Fundus screening device and hardware, software and device upgrade method and apparatus

Abstract

The disclosure discloses an eye fundus screening device, relates to the field of computers and artificial intelligence, and in particular to the fields of deep learning, cloud computing, computer vision, voice technology and the like, and can be applied to scenes such as AI medical treatment.The specific implementation scheme is: a computing resource, wherein the computing resource is configured based on a data model running in the eye fundus screening device; a CPU, wherein the CPU comprises N cores, and the number of cores of the CPU is configured based on an application program running in the eye fundus screening device; and an internal storage device, wherein the internal storage device is configured based on application programs and model files that need to be loaded by the eye fundus screening device.

Description

Technical Field

[0001] This disclosure relates to the fields of computer science and artificial intelligence, particularly deep learning, cloud computing, computer vision, and voice technology, and can be applied to scenarios such as AI-assisted healthcare. Specifically, it relates to a method and apparatus for upgrading hardware, software, and devices, as well as fundus screening equipment, electronic devices, storage media, and computer program products. Background Technology

[0002] Traditional fundus screening methods are implemented either on cloud servers or on local terminal devices. Because these two methods are used in different scenarios, they also present different problems and have different limitations. Summary of the Invention

[0003] This disclosure provides a method and apparatus for upgrading hardware, software and devices, as well as fundus screening devices, electronic devices, storage media and computer program products.

[0004] According to one aspect of this disclosure, a fundus screening device is provided, comprising: computing power resources, wherein the computing power resources are configured based on a data model running in the fundus screening device; a CPU, wherein the CPU includes N cores, and the number of CPU cores is configured based on an application running in the fundus screening device; and an internal storage device, wherein the internal storage device is configured based on the application and model files that the fundus screening device needs to load.

[0005] According to another aspect of this disclosure, a hardware upgrade method is provided, comprising: determining the intended use of a terminal device and at least one data model required to achieve the intended use; predicting the target hardware resources required to achieve the intended use based on the intended use and the at least one data model to obtain a corresponding prediction result; and configuring the required target hardware resources for the terminal device based on the prediction result to achieve a hardware upgrade based on the original terminal device, wherein the original terminal device is also used to achieve the intended use.

[0006] According to another aspect of this disclosure, a software upgrade method is provided, comprising: determining the purpose implemented by a first terminal device and at least one data model for implementing the purpose; obtaining a corresponding second model for a second terminal device by performing a first operation based on a first model in the at least one data model, thereby upgrading the software in the second terminal device, wherein the first operation includes at least one of the following: performing basic network optimization based on the first model to obtain a second model for the second terminal device having a specific basic network; performing model compression based on the first model to obtain a second model for the second terminal device; configuring the first model to be executed in parallel with a first target model when used in the second terminal device, wherein the first target model includes a model directly ported from the at least one data model, or a model obtained by optimizing a third model in the at least one data model; merging the first model with the second target model to obtain a second model for the second terminal device, wherein the second target model includes a model directly ported from the at least one data model, or a model obtained by optimizing a fourth model in the at least one data model.

[0007] According to another aspect of this disclosure, a device upgrade method is provided, including: the hardware upgrade method described in the embodiments of this disclosure; and / or the software upgrade method described in the embodiments of this disclosure.

[0008] According to another aspect of this disclosure, a hardware upgrade apparatus is provided, comprising: a first determining module, configured to determine the intended use of a terminal device and at least one data model required to achieve the intended use; a prediction module, configured to predict the target hardware resources required to achieve the intended use based on the intended use and the at least one data model, to obtain a corresponding prediction result; and a configuration module, configured to configure the required target hardware resources for the terminal device based on the prediction result, to achieve a hardware upgrade based on the original terminal device, wherein the original terminal device is also used to achieve the intended use.

[0009] According to another aspect of this disclosure, a software upgrade apparatus is provided, comprising: a second determining module, configured to determine the purpose implemented by a first terminal device and at least one data model for implementing the purpose; a software upgrade module, configured to obtain a corresponding second model for a second terminal device by performing a first operation based on a first model among the at least one data model, thereby upgrading the software in the second terminal device, wherein the first operation is implemented by at least one of the following units: a network optimization unit, configured to perform basic network optimization based on the first model to obtain a second model for the second terminal device having a specific basic network; a model compression unit, configured to perform model compression based on the first model to obtain a second model for the second terminal device; a parallel processing unit, configured to configure the first model to be executed in parallel with a first target model when used for the second terminal device, wherein the first target model includes a model directly ported from the at least one data model, or a model obtained by optimizing a third model among the at least one data model; and a model merging unit, configured to merge the first model and the second target model to obtain a second model for the second terminal device, wherein the second target model includes a model directly ported from the at least one data model, or a model obtained by optimizing a fourth model among the at least one data model.

[0010] According to another aspect of this disclosure, a device upgrade apparatus is provided, including: the hardware upgrade apparatus described in the embodiments of this disclosure; and / or the software upgrade apparatus described in the embodiments of this disclosure.

[0011] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the methods described in the embodiments of this disclosure.

[0012] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described according to embodiments of this disclosure.

[0013] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method described according to embodiments of this disclosure.

[0014] According to another aspect of this disclosure, a fundus screening device is provided, comprising: hardware resources and software resources, wherein the hardware resources are upgraded using the hardware upgrade method described in the embodiments of this disclosure; and / or the software resources are upgraded using the software upgrade method described in the embodiments of this disclosure.

[0015] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0016] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0017] Figure 1 An exemplary overall architecture diagram suitable for embodiments of this disclosure is shown;

[0018] Figure 2 An exemplary flowchart of a hardware upgrade method according to an embodiment of this disclosure is shown;

[0019] Figure 3 An exemplary flowchart of a software upgrade method according to an embodiment of the present disclosure is shown;

[0020] Figure 4 A flowchart of a device upgrade method according to an embodiment of the present disclosure is illustrated by way of example;

[0021] Figure 5 A block diagram of a hardware upgrade apparatus according to an embodiment of the present disclosure is shown as an example;

[0022] Figure 6 A block diagram of a software upgrade apparatus according to an embodiment of the present disclosure is shown as an example;

[0023] Figure 7 A block diagram of a device upgrade apparatus according to an embodiment of the present disclosure is shown as an example;

[0024] Figure 8 An exemplary block diagram of a fundus screening device according to an embodiment of the present disclosure is shown;

[0025] Figure 9 A block diagram of a fundus screening device according to another embodiment of the present disclosure is shown as an example; and

[0026] Figure 10 A block diagram of an electronic device for implementing various methods and apparatuses of embodiments of the present disclosure is shown as an example. Detailed Implementation

[0027] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0028] Traditional fundus screening methods are implemented either on cloud servers or on local terminal devices. Because these two methods are used in different scenarios, they also present different problems and have different limitations.

[0029] (1) Cloud-based solution: This solution transmits fundus images from the local device to a cloud computing device for processing and then transmits the results back to the local device for display. The cloud computing device can typically be a high-performance server.

[0030] (2) Local solutions, such as x86 PC solutions, typically consist of an x86 CPU and deep learning acceleration devices. Deep learning acceleration devices can be GPUs, FPGAs, or ASICs, etc.

[0031] It should be understood that the drawback of scheme (1) is that it is highly dependent on the network, has a high transmission delay, takes a long time, and requires a high cost for long-term use. It is not suitable for use in remote areas or areas with poor network conditions.

[0032] It should also be understood that the drawback of scheme (2) is that AI computing devices are generally large in size, inconvenient to move and carry, and have high costs, making them unsuitable for the existing market (because the existing market is sensitive to the cost and size of AI computing devices).

[0033] Based on this, the embodiments of this disclosure provide an upgrade scheme including hardware, software, and devices, which can be applied to upgrade scenarios involving hardware, software, and devices related to fundus screening. In the hardware upgrade scheme, hardware resources can be configured based on the intended use of the device and the software resources required to achieve that use. This product upgrade can overcome the aforementioned deficiencies in related technologies.

[0034] The present disclosure will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0035] The overall architecture of the methods and apparatuses suitable for embodiments of this disclosure is described below.

[0036] Figure 1 An exemplary overall architecture diagram suitable for embodiments of this disclosure is shown. It should be noted that... Figure 1The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other environments or scenarios.

[0037] like Figure 1 As shown in this embodiment, the system architecture 100 of the upgraded terminal device (such as an AI computing device) may include hardware upgrades, software upgrades, and personalized function upgrades. Hardware upgrades mainly involve computing power upgrades (such as GPU upgrades), CPU upgrades, and memory upgrades. Software upgrades mainly involve model optimization, model compression, model parallelization, and model merging. Personalized function upgrades mainly involve the expansion of voice control and facial recognition functions.

[0038] It should be noted that the specific technical means used to achieve the above upgrade will be described in detail in the following embodiments of this article, and will not be repeated in this embodiment.

[0039] It should be understood that in this embodiment, the terminal device mentioned above can be a fundus screening device, which is an AI computing device or an AI computing platform. After the upgrade is implemented through the upgrade scheme provided in this embodiment, compared with the original fundus screening device before the upgrade, the upgraded product has a wide range of upgrades in terms of data model, computing power, compatibility and external interfaces. The upgraded product is characterized by high algorithm compatibility, fast running speed, strong scalability, and the addition of personalized functions, which can adapt to more business scenarios of fundus screening.

[0040] It should be noted that, in this embodiment, compatibility refers to the ability to support the development of more complex applications, both current and in the future. External interfaces include input / output interfaces such as video and audio interfaces.

[0041] It should be understood that Figure 1 The hardware upgrades, software upgrades, and personalized feature upgrades shown are merely illustrative. Depending on implementation needs, these upgrades can include other features as well.

[0042] The application scenarios of the various methods and apparatuses suitable for the embodiments of this disclosure are described below.

[0043] It should be understood that the upgrade solutions provided in this disclosure can be used for hardware upgrades, software upgrades, and combined hardware and software upgrades, and are not limited thereto. The following will use a fundus screening device as an example to illustrate in detail the various upgrade solutions provided in this disclosure.

[0044] According to embodiments of this disclosure, this disclosure provides a hardware upgrade method.

[0045] Figure 2A flowchart of a hardware upgrade method according to an embodiment of this disclosure is illustrated.

[0046] like Figure 2 As shown, the hardware upgrade method 200 may include operations S210 to S230.

[0047] In operation S210, the intended use of the terminal device and at least one data model required to achieve that use are determined.

[0048] In operation S220, based on the purpose and at least one data model, the target hardware resources required to realize the purpose are predicted to obtain the corresponding prediction results.

[0049] In operation S230, based on the prediction results, the required target hardware resources are configured for the terminal device to achieve hardware upgrade based on the original terminal device, which is also used to achieve the purpose.

[0050] In operation S210, based on the original terminal equipment before the upgrade, the intended use (i.e., function) that the upgraded product needs to achieve, and at least one data model required to achieve that use, can be determined. In this embodiment, the data model may include image models, such as image detection models, image segmentation models, image classification models, and image recognition models.

[0051] In operations S220 and S230, based on the purpose determined by operation S210 and at least one data model for implementing that purpose, the target hardware resources (GPU, CPU, and memory, etc.) required to implement that purpose in the upgraded product are predicted to obtain the corresponding prediction results, and a certain number / specification of hardware resources are configured for the upgraded product based on the prediction results.

[0052] For example, it is predicted that in product upgrades, the number of CPU cores can be increased to 1.5 times that of the original product, so that programs running on the CPU can achieve greater speedup.

[0053] For example, it is predicted that in product upgrades, the memory can be upgraded to twice that of the original product, which allows the upgraded product to load more applications and model files.

[0054] For example, in predicting product upgrades, computing power (such as GPUs) can be increased up to 20 times that of the original product. This upgrade can not only accelerate the data models currently used in the product, but also provide abundant computing power to support the expansion of applications.

[0055] It should be noted that this embodiment can be used to upgrade the hardware of the embedded AI computing platform / device that implements the fundus screening algorithm.

[0056] Through the embodiments disclosed herein, the hardware resources required for product upgrades can be predicted based on actual application scenarios, and then the hardware resources of the upgraded products can be configured according to the prediction results to achieve upgrades in computing power (such as GPU), CPU, and memory.

[0057] Furthermore, through the embodiments disclosed herein, not only can the shortcomings of cloud solutions provided in related technologies (such as heavy reliance on the network, severe latency, and high cost) be overcome, but the hardware resources in the upgraded product are configured according to actual needs, thus saving costs. At the same time, the size of the product can be reduced to a certain extent, making it easy to carry and move.

[0058] As an optional embodiment, the hardware upgrade may include at least one of the following: computing power upgrade, memory upgrade, CPU upgrade.

[0059] For example, it is predicted that in product upgrades, the number of CPU cores can be increased to 1.5 times that of the original product, so that programs running on the CPU can achieve greater speedup.

[0060] For example, it is predicted that in product upgrades, the memory can be upgraded to twice that of the original product, which allows the upgraded product to load more applications and model files.

[0061] For example, in predicting product upgrades, computing power (such as GPUs) can be increased up to 20 times that of the original product. This upgrade can not only accelerate the data models currently used in the product, but also provide abundant computing power to support the expansion of applications.

[0062] Through the embodiments of this disclosure, after upgrading the hardware resources such as computing power, memory, and CPU in the upgraded product, the running speed of each data model can be accelerated, the inference time can be shortened, and the upgraded product can load more applications and model files.

[0063] According to embodiments of this disclosure, a software upgrade method is provided.

[0064] Figure 3 A flowchart of a software upgrade method according to an embodiment of this disclosure is illustrated.

[0065] like Figure 3 As shown, the software upgrade method 300 may include operations S310 to S320.

[0066] In operation S310, the purpose implemented by the first terminal device and at least one data model for implementing the purpose are determined.

[0067] In operation S320, based on a first model in at least one data model, a corresponding second model for a second terminal device is obtained by performing a first operation, so as to upgrade the software in the second terminal device.

[0068] In operation S320, the first operation 321 may include one or more of operations S3211 to S3214.

[0069] In operation S3211, based on the first model, basic network optimization is performed to obtain a second model with a specific basic network for the second terminal device.

[0070] In operation S3212, based on the first model, model compression is performed to obtain a second model for the second terminal device.

[0071] In operation S3213, when used for a second terminal device, the first model is configured to be executed in parallel with the first target model, wherein the first target model includes a model directly ported from at least one data model, or a model obtained by optimizing a third model in at least one data model.

[0072] In operation S3214, the first model and the second target model are merged to obtain a second model for the second terminal device. The second target model includes a model directly ported from at least one data model, or a model obtained by optimizing the fourth model in at least one data model.

[0073] It should be noted that in this embodiment of the disclosure, the first terminal device is the original terminal device, that is, the terminal device before the upgrade, which can become the original product; the second terminal device is obtained by upgrading the original terminal device, that is, the upgraded terminal device can become the upgraded product.

[0074] Therefore, in operation S310, the intended use / function of the original product and all data models used to implement that use / function can be determined. In operation S320, based on one or more (including all) of the data models determined by operation S310, such as the first model, specified model operations (such as model optimization, model compression, model parallelism, model merging, or one or more of these) can be performed to obtain the corresponding model used in the upgraded product, such as the second model, thereby realizing the software upgrade in the upgraded product.

[0075] Operation S3211 allows for model optimization of any one or more models in the existing product to obtain the corresponding model for the upgraded product. This operation includes the selection of the base network within the model.

[0076] Similarly, operation S3212 can compress any one or more models in the original product to obtain the corresponding model for the upgraded product. In this operation, model compression includes model compression using pruning and / or distillation methods, as well as model optimization based on this compression and incorporating relevant loss functions.

[0077] Similarly, operation S3213 can perform model parallel processing on any two or more models in the original product to obtain the corresponding model for the upgraded product. In this operation, model parallelism includes model parallelism based on model dependencies and / or the computing power required to run the model.

[0078] Similarly, operation S3214 can merge any two or more models in the original product to obtain the corresponding model for the upgraded product. In this operation, model merging includes merging the original models and merging the models obtained after optimization / compression / parallel processing.

[0079] Furthermore, the aforementioned model merging includes combining multiple single-task single models into a single multi-task single model, thereby saving computational resources and training time. For example, models for recognizing apples, bananas, and oranges can be merged into a single object recognition model.

[0080] Specifically, by merging and optimizing data annotations, single-task single-model models can be merged and optimized into multi-task single-model models, which can save both the computational power consumption of the models and the training time of models in multiple scenarios.

[0081] It should be noted that, in order to achieve stability optimization in this embodiment, multiple rounds of optimization operations can be repeated after the above operations are completed to ensure that the overall data model (also known as the algorithm model) can run smoothly on the upgraded product.

[0082] Furthermore, it should be noted that in this embodiment, the models can also be analyzed as a whole, the dependencies between models can be optimized, and the models can be deployed collaboratively in the upgraded product to maximize the utilization of resources in the overall upgraded product.

[0083] Through the embodiments of this disclosure, in the software upgrade of the product, performing model optimization operations can optimize the basic network in the model, making the model perform better and reducing the model size, thereby reducing the model's inference time; performing model compression operations can achieve a high model compression ratio while maintaining high model accuracy; performing model parallel operations can save inference time and make full use of the product's computing resources; performing model merging operations can save the model's computing power consumption and save model training time in multiple scenarios.

[0084] Furthermore, as an optional embodiment, the first operation described above may also include: after obtaining a second model with a specific underlying network, adapting and optimizing the convolution operators and / or pooling operators of the specific underlying network in the second model based on the target hardware resources configured in the second terminal device.

[0085] For example, in this embodiment, for the fundus screening scenario, model optimization may include: for the image detection model, image segmentation model, and image classification model used in the original product, the upgraded product may use a state-of-the-art (SOTA) model in these models. Furthermore, the backbone network (i.e., the base network) of the SOTA model may be optimized, such as by selecting a specific base network in the SOTA model.

[0086] It's worth noting that, for the fundus screening scenario, the original product's algorithm design initially employed models with a large number of parameters to pursue high accuracy. This resulted in the original product's data models consuming a significant amount of memory and having a long inference time. However, in implementing the upgraded product, a thorough analysis of various fundus screening scenarios and the data modules used in the original product was conducted, optimizing each model. Therefore, even with a doubling of the number of data models in the upgraded product, the high accuracy requirements can still be met, while simultaneously significantly reducing the overall memory consumption and inference time of the data models.

[0087] Taking image segmentation models as an example, state-of-the-art (SOTA) models can be used in image segmentation models, and the Efficient Net series in the SOTA network can be used as the backbone network of the SOTA model. Thus, compared with the original product, the image segmentation model in the upgraded product can achieve better performance.

[0088] Furthermore, after determining that the SOTA model uses the EfficientNet series as the backbone network, all convolution operators and / or pooling operators in the backbone network of the model can be adapted and optimized based on the existing hardware resources in the upgraded product, thereby further improving the performance of the upgraded product.

[0089] It should be noted that the range of backbone networks that can be selected in the state-of-the-art (SOTA) model is quite wide. In this embodiment, the Efficient Net series was selected as the backbone network based on rich empirical values ​​and a large number of experimental results, which itself required a great deal of creative work.

[0090] Through the embodiments of this disclosure, in the model optimization stage, after selecting the base network of the data model, it is further adapted to the hardware resources of the upgraded product. All convolution operators and / or pooling operators in the base network are adapted and optimized, which can further improve the performance of the data model in the upgraded product and achieve the goal of making full use of existing computing resources.

[0091] And / or, as an optional embodiment, the first operation described above may further include: after obtaining a second model with a specific base network, optimizing the image size of the input to the second model.

[0092] For example, the image size of the input image can be reduced from 1024x1024 to 512x512. After reducing the image size of the input image in the upgraded product, compared with the performance of the original product, the size of the corresponding data model can be reduced to 1 / 5 of the original, and the inference time can be reduced to 1 / 4 of the original.

[0093] By using the above-described technical means to optimize the model through the embodiments of this disclosure, it is possible to shorten the inference time of the model while maintaining its high accuracy.

[0094] In addition, as an optional embodiment, model compression based on the first model may include the following operations.

[0095] A convolutional layer sensitivity analysis was performed on the first model to identify at least one neuron that is insensitive in the convolutional layer network under the current application scenario.

[0096] At least one neuron is removed from the convolutional layer network to achieve model compression.

[0097] For example, model compression can be achieved based on the PaddleSlim algorithm. In this embodiment, pruning and / or distillation can be used to achieve model compression, resulting in a compression rate of over 50% while maintaining high model accuracy. It should be noted that the PaddleSlim algorithm is a framework / tool ​​for implementing model compression.

[0098] Specifically, in this embodiment, when compressing any data model, the convolutional layer sensitivity analysis of the data model can be performed based on the L1-Norm algorithm to identify at least one neuron in the convolutional layer network of the data model that is insensitive in the current application scenario, and then remove that at least one neuron from the convolutional layer network. It should be noted that the L1-Norm algorithm is the specific algorithm for implementing model compression.

[0099] Through this embodiment, while ensuring model performance, neurons in the convolutional network that are not sensitive to the current application scenario can be removed, and only neurons that are sensitive to the current application scenario can be retained. This can improve the model compression rate and achieve lightweighting of software in upgraded products.

[0100] In addition, as an optional embodiment, model compression based on the first model may include: trimming the hierarchical structure of the first model based on empirical values ​​to achieve model compression.

[0101] For example, during the model compression stage, neurons in the model's convolutional network can be reduced; at the same time, other pruning strategies can be combined to prune the model, such as pruning the model's hierarchical structure based on empirical values; at the same time, distillation strategies can be combined (such as determining whether to use a large or small model based on the actual application scenario) to perform model distillation; and at the same time, loss functions can be combined to perform multiple rounds of iteration to optimize the model.

[0102] For example, using the model compression method provided in this disclosure, the model compression rate can reach more than 50% for a single image detection model, and the model can still maintain high-precision output after compression.

[0103] Through this embodiment, while ensuring model performance, the model can be pruned, distilled, and subjected to multiple rounds of iterative processing based on the loss function. This can also improve the model compression ratio and achieve lightweighting of software in upgraded products.

[0104] Furthermore, as an optional embodiment, the first model and the first target model are not dependent on each other.

[0105] For example, in one embodiment, during the model parallelization phase, for any given data model, all data models in the upgrade product that are not dependent on it can be processed into a state / mode that executes in parallel with this data model, thereby making full use of the existing hardware resources (such as computing power) in the upgrade product.

[0106] Furthermore, as an optional embodiment, the second terminal device has the hardware conditions to satisfy the parallel operation of the first model and the first target model.

[0107] For example, in another embodiment, during the model parallelization phase, for any given data model, in addition to satisfying specific dependency conditions (i.e., models that can run in parallel cannot depend on each other, otherwise model parallelism cannot be achieved), it is also necessary to consider whether the existing hardware resources (such as computing power) in the upgraded product can support the simultaneous operation of all models that satisfy the specific dependency conditions. If it can, then all models that satisfy the specific dependency conditions are processed into models that run in parallel. If it cannot, then model parallelization processing of all models that satisfy the specific dependency conditions is abandoned.

[0108] Through the embodiments of this disclosure, by adopting the above-described technical means, calculations can be performed based on model parameters and hardware resources such as computing power, and the originally independent serially executed models can be optimized into a multi-model parallel mode. This can prevent the performance of the product from being affected by excessive parallelism of models, save the inference time of the models to the greatest extent, and make full use of the existing computing power in the upgraded product.

[0109] Furthermore, as an optional embodiment, the method may further include adding at least one of the following models to the second terminal device to upgrade the software in the second terminal device: a data model for implementing voice control; and a data model for implementing face recognition.

[0110] In this disclosed embodiment, in addition to optimizing / compressing / parallelizing / merging the models used in the original product to upgrade the software in the upgraded product, personalized functions (such as voice control function and face recognition function) can also be extended on this basis to upgrade the software in the upgraded product.

[0111] It should be understood that adding voice control functionality makes the upgraded product more user-friendly, while also saving human resources. That is, adding voice control facilitates user operation and allows the upgraded product to adapt to a wider range of application scenarios. Furthermore, adding facial recognition functionality also makes the upgraded product more user-friendly and, in some scenarios, enables paperless identity verification. For example, after using the upgraded product provided in this disclosure embodiment, patients can use facial recognition instead of a specific paper slip to verify their identity when seeking medical treatment, thus further realizing a new approach to paperless medical care.

[0112] According to embodiments of this disclosure, this disclosure provides a device upgrade method.

[0113] Figure 4 A flowchart of a device upgrade method according to an embodiment of this disclosure is illustrated.

[0114] like Figure 4As shown, the equipment upgrade method may include: operating one or more of S410 to S420.

[0115] Hardware upgrade methods for operating S410.

[0116] Software upgrade method for operating S420.

[0117] It should be understood that in this embodiment, the product may be upgraded only in hardware or software, or both hardware and software may be upgraded simultaneously. This embodiment does not limit this.

[0118] It should be noted that the hardware upgrade method and software upgrade method used in this embodiment can refer to the relevant descriptions in the above embodiments of this disclosure, and will not be repeated here.

[0119] The technical problems solved and the technical effects achieved by the embodiments of this disclosure can also be referred to the relevant descriptions in the above embodiments of this disclosure, which will not be repeated here.

[0120] According to embodiments of this disclosure, this disclosure also provides a hardware upgrade device.

[0121] Figure 5 A block diagram of a hardware upgrade apparatus according to an embodiment of the present disclosure is shown as an example.

[0122] like Figure 5 As shown, the hardware upgrade device 500 may include: a first determining module 510, a prediction module 520, and a configuration module 530.

[0123] The first determining module 510 is used to determine the purpose to be achieved by the terminal device and at least one data model required to achieve that purpose.

[0124] The prediction module 520 is used to predict the target hardware resources required to achieve the purpose based on the purpose and the at least one data model, so as to obtain the corresponding prediction results.

[0125] The configuration module 530 is used to configure the required target hardware resources for the terminal device based on the prediction result, so as to realize the hardware upgrade based on the original terminal device, wherein the original terminal device is also used to achieve this purpose.

[0126] As an optional embodiment, the hardware upgrade includes at least one of the following: computing power upgrade, memory upgrade, CPU upgrade.

[0127] It should be understood that the embodiments of the apparatus portion of this disclosure correspond to the same or similar embodiments of the method portion of this disclosure, and the technical problems solved and the technical effects achieved are also the same or similar. This disclosure will not repeat them here.

[0128] According to embodiments of this disclosure, this disclosure also provides a software upgrade apparatus.

[0129] Figure 6 A block diagram of a software upgrade apparatus according to an embodiment of the present disclosure is shown as an example.

[0130] like Figure 6 As shown, the software upgrade device 600 may include a second determining module 610 and a software upgrade module 620. The function implemented by the software upgrade module 620 in performing the first operation can be achieved by one or more of the network optimization unit 621, the model compression unit 622, the parallel processing unit 623, and the model merging unit 624.

[0131] The second determining module 610 is used to determine the purpose implemented by the first terminal device and at least one data model for implementing that purpose.

[0132] The software upgrade module 620 is used to obtain a corresponding second model for the second terminal device by performing a first operation based on the first model in the at least one data model, so as to realize the software upgrade in the second terminal device.

[0133] The network optimization unit 621 is used to perform basic network optimization based on the first model to obtain a second model with a specific basic network for the second terminal device.

[0134] The model compression unit 622 is used to compress the model based on the first model to obtain a second model for the second terminal device.

[0135] The parallel processing unit 623 is configured to execute the first model in parallel with the first target model when used in the second terminal device, wherein the first target model includes a model directly ported from the at least one data model, or a model obtained by optimizing a third model in the at least one data model.

[0136] The model merging unit 624 is used to merge the first model with the second target model to obtain a second model for the second terminal device, wherein the second target model includes a model directly ported from the at least one data model, or a model obtained by optimizing the fourth model in the at least one data model.

[0137] As an optional embodiment, the network optimization unit is further configured to: after obtaining the second model with a specific basic network, adapt and optimize the convolution operator and / or pooling operator of the specific basic network in the second model based on the target hardware resources configured in the second terminal device.

[0138] As an optional embodiment, the network optimization unit is also used to: optimize the image size of the input to the second model after obtaining the second model with the specific base network.

[0139] As an optional embodiment, the model compression unit includes: a sensitivity analysis subunit for performing convolutional layer sensitivity analysis on the first model to obtain at least one neuron that is insensitive in the convolutional layer network under the current application scenario; and a neuron deletion subunit for deleting the at least one neuron from the convolutional layer network to achieve model compression.

[0140] As an optional embodiment, the model compression unit is also used to: prune the model hierarchy structure of the first model based on empirical values ​​to achieve model compression.

[0141] As an optional embodiment, the first model and the first target model are not dependent on each other.

[0142] As an optional embodiment, the second terminal device has the hardware conditions to satisfy the parallel operation of the first model and the first target model.

[0143] As an optional embodiment, the device further includes a model adding module for adding at least one of the following models to the second terminal device to upgrade the software in the second terminal device: a data model for implementing voice control; and a data model for implementing face recognition.

[0144] It should be understood that the embodiments of the apparatus portion of this disclosure correspond to the same or similar embodiments of the method portion of this disclosure, and the technical problems solved and the technical effects achieved are also the same or similar. This disclosure will not repeat them here.

[0145] According to embodiments of this disclosure, this disclosure also provides a device upgrade apparatus.

[0146] Figure 7 A block diagram of a device upgrade apparatus according to an embodiment of the present disclosure is shown as an example.

[0147] like Figure 7 As shown, the device upgrade apparatus 700 may include at least one of the following: hardware upgrade apparatus 710 and software upgrade apparatus 720.

[0148] It should be noted that the hardware upgrade device 710 in this embodiment can be referred to the relevant description in the above embodiments of this disclosure, and will not be repeated here. The software upgrade device 720 in this embodiment can also be referred to the relevant description in the above embodiments of this disclosure, and will not be repeated here either.

[0149] It should be understood that the embodiments of the apparatus portion of this disclosure correspond to the same or similar embodiments of the method portion of this disclosure, and the technical problems solved and the technical effects achieved are also the same or similar. This disclosure will not repeat them here.

[0150] According to embodiments of this disclosure, this disclosure also provides a fundus screening device.

[0151] Figure 8 A block diagram of a fundus screening device according to an embodiment of the present disclosure is shown as an example.

[0152] like Figure 8 As shown, the fundus screening device 800 may include hardware resources 810 and software resources 820. The hardware resources 810 are upgraded using the hardware upgrade method provided in this disclosure embodiment; and / or the software resources 820 are upgraded using the software upgrade method provided in this disclosure embodiment.

[0153] It should be noted that the hardware resources 810 in this embodiment may include, but are not limited to, computing resources (such as GPUs), CPUs, and memory. The software resources 820 in this embodiment may include the data model and application program in the fundus screening device 800 used to implement the fundus screening function, as well as the model and application program used to implement various extended functions (such as personalized functions such as voice control and facial recognition).

[0154] It should be noted that the fundus screening device 800 can be obtained by upgrading an existing fundus screening device. Specifically, the upgrade of the hardware resources 810 in the fundus screening device 800 can refer to the description of the hardware upgrade method in the above embodiments, and will not be repeated here. Similarly, the upgrade of the software resources 820 in the fundus screening device 800 can also refer to the description of the software upgrade method in the above embodiments, and will not be repeated here either.

[0155] It should be understood that, in this embodiment of the present disclosure, the upgraded fundus screening device 800 is obtained by expanding the voice control function on the basis of the original fundus screening device, which can make the product more user-friendly, and at the same time, using the upgraded product can save human resources.

[0156] It should also be understood that, in this embodiment of the present disclosure, the fundus screening device 800 is upgraded by expanding the face recognition function on the basis of the original fundus screening device. Its practical significance in fundus screening scenarios is that this upgraded product can be widely used in portable scenarios where it is inconvenient to carry barcode scanners, printing paper and network cables, as well as other scenarios to be developed.

[0157] Through the embodiments disclosed herein, compared with the original product, the upgraded product can increase computing power by approximately 20 times while maintaining a basically unchanged size, and the running speed of related algorithms for fundus screening can be increased by approximately 1 time. Furthermore, the upgraded product can accommodate algorithm models of approximately twice the size of the original product. In addition, the upgraded product adds personalized functions such as voice control and facial recognition, enabling its application in a wider range of scenarios.

[0158] According to embodiments of this disclosure, another fundus screening device is also provided.

[0159] Figure 9 A block diagram of a fundus screening device according to another embodiment of the present disclosure is shown as an example.

[0160] like Figure 9 As shown, the fundus screening device 900 may include: computing resources 910, CPU 920 and internal storage device 930.

[0161] The computing power resource is 910, which is configured based on the data model running in the fundus screening device.

[0162] CPU 920, wherein the CPU includes N cores, and the number of CPU cores is based on the configuration of the application running in the fundus screening device.

[0163] Internal storage device (also known as memory) 930, wherein the internal storage device is configured based on the application and model files that the fundus screening device needs to load.

[0164] It should be noted that the fundus screening device configured and implemented in this embodiment is an upgraded product of the original fundus screening device.

[0165] Furthermore, the computing power resource 910 in the upgraded product is not arbitrarily configured, nor is it simply a matter of configuring it to be stronger than the computing power resource in the original fundus screening equipment. Rather, it is configured based on the actual data model that needs to run in the fundus screening equipment. The upgraded product, with this configuration, can exhibit stronger computing power, such as an increase of about 20 times compared to the original product, while also meeting the basic requirement of not wasting the equipment's computing power resources in order to meet the need to improve the running speed of the data model in the current equipment, thereby saving production costs as much as possible.

[0166] Furthermore, the CPU 920 in the upgraded product is not arbitrarily configured, nor is it simply configured to be more powerful than the CPU in the original fundus screening device. Rather, it is configured based on the actual applications that need to run in the fundus screening device. The upgraded product, with this configuration, can exhibit significantly stronger CPU processing capabilities—for example, the CPU in the upgraded product can be several times or even tens of times more powerful than the CPU in the original product—while simultaneously meeting the need to improve program running speed in the current device without wasting the device's CPU processing power, thereby minimizing production costs.

[0167] Furthermore, the memory configuration in the upgraded product is not arbitrary, nor is it simply a matter of configuring it to be stronger than the memory in the original fundus screening device. Rather, it is configured based on the actual program and file storage needs of the fundus screening device. This configuration results in an upgraded product that exhibits significantly stronger storage capabilities—for example, the upgraded product's memory can be several times or even tens of times greater than the original product's—while simultaneously meeting the need to increase program and file storage capacity within the current device without wasting memory, thereby minimizing production costs.

[0168] Through the embodiments disclosed herein, compared with the original product, the upgraded product can increase computing power by approximately 20 times while maintaining a substantially unchanged size, and the running speed of related algorithms for fundus screening can be increased by approximately 1 time. Furthermore, the upgraded product can accommodate algorithm models that are approximately twice the size of the original product.

[0169] As an optional embodiment, the computing resources include: GPU.

[0170] As an optional embodiment, the device may further include at least one of the following: a voice control device for implementing voice control functions on the fundus screening device; and a face recognition device for implementing face recognition functions on the fundus screening device.

[0171] In this embodiment of the disclosure, an upgraded fundus screening device can be obtained by expanding the voice control function on the basis of the original fundus screening device, thereby making the product more user-friendly and saving human resources by using the upgraded product.

[0172] And / or, in this embodiment of the disclosure, an upgraded fundus screening device can be obtained by expanding the facial recognition function on the basis of the original fundus screening device. The practical significance of the facial recognition function in fundus screening scenarios is that this upgraded product can be widely used in portable scenarios where it is inconvenient to carry barcode scanners, printing paper, and network cables, as well as other scenarios to be developed.

[0173] Through the embodiments disclosed herein, the upgraded product adds personalized functions such as voice control and / or facial recognition to the original product, thus enabling it to be applied to a wider range of scenarios.

[0174] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0175] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0176] like Figure 10 As shown, the electronic device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1009 into a random access memory (RAM) 1003. The RAM 1003 may also store various programs and data required for the operation of the electronic device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0177] Multiple components in electronic device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of displays, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0178] The computing unit 1001 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as hardware upgrade methods (or software upgrade methods, or device upgrade methods). For example, in some embodiments, the hardware upgrade method (or software upgrade method, or device upgrade method) may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the hardware upgrade method (or software upgrade method, or device upgrade method) described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform a hardware upgrade method (or a software upgrade method, or a device upgrade method) by any other suitable means (e.g., by means of firmware).

[0179] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0180] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0181] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0182] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0183] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0184] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service ecosystem, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0185] The recording, storage, and application of relevant data in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0186] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0187] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A software upgrade method, comprising: Determine the purpose implemented by the first terminal device and at least one data model for implementing the purpose; Based on the first model in the at least one data model, a corresponding second model for the second terminal device is obtained by performing a first operation, so as to realize the software upgrade in the second terminal device. The first operation includes: Based on the first model, basic network optimization is performed to obtain a second model with a specific basic network for the second terminal device. Based on the target hardware resources configured in the second terminal device, the convolution operators and / or pooling operators of the specific basic network in the second model are adapted and optimized. Based on the first model, by performing convolutional layer sensitivity analysis on the first model, at least one neuron that is insensitive in the convolutional layer network under the current fundus screening application scenario is obtained. The at least one neuron is deleted from the convolutional layer network to achieve model compression and obtain a second model for the second terminal device. When used in the second terminal device, the first model is configured to be executed in parallel with the first target model, wherein the first target model includes a model directly ported from the at least one data model, or a model obtained by optimizing the third model in the at least one data model, and the second terminal device has the hardware conditions to satisfy the parallel operation of the first model and the first target model. The first model and the second target model are merged into a multi-task model to obtain a second model for the second terminal device. The first model and the second target model are single-task models. The second target model includes a model directly ported from the at least one data model, or a model obtained by optimizing the fourth model in the at least one data model.

2. The method according to claim 1, wherein the first operation further comprises: After obtaining the second model with the specific basic network... The image size of the input to the second model is optimized.

3. The method according to claim 1, wherein, Implementing model compression also includes: The model hierarchy structure of the first model is pruned based on empirical values ​​to achieve model compression.

4. The method according to claim 1, wherein, The first model and the first target model are not dependent on each other.

5. The method according to claim 1, further comprising adding at least one of the following models to the second terminal device to achieve software upgrades in the second terminal device: Data model used to implement voice control; A data model used to implement facial recognition.

6. A method for upgrading equipment, comprising: Hardware upgrade methods; The software upgrade method according to any one of claims 1 to 5.

7. The method according to claim 6, wherein, The hardware upgrade method includes: Determine the intended use of the terminal device and at least one data model required to achieve that use; Based on the stated purpose and the at least one data model, predict the target hardware resources required to achieve the stated purpose, and obtain the corresponding prediction results; and Based on the prediction results, the required target hardware resources are configured for the terminal device to achieve a hardware upgrade based on the original terminal device, wherein the original terminal device is also used to achieve the stated purpose.

8. The method according to claim 7, wherein, The hardware upgrade includes at least one of the following: computing power upgrade, memory upgrade, CPU upgrade.

9. A fundus screening device, wherein the fundus screening device is an upgraded version of an existing fundus screening device, the upgrade including software resource upgrade, the software resources being configured using the method described in any one of claims 1 to 5 to achieve the upgrade, the fundus screening device comprising: Computing power resources, wherein the computing power resources are configured based on the data model running in the fundus screening device; CPU, wherein the CPU comprises N cores, and the number of CPU cores is configured based on the application running in the fundus screening device; and An internal storage device, wherein the internal storage device is configured based on the applications and model files that the fundus screening device needs to load.

10. The device according to claim 9, wherein, The computing resources include: GPUs.

11. The device according to claim 9 or 10, further comprising at least one of the following: A voice control device is used to implement voice control functions on the fundus screening device; and A facial recognition device is used to implement facial recognition functionality on the fundus screening device.

12. A software upgrade device, comprising: The second determining module is used to determine the purpose implemented by the first terminal device and at least one data model for implementing the purpose; The software upgrade module is used to obtain a corresponding second model for the second terminal device by performing a first operation based on a first model in the at least one data model, so as to upgrade the software in the second terminal device. The first operation is implemented by the following unit: The network optimization unit is used to perform basic network optimization based on the first model to obtain a second model with a specific basic network for the second terminal device, and to adapt and optimize the convolution operator and / or pooling operator of the specific basic network in the second model based on the target hardware resources configured in the second terminal device. The model compression unit is used to perform convolutional layer sensitivity analysis on the first model to obtain at least one neuron that is insensitive in the convolutional layer network under the current fundus screening application scenario, and delete the at least one neuron from the convolutional layer network to achieve model compression and obtain a second model for the second terminal device. A parallel processing unit is configured to execute the first model in parallel with a first target model when used in the second terminal device, wherein the first target model includes a model directly ported from the at least one data model, or a model obtained by optimizing a third model in the at least one data model, and the second terminal device has hardware conditions that satisfy the parallel operation of the first model and the first target model. The model merging unit is used to merge the first model and the second target model into a multi-task model to obtain a second model for the second terminal device. The first model and the second target model are single-task models. The second target model includes a model directly ported from the at least one data model, or a model obtained by optimizing the fourth model in the at least one data model.

13. The apparatus according to claim 12, wherein the network optimization unit is further configured to: after obtaining the second model having a specific basic network, The image size of the input to the second model is optimized.

14. The apparatus according to claim 12, wherein, The model compression unit is also used for: The model hierarchy structure of the first model is pruned based on empirical values ​​to achieve model compression.

15. The apparatus according to claim 12, wherein, The first model and the first target model are not dependent on each other.

16. The apparatus of claim 12 further includes a model adding module for adding at least one of the following models to the second terminal device to achieve software upgrades in the second terminal device: Data model used to implement voice control; A data model used to implement facial recognition.

17. An equipment upgrade device, comprising: Hardware upgrade device; The software upgrade apparatus according to any one of claims 12 to 16.

18. The apparatus according to claim 17, wherein, The hardware upgrade device includes: The first determining module is used to determine the purpose to be achieved by the terminal device and at least one data model required to achieve the purpose; A prediction module is configured to predict the target hardware resources required to achieve the stated purpose based on the stated purpose and the at least one data model, in order to obtain a corresponding prediction result; and The configuration module is used to configure the required target hardware resources for the terminal device based on the prediction results, so as to realize the hardware upgrade based on the original terminal device, wherein the original terminal device is also used to achieve the purpose.

19. The apparatus according to claim 18, wherein, The hardware upgrade includes at least one of the following: computing power upgrade, memory upgrade, CPU upgrade.

20. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

21. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-8.

22. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-8.

23. A fundus screening device, comprising: Hardware resources and software resources, among which, The hardware resources are configured to enable upgrades using the method described in claim 7 or 8; and / or The software resources are configured to achieve upgrades using the method described in any one of claims 1 to 5.

Images