Model-driven cloud edge-end interconnection integration method, device, equipment and medium

Abstract

The invention relates to the technical field of model-driven cloud end-to-end interconnection integration, and discloses a model-driven cloud end-to-end interconnection integration method and device, equipment and a medium, and the method comprises the steps: obtaining a convolutional neural network; removing a redundant channel of the convolutional neural network based on an attention mechanism module to obtain a target convolutional neural network; and carrying out calculation based on the target convolutional neural network and a cloud edge end interconnection system calculation node to obtain a calculation result. According to the method, after redundant channels of the convolutional neural network are removed, the convolutional neural network is put to different levels of the cloud edge end interconnection system to operate, the resource advantages of the edge and the cloud end are fully utilized, and the related calculation of the convolutional neural network is accelerated.

Description

Technical Field

[0001] The present invention relates to the field of cloud-edge-terminal interconnection integration technology, and in particular to a model-driven cloud-edge-terminal interconnection integration method, device, equipment and medium. Background Art

[0002] Edge computing is a distributed computing architecture that moves computing and data storage resources from central cloud data centers to places closer to data sources. Edge nodes are usually deployed close to data sources or terminal devices, such as routers, base stations, or gateways, to reduce data transmission delays and improve real-time processing capabilities.

[0003] In edge computing, the computing and storage resources of edge nodes are usually more limited than those of cloud data centers. This means that there are limitations when running complex deep learning models or other tasks that require high computing resources on edge nodes. This limitation is mainly reflected in the following aspects: first, computing power: edge nodes are usually equipped with processors with weak performance and cannot handle a large number of complex computing tasks like cloud servers; second, storage space: the storage capacity of edge nodes is also relatively limited, and cannot store large amounts of data or bulky models. This limits the ability of edge nodes to process large-scale data sets or store multiple large models; third, energy consumption limitations: edge devices are often limited by energy consumption and need to find a balance between power consumption and performance. High-performance computing and storage devices usually consume more electricity and are impractical in many edge application scenarios. Summary of the invention

[0004] Based on this, it is necessary to propose a model-driven cloud-edge-end interconnection integration method, device, equipment and medium to address the technical problems in the existing technology that due to limited computing and storage resources, the model may be large in magnitude and require a large amount of resources.

[0005] In a first aspect, a model-driven cloud-edge-device interconnection integration method is provided, the method comprising:

[0006] Get a convolutional neural network;

[0007] Based on the attention mechanism module, redundant channels of the convolutional neural network are removed to obtain a target convolutional neural network;

[0008] Calculations are performed based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system to obtain calculation results.

[0009] In a second aspect, a model-driven cloud-edge-end interconnection integration device is provided, the device comprising:

[0010] An acquisition module, used to acquire a convolutional neural network;

[0011] A redundant channel removal module, used for removing redundant channels of the convolutional neural network based on the attention mechanism module to obtain a target convolutional neural network;

[0012] The computing module is used to perform calculations based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnected system to obtain calculation results.

[0013] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned model-driven cloud-edge-terminal interconnection integration method when executing the computer program.

[0014] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned model-driven cloud-edge-terminal interconnection integration method are implemented.

[0015] The model-driven cloud-edge-end interconnection integration method proposed in the present invention obtains a convolutional neural network, and then removes the redundant channels of the convolutional neural network based on the attention mechanism module to obtain a target convolutional neural network. Finally, the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system are used to perform calculations to obtain calculation results. The present invention can remove the redundant channels of the convolutional neural network and put it into different levels of the cloud-edge-end interconnection system for operation, making full use of the resource advantages of the edge and the cloud, and speeding up the related calculations of the convolutional neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] in:

[0018] Figure 1 An application environment diagram of a model-driven cloud-edge-end interconnection integration method in one embodiment;

[0019] Figure 2 A flowchart of a model-driven cloud-edge-end interconnection integration method in one embodiment;

[0020] Figure 3 A schematic diagram of a convolutional neural network structure of a model-driven cloud-edge-device interconnection integration method in one embodiment;

[0021] Figure 4 It is a structural block diagram of a model-driven cloud-edge-end interconnection integration device in one embodiment;

[0022] Figure 5 is a structural block diagram of a computer device in one embodiment;

[0023] Figure 6 It is a structural block diagram of a computer device in another embodiment. DETAILED DESCRIPTION

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0025] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0026] 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.

[0027] The model-driven cloud-edge-end interconnection integration method provided in the embodiment of the present invention can be applied in the following aspects: Figure 1In an application environment, the client 110 communicates with the server 120 through a network. The server 120 can receive the convolutional neural network through the client 110, and then remove the redundant channels of the convolutional neural network based on the attention mechanism module to obtain the target convolutional neural network, and finally calculate based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system to obtain the calculation results. The present invention can remove the redundant channels of the convolutional neural network and put it into different levels of the cloud-edge-end interconnection system for operation, making full use of the resource advantages of the edge and the cloud, and speeding up the relevant calculations of the convolutional neural network. Among them, the client 110 can be but is not limited to various personal computers, laptops, smart phones, tablets and portable wearable devices. The server 120 can be implemented with an independent server or a server cluster consisting of multiple servers. The present invention is described in detail below through specific embodiments.

[0028] See also Figure 2 As shown, Figure 2 A flow chart of a model-driven cloud-edge-device interconnection integration method provided in an embodiment of the present invention includes the following steps:

[0029] Step S101: obtaining a convolutional neural network;

[0030] Step S102: Based on the attention mechanism module, redundant channels of the convolutional neural network are removed to obtain a target convolutional neural network;

[0031] In this embodiment, an attention mechanism module can be added after each convolutional layer in the convolutional neural network, and the channel weight vector output by the attention mechanism module is sparsified, so that some lower channel weight vectors are approached to 0, so that the channels corresponding to the channel weight vectors approaching 0 are used as redundant channels, and the redundant channels are removed from the convolutional neural network to obtain the target convolutional neural network.

[0032] In one embodiment, the step of removing redundant channels of the convolutional neural network based on the attention mechanism module to obtain a target convolutional neural network includes:

[0033] Step S1021: adding an attention mechanism module to the convolutional layer to be streamlined in the convolutional neural network to obtain a convolutional neural network after the addition;

[0034] As an example, a convolutional neural network includes two sequentially connected convolutional layers 1 and 2. Then, an attention mechanism module is added between convolutional layer 1 and convolutional layer 2 to obtain sequentially connected convolutional layer 1, attention mechanism module, and convolutional layer 2.

[0035] Step S1022: performing network training on the added convolutional neural network, and obtaining the weights of the channels of the convolutional layer in the forward process of the network training;

[0036] Step S1023: Based on the weight of the channel, the sparsity of the channel is calculated, and the redundant channels are determined based on the sparsity, and the redundant channels are removed from the added convolutional neural network to obtain the trained convolutional neural network.

[0037] As an example, the weight corresponding to each channel feature is obtained through the attention mechanism module (Attention Block) in the branch. The output of the attention mechanism module is the channel weight vector. By introducing the sparsification of the weight vector in the attention mechanism module, some lower channel weight vectors are approached to 0. Then, the original output feature of the convolution layer will be multiplied by the weight vector at the channel level, and the final feature map will be used as the input feature map of the subsequent network layer. After sparsification, the output features of the channel with lower weights will not be effectively passed to the next convolution layer.

[0038] refer to Figure 3 During the training process, the feature maps are input into each sub-convolutional layer in the convolutional layer 1 of the added convolutional neural network, and a sub-convolutional layer is selected as the target convolutional layer. The first feature maps output by the sub-convolutional layers other than the target convolutional layer in the convolutional layer 1 are input into the attention mechanism module. The output of the attention module is the channel weight vector. Further, by introducing the sparsification of the weight vector in the attention branch, some lower weights are approached to 0. The output features of the target convolutional layer will be multiplied by the weight vector at the channel level, and the final feature map will be used as the input feature map of the convolutional layer 2. After sparsification, the output features of the channels with lower weights will not be effectively passed to the next convolutional layer 2, so these channels can be safely removed from the convolutional neural network, thereby streamlining the network. The process of the model pruning algorithm based on the attention mechanism is summarized as follows: (1) Add an attention branch to the convolutional layer to be streamlined in the target network; (2) Obtain the channel weights in the forward process of network training, calculate the channel sparsity to perform sparse selection, retain the channels corresponding to more important features, and remove redundant channels from the network; (3) Alternately train and prune the network until the expected pruning target is met, and finally remove the attention branch and continue to train the pruned network until it is fully converged.

[0039] Step S1024: Determine whether the trained convolutional neural network meets the preset pruning target;

[0040] In one embodiment, it is determined whether the training times of the trained convolutional neural network reaches a preset number of times. If so, the preset pruning target is met.

[0041] In another embodiment, it is determined whether the trained convolutional neural network converges. If so, the preset pruning target is met.

[0042] Step S1025: If satisfied, the trained convolutional neural network is used as the target convolutional neural network.

[0043] Step S103: Perform calculations based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system to obtain calculation results.

[0044] In one embodiment, the task program corresponding to the target convolutional neural network is divided into various program blocks; based on each of the program blocks and a preset unloading decision, the program blocks are sent to their respective corresponding cloud-edge-end interconnection system computing nodes for calculation, wherein the cloud-edge-end interconnection system computing nodes perform calculations based on the program blocks to obtain calculation results, and send the calculation results to the user terminal.

[0045] Specifically, (1) Node discovery: Find available cloud-edge-end interconnected system computing nodes for subsequent computing of the uninstalled program. These nodes can be high-performance servers located in a remote cloud computing center or cloud-edge-end interconnected system servers located at the edge of the network. (2) Program segmentation: Divide the task program to be processed, and try to maintain the functional integrity of each part of the program during the segmentation process to facilitate subsequent uninstallation. (3) Uninstallation decision: Uninstallation decision is the most core link in computing offloading. This link mainly solves two major problems: decide whether to uninstall the program? And which parts of the program should be uninstalled to the cloud-edge-end interconnected system computing nodes? Uninstallation strategies can be divided into dynamic unloading and static unloading: the strategy of determining all the program blocks to be uninstalled before executing the uninstallation is a static unloading strategy; the strategy of dynamically planning the uninstallation of the program based on the actual influencing factors during the uninstallation process is a dynamic unloading strategy. (4) Program transmission: After the mobile terminal makes an uninstallation decision, it can hand over the divided computing program to the cloud for execution. There are many ways to transmit the program, which can be transmitted through 3G / 4G / 5G networks or through Wi-Fi. The purpose of program transmission is to transfer the offloaded computing program to the computing node of the cloud-edge-end interconnected system. (5) Execute calculation: The execution mainly adopts the virtual machine solution. After the mobile terminal offloads the computing task to the cloud, the cloud starts a virtual machine for the task, and then the task resides in the virtual machine for execution, and the user end does not feel any change. The computing node of the cloud-edge-end interconnected system calculates the program offloaded to the server. (6) Return of calculation results: The return of calculation results is the last link in the calculation offloading process. The results of the calculation processing performed by the computing node of the cloud-edge-end interconnected system are transmitted back to the user's mobile device terminal. At this point, the calculation offloading process ends and the mobile terminal is disconnected from the cloud.

[0046] It should be noted that the process of computation offloading will be affected by different factors, such as the operation habits of the terminal machine, the communication conditions of the radio channel, the quality of the backhaul connection, the performance of the mobile device, and the availability of the cloud server. Therefore, the key to computation offloading is to specify appropriate offloading decisions. At present, the performance of computation offloading is usually measured by time delay and energy consumption. The calculation of time delay and energy consumption is specifically divided into the following two cases. (1) When computation offloading is not performed, time delay refers to the time spent on performing local computation at the mobile device terminal; energy consumption refers to the energy consumed by performing local computation at the mobile device terminal. (2) When computation offloading is performed, time delay refers to the sum of the transmission time of the offloaded data to the computing node of the cloud-edge-end interconnection system, the execution processing time at the computing node of the cloud-edge-end interconnection system, and the transmission time of the data results processed by the computing node of the cloud-edge-end interconnection system; energy consumption refers to the sum of the transmission energy consumption of the offloaded data to the computing node of the cloud-edge-end interconnection system and the transmission energy consumption of the data results processed by the computing node of the cloud-edge-end interconnection system.

[0047] In addition, the offloading decision is the terminal's decision on whether to offload and how much to offload. The terminal consists of a code parser, a system parser, and a decision engine. There are three steps to make an offloading decision: first, the code parser determines which tasks can be assisted based on the application type and code / data partition; then the system parser is responsible for monitoring various parameters, such as available bandwidth, the size of the data to be offloaded, or the energy consumed by executing local applications; finally, the decision engine determines whether to offload. Generally speaking, there are three solutions for making decisions about computing offloading. 1. Local execution: The entire calculation is completed locally on the terminal; 2. Full offloading: The entire calculation is offloaded and processed by the cloud-edge-end interconnection system; 3. Partial offloading: Part of the calculation is processed locally, while the other part is offloaded to the cloud-edge-end interconnection system server for processing. The factors affecting these three decisions are mainly the energy consumption of the terminal and the delay in completing the computing task. The offloading decision needs to consider the computing delay factor, because the delay will affect the user's experience and may cause the coupled program to fail to run normally due to the lack of the calculation results of this section. Therefore, all offloading decisions must at least meet the time delay limit that the mobile device program can accept.

[0048] Finally, energy consumption must be considered. If energy consumption is too high, the battery of the mobile device terminal will be exhausted quickly. Minimizing energy consumption means minimizing the energy consumption value under the constraint of satisfying the delay condition. For some applications, if it is not necessary to minimize one of the delay or energy indicators, different weighted values ​​can be assigned to the delay and energy consumption indicators according to the specific needs of the program to minimize the sum of the two values, that is, minimize the total cost. This is called the unloading decision that maximizes benefits.

[0049] After the offloading decision is made, the next step is to make reasonable computing resource allocation. Similar to the computing decision, the choice of server-side computing execution location will be affected by whether the application can be split for parallel computing. If the application does not meet the split and parallel computing requirements, only one physical node can be allocated for this calculation. On the contrary, if the application is splittable and supports parallel computing, the offloading program can be distributed on multiple virtual machine nodes for computing.

[0050] The model-driven cloud-edge-end interconnection integration method proposed in this embodiment can divide the large model into parts and manage them. The part requiring small computing resources is completed on the edge, and the part requiring large computing resources is executed on the cloud. This can solve the problems of insufficient edge computing resources in the cloud and edge and too high cloud computing resource load.

[0051] The model-driven cloud-edge-end interconnection integration method proposed in the present invention obtains a convolutional neural network, and then removes the redundant channels of the convolutional neural network based on the attention mechanism module to obtain a target convolutional neural network. Finally, the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system are used to perform calculations to obtain calculation results. The present invention can remove the redundant channels of the convolutional neural network and put it into different levels of the cloud-edge-end interconnection system for operation, making full use of the resource advantages of the edge and the cloud, and speeding up the related calculations of the convolutional neural network.

[0052] See also Figure 4 As shown, in one embodiment, a model-driven cloud-edge-end interconnection integration device is provided, and the device includes:

[0053] An acquisition module 10, used to acquire a convolutional neural network;

[0054] A redundant channel removal module 20, used to remove redundant channels of the convolutional neural network based on the attention mechanism module to obtain a target convolutional neural network;

[0055] The computing module 30 is used to perform calculations based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system to obtain calculation results.

[0056] In one embodiment, the redundant channel removal module 20 is used to: add an attention mechanism module to the convolutional layer to be streamlined in the convolutional neural network to obtain the added convolutional neural network; perform network training on the added convolutional neural network, and obtain the weights of the channels of the added convolutional neural network in the forward process of network training; calculate the sparsity of the channels based on the weights of the channels, determine the redundant channels based on the sparsity, remove the redundant channels from the added convolutional neural network, and obtain the trained convolutional neural network; determine whether the trained convolutional neural network meets the preset pruning target; if so, use the trained convolutional neural network as the target convolutional neural network.

[0057] In one embodiment, the computing module 30 is used to divide the task program corresponding to the target convolutional neural network into various program blocks; based on each of the program blocks and a preset unloading decision, the program blocks are sent to their respective corresponding cloud-edge-end interconnection system computing nodes for calculation, wherein the cloud-edge-end interconnection system computing nodes perform calculations based on the program blocks to obtain calculation results, and send the calculation results to the user terminal.

[0058] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client through a network connection. When the computer program is executed by the processor, it implements the functions or steps of a model-driven cloud-edge-end interconnection integration method service side.

[0059] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 6As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of a model-driven cloud-edge-end interconnection integration method.

[0060] In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are implemented:

[0061] Get a convolutional neural network;

[0062] Based on the attention mechanism module, redundant channels of the convolutional neural network are removed to obtain a target convolutional neural network;

[0063] Calculations are performed based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system to obtain calculation results.

[0064] The present invention can remove redundant channels of the convolutional neural network and run it at different levels of the cloud-edge-end interconnected system, making full use of the resource advantages of the edge and cloud, and speeding up the related calculations of the convolutional neural network.

[0065] In one embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0066] Get a convolutional neural network;

[0067] Based on the attention mechanism module, redundant channels of the convolutional neural network are removed to obtain a target convolutional neural network;

[0068] Calculations are performed based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system to obtain calculation results.

[0069] The present invention can remove redundant channels of the convolutional neural network and run it at different levels of the cloud-edge-end interconnected system, making full use of the resource advantages of the edge and cloud, and speeding up the related calculations of the convolutional neural network.

[0070] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0071] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0072] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0073] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A model-driven cloud-edge-device interconnection integration method, characterized in that: The model-driven cloud-edge-end interconnection integration method includes: Get a convolutional neural network; Based on the attention mechanism module, redundant channels of the convolutional neural network are removed to obtain a target convolutional neural network; Calculations are performed based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnection system to obtain calculation results.

2. The model-driven cloud-edge-end interconnection integration method according to claim 1 is characterized in that: The step of removing redundant channels of the convolutional neural network based on the attention mechanism module to obtain a target convolutional neural network includes: Adding an attention mechanism module to the convolutional layer to be streamlined in the convolutional neural network to obtain a convolutional neural network after the addition; Perform network training on the added convolutional neural network, and obtain the weights of the channels of the added convolutional neural network in the forward process of network training; Based on the weights of the channels, the sparsity of the channels is calculated, redundant channels are determined based on the sparsity, and the redundant channels are removed from the added convolutional neural network to obtain a trained convolutional neural network; Determine whether the trained convolutional neural network meets the preset pruning target; If satisfied, the trained convolutional neural network is used as the target convolutional neural network.

3. The model-driven cloud-edge-end interconnection integration method according to claim 1 is characterized in that: The step of performing calculation based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnected system to obtain a calculation result includes: Divide the task program corresponding to the target convolutional neural network into individual program blocks; Based on each of the program blocks and the preset uninstallation decision, the program blocks are sent to their corresponding cloud-edge-end interconnection system computing nodes for calculation, wherein the cloud-edge-end interconnection system computing nodes perform calculations based on the program blocks, obtain calculation results, and send the calculation results to the user terminal.

4. A model-driven cloud-edge-end interconnection integration device, characterized in that: The model-driven cloud-edge-end interconnection integration device includes: An acquisition module, used to acquire a convolutional neural network; A redundant channel removal module, used for removing redundant channels of the convolutional neural network based on the attention mechanism module to obtain a target convolutional neural network; The computing module is used to perform calculations based on the target convolutional neural network and the computing nodes of the cloud-edge-end interconnected system to obtain calculation results.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the model-driven cloud-edge-end interconnection integration method as described in any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the model-driven cloud-edge-end interconnection integration method as described in any one of claims 1 to 3 are implemented.

Images