and storage medium
CN117237867A (en) Self-adaptive field monitoring video target detection method and system based on feature fusion
CN112862095A (en) Self-distillation learning method and device based on characteristic analysis and readable storage medium
CN114169513B (en) Neural network quantization method and device, storage medium and electronic equipment
CN110458851A (en) Infrared Image Processing Method, device, computer equipment and storage medium
CN113808157B (en) Image processing method and device and computer equipment
WO2021237727A1 (en) Method and apparatus of image processing
CN115631457A (en) Man-machine cooperation abnormity detection method and system in building construction monitoring video
CN112333539B (en) Video real-time target detection method, terminal and server under mobile communication network
US11170260B2 (en) Techniques for determining importance of encoded image components for artificial intelligence tasks
CN110505480B (en) Monitoring scene-oriented fast perception video coding method
CN114549352A (en) Image processing method, image processing device, electronic equipment and storage medium
CN114241253A (en) Model training method, system, server and storage medium for illegal content identification
CN111815658B (en) Image recognition method and device
CN117581263A (en) Multi-modal method and apparatus for segmentation and depth estimation
CN114913513A (en) Method and device for calculating similarity of official seal images, electronic equipment and medium
CN111340137A (en) Image recognition method, device and storage medium
CN110189272B (en) Method, apparatus, device and storage medium for processing image

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination