Screen detection method, computer equipment and computer program product

Abstract

The invention discloses a screen detection method, computer equipment and a computer program product. The screen detection method comprises the following steps: acquiring equipment data of target equipment; obtaining a broken screen prediction result of the screen of the target equipment according to the equipment data through a first detection model; if the broken screen prediction result represents that the risk degree of damage to the screen is higher than a target threshold value, a screen detection result of the target equipment is obtained through a second detection model according to an equipment image of the target equipment, and the screen detection result is used for representing whether the screen is damaged or not. The equipment image is an image obtained by shooting a screen of the target equipment. According to the method, the screen detection efficiency can be better improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic equipment, and more particularly, to a screen detection method, a computer device, and a computer program product. BACKGROUND

[0002] With the rapid progress of science and technology and living standards, electronic devices (such as smart phones, tablet computers, etc.) have become one of the commonly used electronic products in people's lives. The screen is a part of the electronic device used to display the user interface. When people use the electronic device after purchasing the electronic device, the screen is prone to damage. The price of replacing the screen is usually high, so some manufacturers have launched screen breakage insurance for the screen of the electronic device. When the user purchases the screen breakage insurance, the screen of the electronic device usually needs to be detected to avoid the situation that the screen is damaged after the insurance is successfully purchased, but the efficiency of screen detection in the related art needs to be improved. SUMMARY

[0003] The present application provides a screen detection method, a computer device, and a computer program product, which can better improve the efficiency of screen detection.

[0004] In a first aspect, an embodiment of the present application provides a screen detection method, which includes: obtaining device data of a target device; obtaining a screen breakage prediction result of a screen of the target device according to the device data through a first detection model; if the screen breakage prediction result represents that the risk degree of damage of the screen is higher than a target threshold, obtaining a screen detection result of the target device according to a device image of the target device through a second detection model, the screen detection result being used to represent whether the screen is damaged, and the device image being an image obtained by photographing the screen of the target device.

[0005] In a second aspect, an embodiment of the present application provides a screen detection device, which includes: a data obtaining module, a first detection module, and a second detection module. The data obtaining module is configured to obtain device data of a target device. The first detection module is configured to obtain a screen breakage prediction result of a screen of the target device according to the device data through a first detection model. The second detection module is configured to, if the screen breakage prediction result represents that the risk degree of damage of the screen is higher than a target threshold, obtain a screen detection result of the target device according to a device image of the target device through a second detection model, the screen detection result being used to represent whether the screen is damaged, and the device image being an image obtained by photographing the screen of the target device.

[0006] In a third aspect, an embodiment of the present application provides a computer device, comprising: one or more processors; a memory; and one or more application programs stored in the memory and configured to be executed by the one or more processors, the one or more application programs being configured to execute the screen detection method provided in the first aspect.

[0007] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, the computer readable storage medium storing a program code, the program code being executable by a processor to execute the screen detection method provided in the first aspect.

[0008] In a fifth aspect, an embodiment of the present application provides a computer program product, the computer program product comprising a computer program, the computer program being executable by a processor to implement the screen detection method provided in the first aspect.

[0009] The scheme provided in the present application obtains device data of a target device, obtains a screen breakage prediction result of a screen of the target device according to the device data through a first detection model, and if the screen breakage prediction result represents that a risk degree of screen breakage is higher than a target threshold, obtains a screen detection result of the target device according to a device image of the target device through a second detection model, the screen detection result being used to represent whether the screen is damaged, and the device image being an image obtained by photographing the screen of the target device. Thus, in the case where it is determined that the risk degree of screen breakage is higher than the target threshold, the screen is detected according to the device image in a visual detection manner, so that the screen to be detected does not need to be directly detected in the visual detection manner, and therefore the efficiency of screen detection can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort.

[0011] Figure 1 A schematic diagram of an application environment provided by an embodiment of the present application is shown.

[0012] Figure 2 A flowchart of a screen detection method according to an embodiment of the present application is shown.

[0013] Figure 3 A flowchart of a screen detection method according to another embodiment of the present application is shown.

[0014] Figure 4 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0015] Figure 5 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0016] Figure 6 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0017] Figure 7 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0018] Figure 8 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0019] Figure 9 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0020] Figure 10 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0021] Figure 11 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0022] Figure 12 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0023] Figure 13 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application.

[0024] Figure 14 Fig. 1 shows a flow diagram of a screen detection method according to another embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the personnel in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0026] With the development of screen technology, the screens of current electronic devices are becoming increasingly sophisticated. However, in actual use, these screens are prone to breakage, and due to the high cost of screens, replacements are also expensive. Therefore, screen breakage insurance has emerged in the market. Screen breakage insurance covers damage to electronic devices caused by accidental drops or other reasons during the insurance period. In other words, if a consumer purchases screen breakage insurance for their electronic device, they can receive a free replacement of the same type and material screen component if the screen breaks or cracks due to an accident within the coverage period.

[0027] When purchasing screen breakage insurance, the prerequisite is that the screen components (inner and outer screens) are intact. However, consumers often only develop a strong motivation to purchase insurance after the screen is damaged, leading to "adverse selection" fraud. Therefore, accurate, efficient, and convenient screen damage detection is crucial when users purchase screen breakage insurance.

[0028] Currently, when users purchase screen breakage insurance, the service provider typically checks if the screen is damaged: the user takes a picture of the device's screen using another device and uploads it to a backend server. The backend staff then manually review the photo to determine if the screen is intact. If the screen is confirmed to be intact, the device is insured. However, this method relies heavily on manual review, resulting in low efficiency and a high risk of misjudgments. While some service providers have adopted artificial intelligence for screen detection, this typically involves taking a picture of the device's screen and performing visual inspection. However, this method requires acquiring a photo and performing visual inspection for each device, and the efficiency of screen damage detection still needs improvement.

[0029] To address the aforementioned issues, the inventors have proposed a screen detection method, computer device, and computer program product as provided in this application. These methods only detect screen damage based on a device image using visual inspection when the risk of screen damage exceeds a target threshold. This eliminates the need for direct visual inspection of every screen, thus significantly improving screen detection efficiency. The specific screen detection method will be described in detail in subsequent embodiments.

[0030] The scenarios involved in the embodiments of this application will be introduced below.

[0031] like Figure 1 As shown, in Figure 1The scenario shown includes a computer device 100 and an insured target device 200. The computer device 100 can be a server (a network access server), a server cluster (a cloud server) composed of several servers, or a cloud computing center (a database server). The target device 200 can be any device with communication and storage functions, including but not limited to a PC (Personal Computer), a PDA (a tablet computer), a smart television, a smart phone, a smart wearable device, or other smart communication devices with network connection functions.

[0032] The computer device 100 and the target device 200 can interact through a network to receive or send information. In the embodiment of the present application, when the target device 200 needs to purchase a screen breakage insurance, it can initiate a related request to the computer device 100. After the computer device 100 obtains the request of the target device 200, it can detect that the screen of the target device 200 is damaged. When the computer device 100 detects that the screen of the target device 200 is damaged, it can obtain device data of the target device 200. According to the device data, the computer device 100 obtains a screen breakage prediction result of the target device 200 through a first detection model. If the risk degree of screen damage represented by the screen breakage prediction result is higher than a target threshold, the computer device 100 obtains a screen detection result of the target device through a second detection model according to a device image of the target device. The screen detection result is used to represent whether the screen is damaged, and the device image is an image obtained by photographing the screen of the target device.

[0033] The screen detection method provided by the embodiment of the present application will be described in detail below in combination with the drawings.

[0034] Please refer to Figure 2 , Figure 2 The flowchart of the screen detection method provided by an embodiment of the present application is shown. In a specific embodiment, the screen detection method is applied to the computer device described above. The screen detection method will be described in detail below with respect to the flowchart shown. Figure 2 The screen detection method can specifically include the following steps: Step S110: Obtain device data of the target device.

[0035] In the embodiment of the present application, for an electronic device (as a target device) whose screen needs to be detected for damage, device data of the target device can be obtained to determine the risk degree of screen damage of the target device according to the device data. The device data of the target device can be data that can reflect the possibility of screen damage of the device.

[0036] In some embodiments, the device data above can include one or more of identity information of the device, sensor data, hardware feature data, and device behavior data. The identity information of the device can include an identity document (ID) of the device, which can uniquely identify the device so as to confirm the eligibility and support scope of the device for relevant services (e.g., purchasing screen break insurance); the sensor data can include data of an accelerometer and a gyroscope of the electronic device during use, which can reflect whether the device has experienced a violent motion or a fall in the past, for example, can be used to analyze whether the device has experienced a violent motion or a fall in the recent past, and can reflect the risk of damage to the screen of the device; the hardware feature data can include at least one of battery health data, storage capacity, and processor performance data, which can reflect the age of the device and thus can reflect the possibility of damage to the screen of the device to some extent; and the device behavior data can include at least one of historical touch data and historical environment data of the environment in which the device is located, the historical touch data can reflect the historical number of times and frequency of touch operations detected at each position of the screen, and the number of times and frequency of touch operations at some positions of the screen are usually matched with those of the screen of the device of another user, so the historical touch data can also reflect the damage to the screen at some positions. The historical environment data above can include geographic location, temperature, humidity, and pressure, and changes in the geographic location, extreme temperature (e.g., change in the thermal zone), humidity, and pressure in which the device is located in the recent past can also affect the screen, so the historical environment data can also reflect the possibility of damage to the screen.

[0037] In some embodiments, the device data above can be sent to the computer device by the target device. The device data above can be data in the recent N days, for example, device data in the recent 7-14 days, and the device data above can be T+1 offline data, i.e., the latest data up to yesterday, or T+0 offline data, i.e., the latest data up to today.

[0038] Step S120: obtaining a screen break prediction result of the target device according to the device data by the first detection model.

[0039] In the embodiments of the present application, after obtaining the device data above, the computer device can obtain the screen break prediction result of the target device according to the device data through the first detection model. The first detection model above can be a pre-trained model that can determine the risk degree of damage to the screen of the device according to the device data, and the screen break prediction result above is used to represent the risk degree of damage to the screen. The screen break prediction result is determined through the first prediction model according to the device data above, so as to determine whether it is necessary to further determine whether the screen of the target device is damaged through visual detection according to the screen break prediction result.

[0040] In some embodiments, the first detection model above can be obtained based on a large language model (LLM). The large language model is a model based on deep learning and natural language processing technology, which is trained using a large amount of text data to learn the understanding and generation ability of language. The large language model can handle complex natural language tasks such as text classification, question answering, and dialogue. The first detection model above can be obtained by fine-tuning a pre-trained large language model according to sample data. Fine-tuning refers to a process of further training (or adjusting parameters) on the basis of a pre-trained large language model to adapt it to a specific task or data set.

[0041] In a possible implementation, when determining the screen break prediction result of the screen through the first detection model above, a prompt word can be input into the first detection model, and the prompt word is used to prompt the large language model to determine the risk degree of damage to the screen according to the device data above, so as to obtain the screen break prediction result output by the first detection model. The prompt word is an input text or instruction provided to the large language model, which is used to guide the large language model to generate a specific type of response. The prompt word converts natural language text into machine-readable intent and embedding vectors, so that the large model can understand and execute human instructions. Specifically, the prompt word can be a question, a description, a task description, or even a part of a dialogue history record, etc. Its function is to guide the model to generate a reply that meets the expectations or complete a specific task.

[0042] In some embodiments, the first detection model is trained according to a large amount of sample device data of electronic devices labeled as having screen damage and sample device data of electronic devices labeled as not having screen damage. Understandably, the sample device data of electronic devices labeled as having screen damage can be used as positive sample data, and the above sample device data of electronic devices labeled as not having screen damage can be used as negative sample data, so that the first detection model capable of predicting the risk degree of screen damage can be trained according to the positive sample data and the negative sample data. The first detection model can be a tree model, a deep neural network model, or the like, and the specific model type of the first detection model is not limited.

[0043] In some embodiments, the above screen damage prediction result can be a result directly representing the high and low of the risk degree of screen damage, for example, the screen damage prediction result can include high risk, medium risk, and low risk; the above screen damage prediction result can also be a probability value of screen damage, which can represent the size of the possibility of screen damage. Of course, the specific form of the screen damage prediction result output by the first detection result is not limited.

[0044] Step S130: If the screen damage prediction result represents that the risk degree of screen damage is higher than a target threshold, obtaining a screen detection result of the target device by a second detection model according to a device image of the target device, the screen detection result being used to represent whether the screen is damaged, and the device image being an image obtained by photographing the screen of the target device.

[0045] In the embodiments of the present application, after obtaining the screen damage prediction result of the screen by the first detection model according to the device data of the target device, if it is determined that the screen damage prediction result represents that the risk degree of screen damage is higher than a target threshold, it means that the risk degree of screen damage of the target device is relatively high. In order to ensure the accuracy of the detection result, further detection can be performed by visual detection. Therefore, if the screen damage prediction result represents that the risk degree of screen damage is higher than a target threshold, an image obtained by photographing the screen of the target device (as a device image) can be obtained, and a screen detection result of the target device can be obtained by a second detection model according to the device image of the target device, and the screen detection result is used to represent whether the screen is damaged.

[0046] In some embodiments, the first detection model can be a visual detection-based model, which can detect whether the screen of the target device is damaged according to the device image by means of computer vision (CV). The first detection model can be trained according to sample device images labeled as having a damaged screen and sample device images labeled as not having a damaged screen. The first detection model can be a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory network (LSTM), a large vision model (LVM), or the like, and the specific type of the first detection model is not limited.

[0047] In some embodiments, the device image can be an image obtained by photographing the screen of the target device using another device. For example, in the scenario of applying for a screen breakage insurance, a user of the target device to be insured can photograph the screen of the target device using another device, and then send the obtained image to the computer device. The device image can also be an image obtained by photographing a target optical lens using the camera of the target device when the screen of the target device faces the target optical lens. The target optical lens can be a mirror, and the camera can be a front camera of the target device.

[0048] In some embodiments, after obtaining the screen breakage prediction result, if the risk degree of screen damage represented by the screen breakage prediction result is lower than or equal to a target threshold, it indicates that the possibility of screen damage of the target device is low, and therefore the second detection model can not be used for further detection in a visual detection manner, and it can be directly determined that the screen of the target device is not damaged.

[0049] In one possible implementation, when the screen breakage prediction result includes a high-risk result, a medium-risk result, and a low-risk result, it can be determined that the risk degree of screen damage is higher than the target threshold when the screen breakage prediction result is the high-risk result or the medium-risk result. It can be determined that the risk degree of screen damage is lower than or equal to the target threshold when the screen breakage prediction result is the low-risk result.

[0050] In a possible implementation, in the case that the above screen breakage prediction result is a probability value of screen breakage, it can be determined that the risk degree of screen breakage is higher than a target threshold in the case that the above probability value is greater than a target probability threshold; and it can be determined that the risk degree of screen breakage is lower than or equal to the target threshold in the case that the above probability value is not greater than the target probability threshold.

[0051] The screen detection method provided by the embodiments of the present application can determine the risk degree of screen breakage, and then detect whether the screen is damaged by using the visual detection method according to the device image only in the case that the risk degree of screen breakage is higher than the target threshold, so that the screen to be detected does not need to be directly detected by using the visual detection method, and therefore the efficiency of screen detection can be improved. In addition, the different detection models are used to detect in different ways, and therefore the accuracy of detecting whether the screen is damaged can be improved.

[0052] Please refer to Figure 3 , Figure 3 A flowchart of a screen detection method provided by another embodiment of the present application is shown. The screen detection method is applied to the computer device described above, and the following will be described in detail with respect to the flowchart shown in Figure 3 The screen detection method can specifically include the following steps: Step S210: Obtain device data of a target device.

[0053] Step S220: Obtain a first prediction result of the screen by a first agent according to the device data, wherein the first agent tends to predict screen breakage.

[0054] In the embodiments of the present application, the first detection model can be a model adopting a debate framework. The first detection model can include a first agent, a second agent, and a third agent. The first agent, the second agent, and the third agent are all assigned corresponding roles. The role corresponding to the first agent and the role corresponding to the second agent can be mutually opposing roles. The first agent tends to predict that the screen is damaged, the first agent tends to predict that the screen is not damaged, and the role corresponding to the third agent is to arbitrate the results output by the first agent and the second agent, that is, the first detection model is a multi-agent based on adversarial debate. The mutually opposing roles corresponding to the first agent and the second agent can perform adversarial analysis and thinking, thereby improving the accuracy of the obtained screen damage prediction result. Understandably, the first agent amplifies the evidence that the screen is damaged, and tries not to miss the judgment. The goal of the first agent is to maximize the recall rate. In contrast to the role of the first agent, the second agent amplifies the evidence that the screen is not damaged and explains why the screen is not damaged, and tries not to make a mistake. The third agent can combine the outputs of the two and the weight of the evidence to give a final conclusion, that is, to give a screen damage prediction result.

[0055] When the computer device determines the screen damage prediction result of the screen of the target device according to the device data of the target device by using the first detection model, the computer device can acquire, by using the first agent, a first prediction result of the screen indicating whether the screen is damaged according to the device data.

[0056] In some embodiments, the first agent is an LLM-based agent. An LLM agent is an artificial intelligence model trained according to a certain algorithm. The agent can process and generate natural language text. These models are usually very large, with tens of billions or even hundreds of billions of parameters, which enables them to capture the complexity and subtleties of language and perform various language tasks such as text summarization, question answering, text generation, translation, and dialogue systems. The first agent can be fine-tuned on the basis of a pre-trained LLM model to determine the risk degree of screen damage according to the input device data and tend to predict screen damage.

[0057] Step S230: acquiring, by using the second agent, a second prediction result of the screen according to the device data, wherein the first agent tends to predict that the screen is not damaged.

[0058] In the embodiments of the present application, when the computer device determines the screen damage prediction result of the screen of the target device according to the device data of the target device by using the first detection model, the computer device can acquire, by using the second agent, a second prediction result of the screen indicating whether the screen is damaged according to the device data.

[0059] In some embodiments, the second intelligent agent is an LLM-based intelligent agent. The second intelligent agent can be fine-tuned on the basis of a pre-trained LLM model to determine the risk degree of screen damage according to the input device data and tend to predict that the screen is not damaged.

[0060] Step S240: determining, by the third intelligent agent, a screen break prediction result of the screen of the target device according to the first prediction result and the second prediction result.

[0061] In the embodiments of the present application, since the first intelligent agent and the second intelligent agent are opposite roles, and the first intelligent agent tends to predict screen damage and the second intelligent agent tends to predict that the screen is not damaged, the first prediction result and the second prediction result can also be opposite, so that the third intelligent agent can determine the screen break prediction result according to the first prediction result and the second prediction result. That is, the third intelligent agent arbitrates the results output by the first intelligent agent and the second intelligent agent to obtain the final screen break prediction result.

[0062] In some embodiments, the first prediction result includes a first result and a first evidence corresponding to the first result, and the second prediction result includes a second result and a second evidence corresponding to the second result. The first result and the second result can both represent the risk degree of screen damage, the first evidence is the evidence given by the first intelligent agent for the first result, for example, the first result is that the risk degree of screen damage is high, and the first evidence corresponding to the first result can be that a falling event from a high place has occurred in history. Similarly, the second evidence is the evidence given by the second intelligent agent for the second result, for example, the second result is that the risk degree of screen damage is high, and the second evidence corresponding to the second result can be that the height of the falling event is within a safe range.

[0063] In addition, different evidences can be pre-determined with corresponding weights, which are used to reflect the credibility of the evidence. If the credibility of the evidence is higher, the weight of the evidence is larger. If the credibility of the evidence is lower, the weight of the evidence is smaller. For example, in the case that there is no touch data of the top of the screen in the historical touch data, there is evidence of abnormal touch response. In general, when a user uses an electronic device, the user will touch the top status bar and other interface elements. Therefore, it can be determined that the top of the screen is abnormal, and the weight corresponding to the evidence can be set to be high. For another example, in the case that the health degree of the battery is low, it indicates that the electronic device is old, but it is not enough to prove that the screen is abnormal. Therefore, the weight corresponding to the evidence of the low health degree of the battery can be set to be low. When the third intelligent agent determines the screen breakage prediction result according to the first prediction result and the second prediction result, the third intelligent agent can determine the screen breakage prediction result according to the first result, the first evidence, the first weight corresponding to the first evidence, the second result, the second evidence, and the second weight corresponding to the second evidence.

[0064] In the above embodiments, the third intelligent agent can directly take the first result (or the second result) and the first evidence (or the second evidence) as the screen breakage prediction result when the first result is consistent with the second result, and the first evidence is consistent with the second evidence. The third intelligent agent can determine the evidence corresponding to the first result and the second result according to the first evidence, the first weight corresponding to the first evidence, the second evidence, and the second weight corresponding to the second evidence as the final output evidence, and take the first result (or the second result) and the determined evidence as the screen breakage prediction result when the first result is consistent with the second result, and the first evidence is inconsistent with the second evidence. The third intelligent agent can determine the final target evidence according to the first evidence, the first weight corresponding to the first evidence, the second evidence, and the second weight corresponding to the second evidence, and determine the target result matched with the evidence, and take the target result and the target evidence as the screen breakage prediction result when the first result is inconsistent with the second result, and the first evidence is inconsistent with the second evidence. In addition, the third intelligent agent can determine the result matched with the first evidence (or the second evidence) again when the first result is inconsistent with the second result, and the first evidence is consistent with the second evidence, and take the result and the first evidence (or the second evidence) as the screen breakage prediction result.

[0065] In some embodiments, in the case that the first agent, the second agent, and the third agent are all LLM-based agents, the pre-trained LLM-based agents can be fine-tuned according to the positive sample data and the negative sample data. The positive sample data is device data of other devices whose screens have been damaged, and the negative sample data is device data of other devices whose screens have not been damaged. When fine-tuning the above LLM-based agents, the pre-trained LLM-based agents can be further trained (or adjusted parameters) based on the above positive sample data and negative sample data to adapt to the tasks to be performed by the above first agent, the second agent, and the third agent. Understandably, when obtaining sample data, it is usually a small probability event that a user's screen is damaged, so the positive sample data is relatively scarce, and the distribution of various screen damage conditions (such as corner cracking, spider web, local failure, etc.) is extremely uneven, and there is a lack of historical data for some models and user devices, thus causing few-shot and zero-shot problems. By fine-tuning the LLM-based agent, only a small amount of labeled sample data is needed to understand what kind of device data can reflect screen damage, and based on the pre-learned knowledge, it can infer the unobserved screen damage conditions, so it can well cope with the few-shot and zero-shot problems.

[0066] Exemplarily, the roles of each agent in the above first agent, second agent, and third agent will be described by way of example.

[0067] Among them, the role corresponding to the first agent can be a screen damage detection expert, the role corresponding to the second agent can be a device state defense expert, and the role corresponding to the third agent can be a neutral arbitrator. The role corresponding to the first agent is opposite to the role corresponding to the second agent.

[0068] The task target corresponding to the first agent is to find all suspicious signs of screen damage, and not to miss any possible damage situation to maximize the recall rate. The first agent can analyze from the dimensions of drop event, touch accuracy, response delay, etc. to determine the suspicious signs of screen damage. The drop event can be determined according to the height, acceleration, surface contact data, etc. The touch accuracy can be determined according to the deviation and abnormal rate reflected by the historical touch data. The response delay can be determined according to the response rate reflected by the historical touch data. Optionally, the result output by the first agent can include a first predicted score of screen damage and evidence of screen damage.

[0069] The task target corresponding to the second agent can be to question the first agent, defend the device, amplify the screen damage-free and explainable evidence, and find a reasonable explanation. The strategy of the second agent questioning the screen damage-free can include at least one of whether the falling is within a safe range, whether the device is equipped with a protective shell, performance factors of the device (such as storage performance, processing performance, etc.), finding positive evidence, and other evidence of screen damage-free. Optionally, the result output by the second agent can include a second prediction score of screen damage-free and evidence of screen damage-free.

[0070] The task target corresponding to the third agent can be to integrate the prediction results output by the first agent and the second agent, and output a fair screen damage prediction result according to the weight corresponding to the evidence. The decision logic of the third agent can be to directly adopt the results output by the first agent and the second agent in the case that the results output by the first agent and the second agent are consistent; in the case that the results output by the first agent and the second agent are inconsistent, the results output by the first agent and the second agent are integrated for arbitration, and the final screen damage prediction result is output. Optionally, the result output by the third agent can include the risk degree of screen damage and the evidence of screen damage.

[0071] In some embodiments, referring to Figure 4 When the screen detection method provided by the embodiments of the present application is applied to the purchase scene of the screen damage insurance, the detection module can be triggered and started in the target device to collect device data and transmit the device data to the computer device. After the computer device obtains the device data, in the case that the data is not empty, the first detection model is used to obtain a screen damage prediction result according to the device data. After the screen damage prediction result is obtained, the computer device can use the first screen damage judgment module to determine whether the screen damage prediction result represents that the risk degree of screen damage is higher than a target threshold. If the screen damage prediction result represents that the risk degree of screen damage is higher than the target threshold, further detection can be performed by using the second detection model. If the screen damage prediction result represents that the risk degree of screen damage is not higher than the target threshold, the target device can be insured. If the obtained device data is empty, further detection can be performed by using the second detection model.

[0072] S250: If the screen damage prediction result represents that the risk degree of screen damage is higher than a target threshold, a screen detection result of the target device is obtained by using the second detection model according to a device image of the target device. The screen detection result is used to represent whether the screen is damaged, and the device image is an image obtained by photographing the screen of the target device.

[0073] In the embodiments of the present application, step S250 can refer to the content of other embodiments, which will not be described here.

[0074] The screen detection method provided in the embodiments of the present application determines whether the screen is damaged by visual detection according to the device image only when it is determined that the risk degree of screen damage is higher than the target threshold, so that the screen to be detected does not need to be directly detected by visual detection, thereby improving the efficiency of screen detection. In addition, the model used to determine the risk degree of screen damage is a multi-agent based on adversarial debate, and the mutually opposed agents can perform adversarial analysis, thereby improving the accuracy of the determined risk degree of screen damage.

[0075] Please refer to Figure 5 , Figure 5 The flowchart of the screen detection method provided in another embodiment of the present application is shown. The screen detection method is applied to the computer device described above, and the following will be described in detail with respect to the flowchart shown in Figure 5 The screen detection method can specifically include the following steps: Step S310: Obtain device data of a target device.

[0076] In the embodiments of the present application, step S310 can refer to the content of other embodiments, which will not be described here.

[0077] Step S320: Convert the device data into natural language to obtain target text information.

[0078] In the embodiments of the present application, the first detection model can be obtained based on a large language model. When determining the screen breakage prediction result of the screen by the first detection model according to the device data, since most of the data in the device data is structured data, and the first detection model is a large language model, the device data can be converted into natural language to obtain target text information, so that the first detection model can better understand the state of the screen of the device according to the target text information. Since the first detection model is a large language model, it does not need to train the model with a large amount of sample data, and can well cope with the problem of few samples and zero samples.

[0079] In some embodiments, in order to better improve the understanding ability of the first detection model to the device data, the low-level and context-free structured data (i.e. the above device data) can be promoted to high-level and semantic information, thereby activating the reasoning ability of the large language model. The above device data can be converted into target text information by a rule-based conversion method.

[0080] In a possible implementation, a target template for converting the device data into the target text information can be determined; when the device data is converted into the target text information, the obtained device data can be preprocessed and filled into the target template, i.e., instantiated, to obtain the target text information. The preprocessing of the device data can include at least one of filtering invalid data, data classification, and data statistics.

[0081] Optionally, referring to FIG. 8, Figure 6 As shown in FIG. 8, when determining a target template for converting the device data into the target text information for inputting into the first detection model, text planning can be performed first to determine core information to be expressed and a logical order of the core information; then sentence planning is performed to determine a specific sentence structure and grammar for the core information, to obtain the target template. After obtaining the target template, instantiation can be performed to obtain the target text information for inputting into the first detection model.

[0082] The following examples are used to illustrate the conversion of device data into target text information by using a rule-based conversion method.

[0083] For example, a rule case can be: Recent days change Touch screen change: Touch response mean rises from {touch_response_mean_ms1} ms to {touch_response_mean_ms2} ms (Δ{touch_fluctuation} ms), and slow response ratio rises from {slow_response_ratio1} % to {slow_response_ratio2} % (Δ{slow_fluctuation} %); Drop / impact: records {drop_event_count} times, respectively, at {datetime1} and {datetime2}; Usage and aging: charging frequency increased from {charging_frequency1} times to { charging_frequency1} times / day (Δ{charging_fluctuation} times), single use duration decreased from { duration_use1} hours to { duration_use2} hours (Δ{duration_fluctuation} hours), and the range of environmental temperature changes in the area involved in the recent trajectory expanded from {temperature_changes1} to {temperature_changes2} (Δ{temperature_fluctuation} temperature width).

[0084] Application cases can be: Changes in the past 7 days Touch screen changes: average touch response time increased from 31.0 ms to 43.5 ms (Δ +12.5 ms), and the proportion of slow response increased from 8% to 18% (Δ +10%); Drop / impact: recorded 2 times (max 22g), on October 6, 2025 and October 10, 2025; Usage and aging: charging frequency increased from 1.2 times to 1.8 times / day (Δ +0.6 times), single use duration decreased from 12.5 hours to 9.3 hours (Δ -3.2 hours), and the range of environmental temperature changes in the area involved in the recent trajectory expanded from 15-28℃ to -5℃ to 35℃ (Δ 40℃ temperature width).

[0085] In the above example, the rule case is the above target template determined by the rule-based conversion method, and the application case is the target text information obtained by filling the preprocessed device data into the above target template.

[0086] Step S330: input the target text information into the first detection model to obtain the screen breakage prediction result output by the first detection model, wherein the first detection model is obtained based on a large language model.

[0087] In the embodiments of the present application, after obtaining the target text information obtained by converting the device data, the target text information can be input into the first detection model, so as to obtain the screen breakage prediction result output by the first detection model.

[0088] Step S340: if the screen breakage prediction result represents that the risk degree of damage of the screen is higher than a target threshold, obtaining a screen detection result of the target device by a second detection model according to a device image of the target device, wherein the screen detection result is used to represent whether the screen is damaged, and the device image is an image obtained by photographing the screen of the target device.

[0089] In the embodiment of the present application, step S340 can refer to the content of other embodiments, which will not be repeated here.

[0090] It should be noted that the screen detection method provided in the embodiment of the present application can be combined with the screen detection method provided in the previous embodiment. For example, the first agent, the second agent and the third agent in the previous embodiment can be LLM-based agents, which can convert device data into natural language, obtain target text information, and then input the target text information into the first agent and the second agent to obtain the first prediction result output by the first agent and the second prediction result output by the second agent, and then determine the screen break prediction result according to the first prediction result and the second prediction result through the third agent.

[0091] The screen detection method provided in the embodiment of the present application determines that the risk degree of screen damage is higher than the target threshold value, and then detects whether the screen is damaged according to the device image in a visual detection manner, so that the screen to be detected does not need to be directly detected by the visual detection manner, thereby better improving the efficiency of screen detection. In addition, the first detection model is obtained based on a large language model, so that a large number of sample data are not needed to train the model, and the problem of few samples and zero samples can be better solved, thereby reducing the implementation difficulty and cost.

[0092] Please refer to Figure 7 , Figure 7 A flowchart of a screen detection method provided in another embodiment of the present application is shown. The screen detection method is applied to the above computer device, and the following will be described in detail with respect to the flowchart shown in Figure 7 The screen detection method can specifically include the following steps: Step S410: Obtain device data of a target device.

[0093] Step S420: Obtain a screen break prediction result of a screen of the target device by a first detection model according to the device data.

[0094] In the embodiment of the present application, steps S410 and S420 can refer to the content of other embodiments, which will not be repeated here.

[0095] Step S430: If the screen break prediction result represents that the risk degree of screen damage is higher than a target threshold value, obtain a damage detection result of the screen by each visual detection model in a plurality of visual detection models according to a device image of the target device, and obtain a plurality of damage detection results.

[0096] In the embodiments of the present application, the second detection model can include a voting agent and a plurality of visual detection models. The visual detection model is a model for detecting whether the screen is damaged by means of computer vision. The voting agent is used to aggregate the detection results output by the plurality of visual detection models, perform voting, and determine the final screen detection result according to the number of votes. When the computer device determines the screen detection result by using the second detection model, the device image can be input to each visual detection model to obtain the damage detection result output by each visual detection model. By using a plurality of visual detection models to respectively detect whether the screen is damaged according to the device image, accidental misjudgment (for example, interference caused by occlusion, glare artifact) of a single model can be avoided, and the accuracy of screen detection can be improved.

[0097] In some embodiments, the above visual detection model can be a large visual model. The large visual model is a large-scale model trained and inferred based on pixel-level visual data. The large visual model draws on two features of a large language model (large-scale training and context prompt). The large visual model aims to learn general visual knowledge and adapt to multiple visual tasks and scenes.

[0098] In a possible implementation, the plurality of visual detection models above can be large visual models of different manufacturers. The number of the plurality of visual detection models is at least 3 (i.e., greater than or equal to 3), and the number of the plurality of visual detection models is odd, so that when the damage detection results output by the plurality of visual detection models are voted, the damage detection result with the most votes can be determined, and the situation where the number of two detection results is the same can be avoided.

[0099] In some embodiments, the visual detection model can output a detection result (as a sub-detection result) for representing whether the screen is damaged, and can output corresponding description information in the case that the sub-detection result represents that the screen is damaged. The description information can include a damage type of the screen existing damage, and evidence corresponding to the damage type. The damage type can include: existing cracks, scratches, light distortion, bad points, color spots, green lines, screen edge lifting, and screen fragmentation. The evidence corresponding to the damage type can include specific features, positions, and degrees of the damage type. Correspondingly, the visual detection model can check from multiple dimensions according to the device image. The multiple dimensions can include crack detection, scratch detection, detection of abnormal gloss reflection, detection of display abnormalities, and detection of structural integrity. The crack detection can include detection of the type, length, and position of the crack. The scratch detection can include detection of the visibility, number, and area of the scratch. The detection of abnormal gloss reflection can include detection of light distortion and interruption. The detection of structural integrity can include detection of screen edge lifting and fragmentation.

[0100] Step S440: determining, by the voting intelligent agent, a screen detection result of the target device according to the voting principle and the plurality of damage detection results, the screen detection result being used to represent whether the screen is damaged, and the device image being an image obtained by photographing the screen of the target device.

[0101] In the embodiments of the present application, after obtaining the damage detection results output by each of the above visual detection models, the voting intelligent agent can determine the screen detection result of the target device based on the voting rule and the plurality of damage detection results. That is, the voting intelligent agent can vote on different damage detection results, and obtain the screen detection result based on the damage detection result with the most votes.

[0102] In some embodiments, the damage detection result includes a sub-detection result used to represent whether the screen is damaged, and description information in the case that the sub-detection result represents that the screen is damaged. The voting intelligent agent can determine the sub-detection result with the most votes as the screen detection result of the target device according to the voting principle and the plurality of damage detection results, and determine the evidence information that the screen is damaged based on the description information corresponding to the sub-detection result with the most votes in the case that the screen detection result represents that the screen is damaged.

[0103] In the above embodiments, for the sub-detection results in the plurality of damage detection results, the first sub-detection result representing that the screen is damaged and the second sub-detection result representing that the screen is not damaged can be voted on. If the sub-detection result in any damage detection result is the first sub-detection result, the number of votes for the first sub-detection result is increased by 1, and if the sub-detection result in any damage detection result is the second sub-detection result, the number of votes for the second sub-detection result is increased by 1. After traversing each sub-detection result in the plurality of damage detection results, the number of votes for the first sub-detection result and the number of votes for the second sub-detection result can be obtained, and the sub-detection result with the most votes can be determined as the screen detection result of the target device.

[0104] In addition, after the screen detection result is determined, in the case that the screen detection result represents that the screen is damaged, the evidence information that the screen is damaged can also be determined according to the description information corresponding to the above sub-detection result with the most votes. It can be understood that the voting intelligent agent can determine the damage type of the screen, the specific features of the damage type, the damage location and the damage degree as the evidence information that the screen is damaged by comprehensively considering the description information corresponding to the sub-detection result with the most votes.

[0105] For example, the result obtained by the second detection model after detecting the device image with the picture number 1 is shown in the following table, The screen detection result represents whether the screen is damaged, i.e., whether the screen is damaged, a fault type, and a detection process are evidence information in the case of damage of the screen.

[0106] Optionally, in the case that the screen detection result represents damage of the screen, the device image can also be saved. When the computer device receives an evidence query request of the target device, the computer device can return the evidence information and the device image to the target device for viewing by a user corresponding to the target device.

[0107] In some embodiments, the screen detection method provided by the embodiments of the present application is applied to a purchase scenario of a screen breakage insurance. Please refer to Figure 8 After the machine starts, a device image can be captured and provided to the computer device. The computer device can obtain a screen detection result by using a second detection model and according to the device image. After obtaining the screen detection result, the computer device can determine whether the screen detection result represents damage of the screen by using a second screen breakage judgment module. If the screen detection result represents damage of the screen, the target device can be refused to purchase the screen breakage insurance. If the screen detection result represents that the screen is not damaged, the target device can be insured.

[0108] The screen detection method provided by the embodiments of the present application determines the risk degree of damage of the screen. If the risk degree of damage of the screen is higher than a target threshold, the screen detection method detects whether the screen is damaged according to the device image in a visual detection manner. Therefore, the screen to be detected does not need to be directly detected in the visual detection manner, and thus the efficiency of screen detection can be improved. In addition, when the screen detection method detects whether the screen is damaged according to the device image in the visual detection manner, a plurality of visual detection models detect according to the device image, and the final screen detection result is determined by using a voting manner. Therefore, the accuracy of screen detection can be improved.

[0109] Please refer to Figure 9 , Figure 9 A flowchart of a screen detection method provided by another embodiment of the present application is shown. The screen detection method is applied to the computer device described above. The screen detection method will be described in detail below with reference to the flowchart shown in Figure 9 The screen detection method can specifically include the following steps: Step S510: Obtain device data of a target device.

[0110] Step S520: Obtain a screen breakage prediction result of the target device by using a first detection model according to the device data.

[0111] In the embodiments of the present application, steps S510 and S520 can refer to the content of other embodiments, which will not be described herein again.

[0112] Step S530: If the screen break prediction result represents that the risk degree of the screen break is higher than the target threshold, identity verification is performed on the obtained device image to obtain a verification result.

[0113] In the embodiments of the present application, after the above screen break prediction result is obtained, if the screen break prediction result represents that the risk degree of the screen break is higher than the target threshold, a device image obtained by photographing the screen of the target device can be obtained, and identity verification can be performed on the obtained device image to obtain a verification result, so as to verify whether the device image belongs to the target device, i.e., whether the device in the device image is the target device, thereby avoiding obtaining a fake image of the target device and ensuring the authenticity of the obtained device image.

[0114] In some embodiments, the computer device can send verification content to be displayed to the target device; the target device can display the verification content after receiving the verification content, photograph the screen of the target device on which the verification content is displayed to obtain a device image, and then send the device image to the computer device; the computer device can determine whether the verification content in the image matches the verification content sent by the computer device to the target device according to the region of the screen in the device image; if they match, it can be determined that the verification result is a first result representing that the device image is obtained by photographing the screen of the target device; if they do not match, it can be determined that the verification result is a first result representing that the device image is not obtained by photographing the screen of the target device. In this way, the authenticity of the device image sent by the client can be ensured, and the situation that the client forges a device image can be avoided.

[0115] In a possible implementation, the above verification content can include at least one of a two-dimensional code, a verification code, and a verification pattern.

[0116] In a possible implementation, the above verification content can include a marker image, and the marker image includes at least one sub-marker. The sub-marker can be a pattern having a certain shape. Each sub-marker can have one or more feature points, where the shape of the feature point is not limited and can be a dot, a ring, a triangle, or other shapes. In addition, the distribution rules of the sub-markers in different markers are different, so each marker can have different identity information. When generating the above marker image, a corresponding marker image can be generated according to the device identifier of the target device or the user identifier corresponding to the target device, so that the identity information corresponding to the combination of the sub-markers in the marker image corresponds to the device identifier or the user identifier of the target device. In this way, the marker images sent for different devices can be different, thereby better avoiding the situation of forging a device image.

[0117] In some embodiments, the target device can display an inspection machine control in the interface; after detecting an operation on the inspection machine control, guidance information for guiding the shooting of the above device image can be displayed to guide the user to use other devices to shoot the target device, or to shoot the target device against a mirror, and after the device image is obtained, the device image is sent to the above computer device.

[0118] Step S540: If the verification result represents that the device image is an image obtained by shooting the screen of the target device, a screen detection result of the target device is obtained from the device image by a second detection model.

[0119] In the embodiments of the present application, after obtaining the above verification result, if the verification result represents that the device image is an image obtained by shooting the screen of the target device, the device image is a real image of the target device, and therefore a screen detection result of the target device can be obtained from the device image by a second detection model.

[0120] The screen detection method provided in the embodiments of the present application determines whether the risk degree of screen damage is higher than a target threshold, and then detects whether the screen is damaged by visual detection according to the device image, so that the screen to be detected does not need to be directly detected by visual detection, thereby improving the efficiency of screen detection. In addition, the device image is verified when detected by visual detection, so that the device image is not falsified, the authenticity of the device image is ensured, and the authenticity of the screen detection result is ensured.

[0121] Please refer to Figure 10 , Figure 10 A flowchart of a screen detection method provided in another embodiment of the present application is shown. The screen detection method is applied to the above computer device, and the following will be described in detail with respect to the flowchart shown in Figure 10 The screen detection method can specifically include the following steps: Step S610: In response to an insurance application request for screen damage initiated by the target device, device data of the target device is obtained.

[0122] In the embodiments of the present application, the screen detection method provided in the embodiments of the present application is applied to the purchase scene of screen damage insurance, and the target device can initiate an insurance application request for screen damage to the computer device. Correspondingly, the computer device can obtain the insurance application request for screen damage initiated by the target device. After obtaining the insurance application request, the device data of the target device can be obtained.

[0123] In some embodiments, the target device can display a purchase interface of the screen break insurance; according to a detected purchase operation in the purchase interface, a computer device can be initiated to request a purchase request of the screen break insurance of the target device. And the target device can trigger a starting detection module, the detection module can collect device data, and send the device data to the computer device, and correspondingly, the computer device can obtain the device data of the target device.

[0124] Step S620: obtaining a screen break prediction result of the target device according to the device data through a first detection model.

[0125] Step S630: if the screen break prediction result represents that the risk degree of the screen break is higher than a target threshold, obtaining a screen detection result of the target device according to a device image of the target device through a second detection model, the screen detection result is used to represent whether the screen is damaged, and the device image is an image obtained by shooting the screen of the target device.

[0126] In the embodiments of the present application, steps S620 and S630 can refer to the content of other embodiments, which will not be repeated here.

[0127] Step S640: if the screen detection result represents that the screen is not damaged, determining that the underwriting result of the purchase request is a first result, and the first result is used to represent that the underwriting is passed.

[0128] Step S650: if the screen detection result represents that the screen is damaged, determining that the underwriting result of the purchase request is a second result, and the second result is used to represent that the underwriting is not passed.

[0129] In the embodiments of the present application, after obtaining the screen detection result of the target device, according to the obtained screen detection result; if the screen detection result represents that the screen is not damaged, it can be determined that the underwriting result of the purchase request is a first result, and the first result is used to represent that the underwriting is passed; if the screen detection result represents that the screen is not damaged, it can be determined that the underwriting result of the purchase request is a second result, and the second result is used to represent that the underwriting is not passed.

[0130] In some embodiments, after determining the above underwriting result, it can be determined whether to insure the screen break insurance of the above target device according to the underwriting result. If the underwriting result is the first result, the insurance can be performed; if the underwriting result is the second result, the insurance is not performed.

[0131] The screen detection method provided in the embodiments of the present application can be applied to a purchase scenario of a screen damage insurance. In the case that the risk degree of screen damage is higher than a target threshold, the screen damage is detected in a visual detection manner according to a device image, so as to improve the underwriting efficiency.

[0132] The screen detection method related to the foregoing embodiments is described below in combination with a purchase scenario of a screen damage insurance.

[0133] As shown in Figure 11 When the target device purchases the screen damage insurance, the device data can be collected by the starting detection module, and the device data is transmitted to the computer device. The computer device can obtain a screen damage prediction result of the target device according to the device data by using the first detection model. The computer device determines whether the risk degree of screen damage is higher than a target threshold according to the screen damage prediction result by using the first screen damage judgment module. If the risk degree is not higher than the target threshold, the insurance is underwritten. If the risk degree is higher than the target threshold, the screen is detected in a visual detection manner according to a device image of the target device by using the second detection model, so as to obtain a screen detection result. The computer device determines whether the screen is damaged according to the screen detection result by using the second screen damage judgment module. If the screen is not damaged, the insurance is underwritten. If the screen is damaged, the insurance is not underwritten. By using the screen detection method provided in the embodiments of the present application, the error rate of screen detection can be effectively reduced by two-stage detection. When the screen is detected in a visual detection manner, a plurality of visual detection models are used for detection, so as to accurately identify the damage of the screen, and the accuracy and robustness of screen damage identification are significantly improved. In the detection process, manual intervention is not required, and the screen detection is automated and intelligent, so as to significantly improve the detection efficiency and response speed, thereby improving the user satisfaction and effectively reducing the cost of manual detection and auditing.

[0134] Please refer to Figure 12 which shows a structural block diagram of a screen detection apparatus 700 provided in the embodiments of the present application. The screen detection apparatus 700 uses the computer device described above. The screen detection apparatus 700 includes a data acquisition module 710, a first detection module 720, and a second detection module 730. The data acquisition module 710 is configured to acquire device data of a target device. The first detection module 720 is configured to obtain a screen damage prediction result of the target device according to the device data by using a first detection model. The second detection module 730 is configured to obtain a screen detection result of the target device according to a device image of the target device by using a second detection model, if the screen damage prediction result indicates that the risk degree of screen damage is higher than a target threshold. The screen detection result is used to indicate whether the screen is damaged. The device image is an image obtained by photographing the screen of the target device.

[0135] In some embodiments, the first detection model comprises a first agent, a second agent, and a third agent. The first detection module 720 can be specifically configured to obtain, by the first agent, a first prediction result of the screen according to the device data, wherein the first agent tends to predict that the screen is damaged; obtain, by the second agent, a second prediction result of the screen according to the device data, wherein the first agent tends to predict that the screen is not damaged; and determine, by the third agent, a screen break prediction result of the screen according to the first prediction result and the second prediction result.

[0136] In a possible implementation, the first prediction result comprises a first result and a first evidence corresponding to the first result, and the second prediction result comprises a second result and a second evidence corresponding to the second result. The first detection module 720 can be specifically configured to determine, by the third agent, a screen break prediction result of the screen according to the first result, the first evidence, a first weight corresponding to the first evidence, the second result, the second evidence, and a second weight corresponding to the second evidence.

[0137] In some embodiments, the second detection model comprises a voting agent and a plurality of visual detection models. The second detection module 730 can be specifically configured to obtain, by each visual detection model in the plurality of visual detection models, a damage detection result of the screen according to a device image of the target device, to obtain a plurality of damage detection results; and determine, by the voting agent, a screen detection result of the target device according to a voting principle and the plurality of damage detection results.

[0138] In a possible implementation, the damage detection result comprises a sub-detection result for characterizing whether the screen is damaged, and description information in a case where the sub-detection result characterizes that the screen is damaged. The second detection module 730 can be specifically configured to determine, by the voting agent, a sub-detection result with the most votes as the screen detection result of the target device according to a voting principle and the plurality of damage detection results, and determine, in a case where the screen detection result characterizes that the screen is damaged, evidence of the screen being damaged based on description information corresponding to the sub-detection result with the most votes.

[0139] In some embodiments, the first detection model is obtained based on a large language model, and the first detection module 720 can be specifically configured to convert the device data into natural language to obtain target text information; and input the target text information into the first detection model to obtain the screen break prediction result of the screen output by the first detection model.

[0140] In some embodiments, the second detection module 730 can be specifically configured to authenticate the obtained device image, to obtain an authentication result; if the authentication result indicates that the device image is an image obtained after the screen of the target device is photographed, the second detection model is used to obtain the screen detection result of the target device according to the device image.

[0141] In some embodiments, the screen detection apparatus 700 can further include an underwriting result determination module. The data acquisition module 710 can be specifically configured to acquire the device data of the target device in response to an insurance application request of the screen breakage insurance initiated by the target device. The underwriting result determination module can be specifically configured to determine an underwriting result of the insurance application request as a first result if the screen detection result indicates that the screen is not damaged, the first result being used to indicate that the underwriting is passed; and determine the underwriting result of the insurance application request as a second result if the screen detection result indicates that the screen is damaged, the second result being used to indicate that the underwriting is failed.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described apparatuses and modules can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein.

[0143] In several embodiments provided in the present application, the coupling between the modules can be electrical, mechanical or other forms of coupling.

[0144] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically independently, or two or more modules can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software functional module.

[0145] To sum up, the scheme provided in the present application acquires the device data of the target device, acquires the screen breakage prediction result of the screen of the target device according to the device data through the first detection model, acquires the screen detection result of the target device according to the device image of the target device through the second detection model if the screen breakage prediction result indicates that the risk degree of screen damage is higher than a target threshold, and the screen detection result is used to indicate whether the screen is damaged, and the device image is an image obtained by photographing the screen of the target device. Thus, in the case where it is determined that the risk degree of screen damage is higher than the target threshold, the screen is detected according to the device image in the visual detection manner only when the risk degree of screen damage is higher than the target threshold, so that the screen to be detected does not need to be directly detected in the visual detection manner, and therefore the efficiency of screen detection can be improved better.

[0146] For reference Figure 13FIG. 1 is a structural block diagram of a computer device according to an embodiment of the present application. The computer device 100 can include one or more of the following components: a processor 110, a memory 120, and one or more application programs, wherein the one or more application programs can be stored in the memory 120 and configured to be executed by the one or more processors 110, and the one or more application programs are configured to perform the method described in the foregoing method embodiments.

[0147] The processor 110 can include one or more processing cores. The processor 110 connects various parts within the computer device 100 by various interfaces and lines, performs various functions of the computer device 100 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 120, and calling data stored in the memory 120. Optionally, the processor 110 can be implemented in at least one of a hardware form of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 110 can integrate a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU mainly processes operating systems, user interfaces, and application programs, etc.; the GPU is responsible for rendering and drawing display content; and the modem is used for processing wireless communication. It can be understood that the above-mentioned modem can also not be integrated into the processor 110, but be implemented by a separate communication chip.

[0148] The memory 120 can include a random access memory (RAM) and can also include a read-only memory (ROM). The memory 120 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 120 can include a program storage area and a data storage area, wherein the program storage area can store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing each of the following method embodiments, etc. The data storage area can also store data created by the computer device 100 in use (such as a phone book, audio and video data, chat record data, etc.).

[0149] Reference is made to Figure 14FIG. 8 is a structural block diagram of a computer readable storage medium according to an embodiment of the present application. The computer readable medium 800 stores program codes, which can be invoked by a processor to execute the methods described in the above method embodiments.

[0150] The computer readable storage medium 800 can be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read-Only Memory), an EPROM, a hard disk, or a ROM. Alternatively, the computer readable storage medium 800 comprises a non-transitory computer readable medium. The computer readable storage medium 800 has a storage space for program codes 810 to execute any of the above methods. These program codes can be read from or written to one or more computer program products. The program codes 810 can be compressed in an appropriate form, for example.

[0151] The embodiments of the present application further provide a computer program product comprising a computer program, which, when executed by a processor, implements the screen detection method provided by the above embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate but not limit the technical solutions of the present application; even though the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or equivalently replace some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A screen detection method characterized by, The method comprises: obtaining device data of a target device; obtaining a screen breakage prediction result of a screen of the target device according to the device data by a first detection model; if the screen breakage prediction result represents that a risk degree of damage of the screen is higher than a target threshold, obtaining a screen detection result of the target device according to a device image of the target device by a second detection model, the screen detection result being used to represent whether the screen is damaged, the device image being an image obtained by photographing the screen of the target device.

2. The method of claim 1, wherein, The first detection model comprises a first agent, a second agent and a third agent, and the obtaining of the screen breakage prediction result of the screen of the target device according to the device data by the first detection model comprises: obtaining a first prediction result of the screen according to the device data by the first agent, wherein the first agent tends to predict that the screen is damaged; obtaining a second prediction result of the screen according to the device data by the second agent, wherein the first agent tends to predict that the screen is not damaged; determining the screen breakage prediction result of the screen according to the first prediction result and the second prediction result by the third agent.

3. The method of claim 2, wherein, The first prediction result comprises a first result and first evidence corresponding to the first result, the second prediction result comprises a second result and second evidence corresponding to the second result, and the determining of the screen breakage prediction result according to the first prediction result and the second prediction result by the third agent comprises: determining the screen breakage prediction result according to the first result, the first evidence, a first weight corresponding to the first evidence, the second result, the second evidence and a second weight corresponding to the second evidence by the third agent.

4. The method of claim 1, wherein, The second detection model comprises a voting agent and a plurality of visual detection models, and the obtaining of the screen detection result of the target device according to the device image of the target device by the second detection model comprises: obtaining damage detection results of the screen according to the device image of the target device by each visual detection model in the plurality of visual detection models to obtain a plurality of damage detection results; determining the screen detection result of the target device according to a voting principle and the plurality of damage detection results by the voting agent.

5. The method of claim 4, wherein, The damage detection result comprises a sub-detection result used to represent whether the screen is damaged and description information in a case where the sub-detection result represents that the screen is damaged, and the determining of the screen detection result of the target device according to the voting principle and the plurality of damage detection results by the voting agent comprises: determining, as the screen detection result of the target device, a sub-detection result with the most votes according to the voting principle and the plurality of damage detection results, and determining, in a case where the screen detection result represents that the screen is damaged, evidence information of damage of the screen based on description information corresponding to the sub-detection result with the most votes.

6. The method of claim 1, wherein, The first detection model is obtained based on a large language model, and the screen cracking prediction result of the screen of the target device is obtained by the first detection model according to the device data, including: Converting the device data into natural language to obtain target text information; The target text information is input into the first detection model to obtain the screen cracking prediction result output by the first detection model.

7. The method according to any one of claims 1 to 6, characterized in that, Before the screen detection result of the target device is obtained by the second detection model according to the device image of the target device, the method further includes: Identity verification is performed on the obtained device image to obtain a verification result; If the verification result indicates that the device image is an image obtained after the screen of the target device is photographed, a screen detection result of the target device is obtained by the second detection model according to the device image.

8. The method according to any one of claims 1 to 6, characterized in that, The device data of the target device is obtained, including: In response to an insurance application request of the target device for a screen cracking insurance, device data of the target device is obtained; After the screen detection result of the target device is obtained by the second detection model according to the device image of the target device, the method further includes: If the screen detection result indicates that the screen is not damaged, a first result is determined as the underwriting result of the insurance application request, and the first result is used to represent that the underwriting is passed; If the screen detection result indicates that the screen is damaged, a second result is determined as the underwriting result of the insurance application request, and the second result is used to represent that the underwriting is not passed.

9. A computer device, comprising: Including: One or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method of any one of claims 1-8.

10. A computer program product, characterised in that, The computer program product includes a computer program, and the computer program is executed by a processor to implement the method of any one of claims 1-8.

Images