AI vision module identifies oil smoke accuracy rate detection method and device and range hood

Abstract

The application provides an AI vision module oil fume recognition accuracy detection method and device and a range hood, the AI vision module oil fume recognition accuracy detection method obtains n oil fume sample images collected by an AI vision module in a cooking process of a cooking appliance, determines the number of negative samples, the number of sample misrecognition and the number of effective samples in the n oil fume sample images, then calculates the misrecognition rate of the corresponding negative samples and the effective recognition rate of the effective samples based on the determined values, and finally determines whether the AI vision module is qualified for recognizing oil fume based on the misrecognition rate and the effective recognition rate. The application processes the collected multiple oil fume sample images to obtain the numbers of multiple invalid and effective samples, further calculates the misrecognition rate and the effective recognition rate and determines the accuracy based on the calculated values, so that higher detection precision and efficiency are achieved, the recognition error is reduced, automatic detection, quantitative grading and reduction of artificial cost are realized, and the detection precision and efficiency are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of kitchen appliance technology, and in particular to a method, device, and range hood for detecting the accuracy of oil fume recognition using an AI vision module. Background Technology

[0002] Kitchen appliances such as range hoods and cooktops are essential appliances in modern kitchens, providing convenience and assistance for users' cooking. The range hood is an important component of the range hood and cooktop, responsible for removing cooking fumes from the kitchen.

[0003] With the diversification of cooking methods, the amount of cooking fumes varies. When the fumes are heavy, people often need to manually operate the range hood to increase the setting to prevent them from escaping. Conversely, when the fumes are light, they need to manually decrease the setting to avoid wasting energy. This method has low automation, suffers from lag in manual adjustment leading to fume escape and energy waste, and the frequent manual operation during cooking results in a relatively poor user experience.

[0004] In response, some manufacturers have launched range hoods with AI visual fume recognition capabilities. These hoods use AI vision technology to identify the amount of cooking fumes and adjust the fan speed accordingly, significantly improving the user experience. To ensure reliability, the AI ​​visual fume recognition function needs performance verification, but currently this can only be done through actual cooking tests and manual judgment. However, manual judgment, based on individual testers' subjective perception of the relationship between fume levels and fan speed adjustments, can lead to significant variations in test results due to individual differences. Furthermore, the verification testing process is time-consuming and inefficient. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0006] Therefore, one objective of this invention is to propose an accuracy detection method for AI vision module recognition of cooking fumes. This method first acquires multiple cooking fume sample images collected by the AI ​​vision module during cooking on a stove. It then determines the number of negative cooking fume sample images, the number of misidentified negative cooking fume sample images, and the number of valid cooking fume sample images. Based on these determined values, it calculates the misidentification rate of the corresponding negative cooking fume sample images and the effective recognition rate of the valid cooking fume sample images. Finally, based on the misidentification rate and the effective recognition rate, it determines whether the AI ​​vision module's recognition of cooking fumes is qualified. Compared to traditional manual judgment methods, this AI vision-based cooking fume accuracy detection method quantifies and calibrates the amount of cooking fumes in the cooking scene, reducing errors from human perception and significantly reducing manual labor, thus improving detection efficiency. Furthermore, by recognizing and summarizing multiple cooking fume sample images and calculating the recognition ratio, unlike instantaneous judgment based on a single image, it avoids accidental interference from lighting and fumes in a single frame, reducing the random error of single recognition and effectively enhancing the reliability and accuracy of recognition. By leveraging automated data comparison and accuracy statistics mechanisms, it avoids discrepancies caused by subjective human judgment, while reducing manual operation costs. This overcomes the pain points of low testing efficiency and difficulty in data reproduction, comprehensively improving detection accuracy and testing effectiveness.

[0007] Therefore, the second objective of this invention is to provide an accuracy detection device for AI vision module to identify cooking fumes.

[0008] Therefore, the third objective of this invention is to provide a smoke hood.

[0009] To achieve the above objectives, a first aspect of the present invention provides a method for detecting the accuracy of an AI vision module in identifying cooking fumes, used in a range hood. The method includes: acquiring n cooking fume sample images collected by the AI ​​vision module during cooking on a stove; determining the number of negative cooking fume sample images and the number of misidentified negative cooking fume sample images among the n images, and determining the number of valid cooking fume sample images among the n images; determining the misidentification rate of the negative cooking fume sample images based on the number of negative images and the number of misidentified negative images, and determining the valid identification rate based on the number of valid cooking fume sample images; and determining whether the AI ​​vision module's identification of cooking fumes is qualified based on the misidentification rate and the valid identification rate.

[0010] According to an embodiment of the present invention, the accuracy detection method for identifying cooking fumes using an AI vision module first acquires n cooking fume sample images collected by the AI ​​vision module during cooking on a stove. The number of negative cooking fume sample images, the number of misidentified negative cooking fume sample images, and the number of valid cooking fume sample images are determined from the n images. Then, the misidentification rate of the corresponding negative cooking fume sample images is calculated based on the number of negative cooking fume sample images and the number of misidentified negative images. The effective recognition rate is calculated based on the number of valid cooking fume sample images. Finally, the accuracy of the AI ​​vision module's identification of cooking fumes is determined based on the misidentification rate and the effective recognition rate. This method obtains multiple... The number of invalid and valid samples is used to further calculate the false recognition rate and valid recognition rate. Based on the calculated values, the accuracy of the AI ​​vision module in recognizing cooking fumes is judged. Compared with traditional manual judgment methods, this AI vision recognition method for cooking fumes quantifies and calibrates the amount of cooking fumes in the cooking scene, reducing errors from human perception and significantly reducing manual labor, thus improving detection efficiency. Furthermore, it identifies and summarizes multiple cooking fume sample images, calculating the recognition ratio. Unlike instantaneous judgment based on a single image, this method avoids accidental interference from lighting and fumes in a single frame, reducing random errors in single recognition and effectively enhancing the reliability and accuracy of recognition. With its automated data comparison and accuracy statistics mechanism, it avoids the results differences caused by subjective human judgment, while reducing manual operation costs. This overcomes the pain points of low testing efficiency and difficulty in data reproduction, comprehensively improving detection accuracy and testing effectiveness.

[0011] In some embodiments, determining the number of valid oil fume sample images among the n oil fume sample images includes: determining the oil fume level calibration value and oil fume level detection value of each oil fume sample image; calculating the absolute value of the difference between the oil fume level calibration value and the oil fume level detection value of each oil fume sample image; oil fume sample images whose absolute value of the difference satisfies a preset condition are taken as valid oil fume sample images, and the number of valid oil fume sample images is counted.

[0012] In some embodiments, the preset condition includes: the absolute value of the difference is less than or equal to a preset value.

[0013] In some embodiments, determining the fume level calibration value of each of the fume sample images includes: obtaining the fume level calibration value corresponding to each of the fume sample images based on a two-dimensional mapping relationship between pre-calibrated different fume sample images and their corresponding fume level calibration values.

[0014] In some embodiments, determining the fume level detection value of each of the fume sample images includes: inputting each of the fume sample images into a pre-trained fume level detection model to obtain the corresponding fume level detection value, wherein the fume level detection model is trained based on multiple sets of fume sample images and their corresponding fume level detection values.

[0015] In some embodiments, determining the effective recognition rate based on the number of effective oil fume sample images includes: taking the quotient of the number of effective oil fume sample images and the number of n oil fume sample images as the effective recognition rate.

[0016] In some embodiments, determining whether the AI ​​vision module's identification of cooking fumes is qualified based on the false recognition rate and the effective recognition rate includes: when the effective recognition rate is greater than or equal to a first preset threshold and the false recognition rate is less than or equal to a second preset threshold, determining that the AI ​​vision module's identification of cooking fumes is qualified; otherwise, determining that the AI ​​vision module's identification of cooking fumes is unqualified.

[0017] In some embodiments, determining the number of negative oil fume sample images among the n oil fume sample images includes: taking images that do not contain the stove among the n oil fume sample images as negative oil fume sample images, and counting the number of negative oil fume sample images.

[0018] To achieve the above objectives, a second aspect of the present invention provides an accuracy detection device for identifying cooking fumes using an AI vision module, for use in a range hood. The device includes: an acquisition module for acquiring n cooking fume sample images collected by the AI ​​vision module during cooking on a stove; a first determination module for determining the number of negative cooking fume sample images and the number of misidentified negative cooking fume sample images among the n images, and determining the number of valid cooking fume sample images among the n images; a second determination module for determining the misidentification rate of the negative cooking fume sample images based on the number of negative images and the number of misidentified negative images, and determining the effective recognition rate based on the number of valid cooking fume sample images; and a judgment module for judging whether the AI ​​vision module's identification of cooking fumes is qualified based on the misidentification rate and the effective recognition rate.

[0019] The AI ​​vision module's accuracy detection device for identifying cooking fumes according to embodiments of the present invention is used to implement the AI ​​vision module's accuracy detection method for identifying cooking fumes according to the above embodiments of the present invention. First, it acquires n cooking fume sample images collected by the AI ​​vision module during cooking on a stove. It then determines the number of negative cooking fume sample images, the number of misidentified negative cooking fume sample images, and the number of valid cooking fume sample images among the n images. Next, it calculates the misidentification rate of the corresponding negative cooking fume sample images based on the number of negative cooking fume sample images and the number of misidentified negative cooking fume sample images. Finally, it calculates the effective recognition rate based on the number of valid cooking fume sample images. Finally, it determines whether the AI ​​vision module's identification of cooking fumes is qualified based on the misidentification rate and the effective recognition rate. This method detects the accuracy of multiple cooking fume samples collected. The method processes and judges smoke sample images to obtain the number of invalid and valid samples. It then calculates the false recognition rate and valid recognition rate, and uses these calculated values ​​to determine the accuracy of the AI ​​vision module in recognizing cooking fumes. Compared to traditional manual judgment methods, this AI visual recognition method quantifies and calibrates the amount of cooking fumes in a cooking scenario, reducing errors from human perception and significantly decreasing manual labor, thus improving detection efficiency. Furthermore, it identifies and summarizes multiple smoke sample images, calculating the recognition ratio. Unlike instantaneous judgment based on a single image, this method avoids accidental interference from lighting and smoke morphology in a single frame, reducing random errors in single-recognition and effectively enhancing the reliability and accuracy of recognition. Simultaneously, through automated data comparison and accuracy statistics mechanisms, it avoids discrepancies caused by subjective human judgment, reduces manual operation costs, overcomes the pain points of low testing efficiency and difficulty in data reproducibility, and comprehensively improves detection accuracy and testing efficiency.

[0020] To achieve the above objectives, a third aspect of the present invention provides a range hood, comprising: a range hood body; an AI vision module disposed on the range hood body for acquiring images of oil fume samples generated by a cooktop; and an accuracy detection device for identifying oil fumes using the AI ​​vision module as described in the above embodiments of the present invention; or, a controller for executing an accuracy detection method for identifying oil fumes using the AI ​​vision module as described in the above embodiments of the present invention.

[0021] A range hood according to an embodiment of the present invention includes a range hood body; an AI vision module disposed on the range hood body for acquiring images of oil fume samples generated by the stove during cooking; and an accuracy detection device for identifying oil fumes using the AI ​​vision module as described in the above embodiment of the present invention; or, a controller, the controller being configured to execute an accuracy detection method for identifying oil fumes using the AI ​​vision module as described in the above embodiment of the present invention, which first acquires n oil fume sample images acquired by the AI ​​vision module during cooking on the stove, determines the number of negative oil fume sample images, the number of misidentified negative oil fume sample images, and the number of valid oil fume sample images among the n oil fume sample images, then calculates the misidentification rate of the corresponding negative oil fume sample images based on the number of negative oil fume sample images and the number of misidentified negative oil fume sample images, calculates the effective recognition rate based on the number of valid oil fume sample images, and finally... The false recognition rate and effective recognition rate determine whether the AI ​​vision module's identification of cooking fumes is qualified. This is achieved by processing multiple collected cooking fume sample images to determine the number of invalid and valid samples, further calculating the false recognition rate and effective recognition rate, and using these calculated values ​​to determine the accuracy of the AI ​​vision module's cooking fume identification. Compared to traditional manual judgment methods, this AI vision-based cooking fume accuracy detection method quantifies and calibrates the amount of cooking fumes in a cooking scenario, reducing errors from human perception and significantly reducing manual labor, thus improving detection efficiency. Furthermore, by identifying and summarizing multiple cooking fume sample images and calculating the recognition ratio, unlike instantaneous judgment from a single image, it avoids accidental interference from lighting and fumes in a single frame, reducing random errors in single-recognition and effectively enhancing the reliability and accuracy of identification. Simultaneously, through automated data comparison and accuracy statistics mechanisms, it avoids the results differences caused by subjective human evaluation, reduces manual operation costs, overcomes the pain points of low testing efficiency and difficulty in data reproduction, and comprehensively improves detection accuracy and testing efficiency.

[0022] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of a method for detecting the accuracy of an AI vision module in recognizing cooking fumes according to an embodiment of the present invention; Figure 2 This is a flowchart of an AI vision module accuracy detection method for recognizing cooking fumes according to a specific embodiment of the present invention; Figure 3 This is a flowchart of a detection accuracy algorithm according to a specific embodiment of the present invention; Figure 4This is a schematic diagram of the structure of a fan speed control device for a smoke hood according to an embodiment of the present invention; Figure 5 A schematic diagram of a range hood according to an embodiment of the present invention; Figure 6 A schematic diagram of the structure of a range hood according to another embodiment of the present invention.

[0024] Figure label: 1000 - AI vision module accuracy detection device for identifying cooking fumes; 100 - Acquisition module; 200 - First determination module; 300 - Second determination module; 400 - Judgment module; 2000 - Range hood; 2100 - Range hood body; 2200 - AI vision module; 2300 - Controller. Detailed Implementation

[0025] The embodiments of the present invention are described in detail below. The embodiments described with reference to the accompanying drawings are exemplary. The embodiments of the present invention are described in detail below.

[0026] The following is for reference. Figures 1-6 This invention describes a method, apparatus, and range hood for detecting the accuracy of AI vision module recognition of cooking fumes according to embodiments of the present invention.

[0027] In one embodiment of the present invention, an accuracy detection method for identifying cooking fumes using an AI vision module is provided for use in range hoods.

[0028] Figure 1 This is a flowchart of a method for detecting the accuracy of an AI vision module in recognizing cooking fumes according to an embodiment of the present invention, as shown below. Figure 1 As shown, the accuracy detection method for the AI ​​vision module to identify cooking fumes includes: Step S10: Obtain n oil fume sample images collected by the AI ​​vision module during the cooking process on the stove.

[0029] In a specific embodiment, a fixed image sampling period is set, and the AI ​​vision module is controlled to continuously capture images of the cooking area of ​​the stove through a camera at preset time intervals. Multiple frames of oil fume sample images corresponding to multiple sampling periods are sequentially acquired, and the acquired oil fume sample images are uniformly summarized and stored to provide complete time-series image data for subsequent oil fume state analysis, concentration level calibration values, and real-time detection values ​​of oil fume concentration levels. The AI ​​vision module may include, for example, a visible light high-definition camera.

[0030] Specifically, the AI ​​vision module's method for detecting the accuracy of oil fume identification first sets a uniform image sampling time interval as the sampling period. When the stove is in cooking mode, the AI ​​vision module is activated and continuously captures images of the cooking area through the camera. Each time a sampling period is completed, an oil fume sample image is stored. This process is repeated for n consecutive sampling periods, resulting in n oil fume sample images. This ensures that the captured images can completely reproduce the dynamic changes in the generation and diffusion of oil fumes at different times, providing complete time-series image data for subsequent oil fume status analysis, concentration level calibration, and detection values.

[0031] Step S20: Determine the number of negative oil fume sample images and the number of misidentified negative oil fume sample images in the n oil fume sample images, and determine the number of valid oil fume sample images in the n oil fume sample images.

[0032] Specifically, the accuracy detection method for identifying cooking fumes using this AI vision module involves determining the number of negative cooking fume sample images among n cooking fume sample images collected by the AI ​​vision module. These negative cooking fume sample images can include, for example, images without cooking fumes and images with no reference value for cooking fume identification. Further, the method determines the number of misidentified negative cooking fume sample images. These misidentified negative cooking fume sample images can include, for example, images without cooking fumes incorrectly identified as valid cooking fume sample images containing cooking fumes. By statistically analyzing the number of negative samples and misidentified negative cooking fume sample images among the n cooking fume sample images, the method accurately distinguishes between images without cooking fumes and valid detection images. This helps to eliminate abnormal situations where the algorithm misidentifies images without cooking fumes as containing cooking fumes, avoids invalid images interfering with the overall detection results, and improves recognition efficiency and accuracy.

[0033] Furthermore, by determining the number of negative oil fume sample images among the n oil fume sample images collected by the AI ​​vision module, the number of valid oil fume sample images among the n oil fume sample images is also determined. This AI vision module's method for detecting the accuracy of oil fume recognition accurately counts the total number of valid samples and comprehensively summarizes the statistical data of the three types of samples, providing detailed and reliable data for subsequent accuracy detection of oil fume recognition.

[0034] Specifically, relying solely on a single image of cooking fumes for instantaneous judgment is easily affected by sudden factors such as instantaneous fume morphology, changes in ambient lighting, and shifts in shooting angle. This makes it impossible to objectively reflect the true state of cooking fumes under continuous operating conditions. Therefore, this embodiment of the invention uses the image as a sample of cooking fumes for reference in subsequent screening and processing. This facilitates a comprehensive restoration of the actual changes in cooking fumes, reduces the judgment bias caused by accidental factors in a single frame, accurately screens effective analysis samples, and improves the comprehensiveness and objectivity of the overall evaluation.

[0035] Step S30: Determine the false recognition rate of negative oil fume sample images based on the number of negative oil fume sample images and the number of false recognitions, and determine the effective recognition rate based on the number of effective oil fume sample images.

[0036] In a specific embodiment, the accuracy detection method for the AI ​​vision module in recognizing cooking fumes is based on the determined number of negative cooking fume sample images, denoted as t, and the number of false negative samples, denoted as a. The quotient of these two values ​​is then used to determine the accuracy of the recognition. The false recognition rate (MRR) of negative oil fume sample images can be understood as the ratio of samples that incorrectly identify oil fume in smoke-free scenes to all negative samples. This metric measures the method's ability to distinguish smoke-free scenes. Simultaneously, based on the number of valid oil fume sample images *m*, the quotient is obtained by comparing the number of valid oil fume sample images to the total number of collected samples *n*. This is recorded as the effective recognition rate. The effective recognition rate can objectively evaluate the effectiveness of this oil fume identification and detection, intuitively determine the reliability of the identification results, and thus judge the rationality and applicability of the entire detection method.

[0037] Step S40: Determine whether the AI ​​vision module's recognition of cooking fumes is qualified based on the false recognition rate and the effective recognition rate.

[0038] In a specific embodiment, a qualified threshold for the effective recognition rate and a qualified threshold for the negative sample false recognition rate are preset. The values ​​of the two indicators obtained from the detection are compared with the corresponding thresholds. When both indicator values ​​meet the qualified conditions, the oil fume recognition result is determined to be qualified; if either indicator fails to meet the threshold requirement, the recognition result is determined to be unqualified.

[0039] Specifically, the AI ​​vision module's method for detecting the accuracy of oil fume recognition offers multiple practical advantages through a dual-threshold approach. By simultaneously verifying both the effective recognition rate and the false recognition rate—two core indicators—it eliminates the limitations of a single evaluation dimension, enabling comprehensive detection of oil fume recognition accuracy. The standardized and fixed judgment criteria avoid biases caused by subjective human judgment, ensuring a rigorous and standardized evaluation process and guaranteeing the overall stability and practicality of the oil fume recognition system.

[0040] Specifically, the method for detecting the accuracy of oil fume recognition by the AI ​​vision module involves first acquiring n oil fume sample images collected by the AI ​​vision module during cooking on a stove. The number of negative oil fume sample images, the number of misidentified negative oil fume sample images, and the number of valid oil fume sample images are then determined. Next, the misidentification rate of the corresponding negative oil fume sample images is calculated based on the number of negative oil fume sample images and the number of misidentified negative oil fume sample images. The effective recognition rate is calculated based on the number of valid oil fume sample images. Finally, the accuracy of the oil fume recognition by the AI ​​vision module is determined based on the misidentification rate and the effective recognition rate. This process, by processing and judging multiple collected oil fume sample images, yields multiple results without... The method calculates the false recognition rate and effective recognition rate based on the effective sample count and the calculated values ​​to determine the accuracy of the AI ​​vision module in recognizing cooking fumes. Compared to traditional manual judgment methods, this AI vision-based cooking fume recognition accuracy detection method quantifies and calibrates the amount of cooking fumes in a cooking scenario, reducing errors from human perception and significantly reducing manual labor, thus improving detection efficiency. Furthermore, it identifies and summarizes multiple cooking fume sample images, calculating the recognition ratio. Unlike instantaneous judgment based on a single image, this method avoids accidental interference from lighting and fumes in a single frame, reducing random errors in single-recognition and effectively enhancing the reliability and accuracy of recognition. Simultaneously, through automated data comparison and accuracy statistics mechanisms, it avoids discrepancies caused by subjective human judgment, reduces manual operation costs, overcomes the pain points of low testing efficiency and difficulty in data reproduction, and comprehensively improves detection accuracy and testing efficiency.

[0041] Figure 3 This is a flowchart of a detection accuracy algorithm according to a specific embodiment of the present invention, such as... Figure 3 As shown, in one embodiment of the present invention, determining the number of valid oil fume sample images among n oil fume sample images includes: determining the oil fume level calibration value and oil fume level detection value of each oil fume sample image; calculating the absolute value of the difference between the oil fume level calibration value and the oil fume level detection value of each oil fume sample image; oil fume sample images whose absolute value of the difference satisfies a preset condition are taken as valid oil fume sample images, and the number of valid oil fume sample images is counted.

[0042] In a specific embodiment, the detection program relies on a pre-trained recognition model to perform image-based oil fume level detection. The program retrieves image data collected on-site and inputs it into the model for analysis. This model has previously undergone iterative training using a vast number of oil fume sample images under various cooking conditions, paired with corresponding correct oil fume level values, to fully learn the characteristic patterns corresponding to various oil fume forms and concentrations. Based on this mature model, feature extraction and comparison calculations are performed, and finally, the oil fume level detection value corresponding to each image is automatically output. , , ..., , Similarly, the calibration values ​​for oil fume sample images are also obtained by using a pre-trained model. By inputting the collected oil fume sample images, a calibration value for the oil fume level can be obtained. , , ..., , .

[0043] Furthermore, after obtaining the above values, the absolute value of the difference between the oil fume level calibration value and the oil fume level detection value of each oil fume sample image is calculated. By calculating the absolute value of the difference between the output results of the two models, the deviation of the detection results of the two recognition models can be directly compared, quantifying the degree of difference between the two recognition data. This difference can be used to determine the stability and consistency of the model detection. The smaller the difference, the higher the fit of the model recognition result and the more stable the detection performance. At the same time, it can also quickly identify model operation deviations and recognition anomalies, providing reliable data support for model accuracy verification and algorithm optimization.

[0044] In a specific embodiment, when the absolute value of the above difference meets a preset condition, the preset condition may include a threshold judgment condition. Oil fume sample images that meet the preset condition are considered valid oil fume sample images, and the number m of valid oil fume sample images is counted. This AI vision module's method for detecting the accuracy of oil fume recognition allows for the manual setting of a deviation judgment threshold, and the evaluation scale can be flexibly adjusted according to the required detection accuracy.

[0045] In one embodiment of the present invention, the preset condition includes: the absolute value of the difference is less than or equal to a preset value.

[0046] In a specific embodiment, after calculating the absolute value of the difference, the accuracy detection method for the AI ​​vision module in recognizing cooking fumes determines whether the absolute value of the difference is less than or equal to a preset value. This preset value can be flexibly set according to actual needs. In a specific embodiment, the preset value is, for example, set to 1, that is, when the absolute value of the difference is less than or equal to 1, i.e. At that time, oil fume sample images that are considered accurate within the specified range and meet the preset conditions are considered valid oil fume sample images. In a specific embodiment, setting the preset value reasonably, such as, but not limited to, 1, can tighten the allowable range of data deviation, filter out detection results with smaller deviations, effectively eliminate abnormal data with large errors, and facilitate the judgment of errors caused by calibration errors. This adjustable threshold setting mode adapts to different levels of stringency in detection standards, controls the data screening threshold as needed, further improves the accuracy and reliability of the retained data, and ensures that the final identification results have higher reference value.

[0047] In one embodiment of the present invention, determining the fume level calibration value of each fume sample image includes: obtaining the fume level calibration value corresponding to each fume sample image based on the two-dimensional mapping relationship between different pre-calibrated fume sample images and their corresponding fume level calibration values.

[0048] In a specific embodiment, the AI ​​vision module's method for detecting the accuracy of oil fume recognition involves pre-collecting oil fume sample images under different concentrations and cooking conditions, and simultaneously matching them with corresponding oil fume level calibration values. For example, this dataset serves as the basis for model training, establishing a two-dimensional mapping relationship between image features and oil fume level values. During actual detection, the on-site collected oil fume sample images are directly input into the fully trained recognition model. The model automatically retrieves the preset mapping relationship, analyzes various image feature parameters, and processes them according to predetermined rules, directly outputting the oil fume level calibration result matched to the image. Furthermore, based on the two-dimensional mapping relationship between pre-calibrated images of different oil fume samples and their corresponding oil fume level calibration values, the oil fume level calibration value corresponding to each oil fume sample image is obtained. The level determination is then completed using a trained model, overcoming the limitations of manual comparison and evaluation. The judgment standard remains consistent and standardized, significantly reducing errors caused by human judgment. Level results are quickly obtained from image input, with a fast detection response speed, meeting the needs of continuous batch image detection. It can achieve quantitative grading of oil fume status, intuitively and accurately defining the degree of oil fume pollution, making oil fume monitoring work standardized and intelligent, and effectively improving the accuracy of oil fume level identification and overall work efficiency.

[0049] In one embodiment of the present invention, determining the fume level detection value of each of the fume sample images includes: inputting each fume sample image into a pre-trained fume level detection model to obtain the corresponding fume level detection value, wherein the fume level detection model is trained based on multiple sets of fume sample images and their corresponding fume level detection values.

[0050] Specifically, the training of this oil fume level detection model can be achieved, for example, by collecting a large number of oil fume sample images with different concentrations, resolutions, and shooting environments, and accurately labeling all oil fume sample images with their corresponding oil fume levels, forming a complete training dataset with a one-to-one correspondence between image samples and corresponding oil fume level values. This dataset is then used to fully train the oil fume level detection model, allowing the model to deeply learn the intrinsic mapping relationship between key features of oil fume sample images, such as turbidity, light transmittance, and smoke texture, and the oil fume level. This continuously optimizes the model parameters, ultimately resulting in a stable and reliable oil fume level detection model. In the actual detection phase, the oil fume sample images collected by all AI vision modules are input one by one into the trained detection model. The model automatically extracts multi-dimensional features from the images, combines them with the learned correspondences for intelligent analysis and calculation, and finally accurately outputs the oil fume level detection value corresponding to each oil fume sample image.

[0051] In a specific embodiment, by inputting images of various oil fume samples through a pre-trained oil fume level detection model, the corresponding oil fume level detection value can be obtained. The entire detection method relies on a pre-trained and mature AI model to complete intelligent judgment, eliminating the need for subjective scoring and comparison by humans throughout the process. This completely avoids problems such as inconsistent human evaluation standards, large deviations in visual judgment, and misjudgments caused by differences in experience. The oil fume level detection model can continuously process massive amounts of oil fume sample images in batches. Inputting images automatically outputs quantitative level results, with fast detection speed and strong adaptability, greatly improving the overall work efficiency of oil fume level detection. Furthermore, based on training and fitting with massive amounts of real samples, feature recognition is more comprehensive and level matching is more accurate. It can stably and standardizedly complete the quantitative assessment of oil fume levels, realizing objective, intelligent, and refined detection of oil fume pollution levels, effectively ensuring the stability and reliability of detection results.

[0052] In one embodiment of the present invention, determining the effective recognition rate based on the number of effective oil fume sample images includes: taking the quotient of the number of effective oil fume sample images and the number of n oil fume sample images as the effective recognition rate.

[0053] In a specific embodiment, when the absolute value of the difference between the detected value and the calibration value is less than or equal to a preset value, the number of samples meeting this condition is counted as m, which are considered valid oil fume sample images. Then, the quotient A of the number of valid oil fume sample images m and the total number of oil fume sample images n collected by the AI ​​vision module is calculated as the effective oil fume recognition rate of the AI ​​vision module. Furthermore, the effective recognition rate of this oil fume sample image is calculated by dividing the number of valid oil fume sample images obtained from the statistics by the total number of oil fume sample images participating in the test. Therefore, this quantitative calculation method for identifying indicators, with standardized calculation rules, yields objective and accurate data that directly reflects the actual effectiveness of image recognition work. It also provides a reliable data benchmark for subsequent assessment of the pass / fail status of oil fume recognition results.

[0054] In one embodiment of the present invention, determining whether the AI ​​vision module's recognition of cooking fumes is qualified based on the false recognition rate and the effective recognition rate includes: when the effective recognition rate is greater than or equal to a first preset threshold and the false recognition rate is less than or equal to a second preset threshold, determining that the AI ​​vision module's recognition of cooking fumes is qualified; otherwise, determining that the AI ​​vision module's recognition of cooking fumes is unqualified.

[0055] Specifically, by pre-setting a passing threshold for the effective recognition rate and a passing threshold for the negative sample misidentification rate, the measured values ​​of these two indicators are compared with their corresponding thresholds. When the effective recognition rate A is greater than or equal to the first preset threshold, it indicates that the accuracy of the oil fume sample image recognition meets the standard, and when the misidentification rate MA is less than or equal to the second preset threshold, it indicates that the image misjudgment is within a controllable range, meaning that the model's error probability meets the standard, with low recognition deviation and high judgment reliability. That is, when both conditions are met... ,and ( and When all values ​​are preset, the AI ​​vision module is deemed to have qualified to identify the cooking fumes.

[0056] Furthermore, the AI ​​vision module's ability to identify cooking fumes is determined based on the false recognition rate and the effective recognition rate. If the effective recognition rate A is greater than or equal to the first preset threshold, but the false recognition rate MA does not meet the requirement of being less than or equal to the second preset threshold; or the false recognition rate MA is less than or equal to the second preset threshold, but the effective recognition rate A does not meet the requirement of being greater than or equal to the first preset threshold; or the effective recognition rate A does not meet the requirement of being greater than or equal to the first preset threshold and the false recognition rate MA does not meet the requirement of being less than or equal to the second preset threshold, then in all three cases, the AI ​​vision module's identification of cooking fumes is deemed unqualified. Therefore, only when both the effective recognition rate meets the standard and the false recognition rate does not exceed the limit standard can the overall result of this cooking fume sample image recognition detection be considered qualified. This ensures both the basic ability of normal recognition and strict control over erroneous judgments, achieving comprehensive recognition performance that meets usage specifications.

[0057] The AI ​​vision module's method for detecting the accuracy of oil fume recognition employs a dual-indicator joint verification approach, making the evaluation criteria more rigorous and comprehensive, avoiding the bias caused by a single indicator. This ensures both the overall output quality of the recognition and limits the frequency of errors, significantly improving the reliability of the detection results.

[0058] In one embodiment of the present invention, determining the number of negative oil fume sample images among n oil fume sample images includes: taking images that do not contain the stove among the n oil fume sample images as negative oil fume sample images, and counting the number of negative oil fume sample images.

[0059] Specifically, the AI ​​vision module collects n sample images of cooking fumes and performs feature detection on each image sequentially. For example, algorithms can be used to extract object contours and shapes, which are then compared to the standard contour of a stove for verification. Alternatively, detection algorithms can be used to define bounding boxes within the images to check for the presence of a stove target object. If a stove contour feature cannot be matched, the image does not show the stove's shape, or no corresponding detection box is generated, the image is designated as a negative cooking fume sample. After all images are screened, the total number of negative cooking fume sample images is tallied. By calculating the number of negative samples and combining this with the negative sample determination results, the false recognition rate is calculated to check whether the model misclassifies images without a stove as having a target. This also helps to categorize samples, improve the sample dataset classification system, provide foundational data for overall recognition performance evaluation, and reduce misjudgments caused by irrelevant images. This further enhances the performance evaluation dimensions and ensures the reliability of subsequent recognition results.

[0060] The following describes the accuracy detection method for identifying cooking fumes using the AI ​​vision module of the present invention, in conjunction with specific embodiments. Figure 2 This is a flowchart of a method for detecting the accuracy of an AI vision module in recognizing cooking fumes according to a specific embodiment of the present invention, as shown below. Figure 2 As shown in this specific embodiment, the method for detecting the accuracy of oil fume recognition by the AI ​​vision module includes the following steps: S1: Initialization; S2: Enable camera streaming; S3: Camera streaming is normal; S4: Check if any other process is using the streaming process; S5: Close non-camera processes; S6: Investigate other possible causes; S7: Start the image acquisition program; S8: Collect images of cooking fumes samples according to the acquisition frame rate; S9: Store the collected images of the oil fume samples; S10: Oil fume sample image oil fume level calibration, output calibration value; S11: Detect the oil fume level of the collected oil fume sample image and output the detection value; S12: Classify and statistically analyze the information on the level of cooking fumes, calculate its accuracy, and determine whether the test results are qualified.

[0061] In this specific embodiment, the control flow description of the AI ​​vision module's accuracy detection method for identifying cooking fumes includes first entering S1, then entering S2, if S3 is true, then entering S6, otherwise executing S4, if S4 is true, then executing S5 and then S2, otherwise executing S6 and then S2 again, then entering S7, then S8, then S9 and then S10, then S11, and finally executing S12.

[0062] Figure 3 This is a flowchart of a detection accuracy algorithm according to a specific embodiment of the present invention, such as... Figure 3 As shown in this specific embodiment, step S12, which involves classifying and statistically analyzing the oil fume level information and calculating its accuracy, includes: S121: Initialization; S122: The total number of samples collected is ; S123: The total number of negative samples (non-oil fume samples) collected is ; S124: The oil fume detection level values ​​of the collected images are as follows: , , ..., , ; S125: The oil fume rating values ​​of the collected images are as follows: , , ..., , ; S126: The number of misidentifications of the collected negative samples is

[0063] S127: Calculate the difference between the detected value and the calibration value. ; S128: The number of samples is ; S129: Effective recognition rate ; S1210: The false recognition rate of negative samples is... ; S1211: Judgment ,and ( and (The value is set); if it meets the requirements, proceed to step S1212; if it does not meet the requirements, proceed to step S1213. S1212: Pass; S1213: Unqualified.

[0064] In this specific embodiment, the AI ​​vision module's method for detecting the accuracy of oil fume identification has the following advantages compared to existing technologies: (1) The AI ​​vision module's method for detecting the accuracy of oil fume recognition calculates the recognition ratio by recognizing and summarizing multiple oil fume sample images. This method differs from instantaneous judgment based on a single image, avoiding accidental interference from light and shadow and oil fume morphology in a single frame, reducing the accidental error of a single recognition, and effectively enhancing the reliability and accuracy of recognition judgment. (2) The AI ​​vision module's method for detecting the accuracy of oil fume identification makes a judgment on whether the accuracy conditions meet the preset threshold range (accuracy is considered within the range), which is beneficial for judging errors caused by calibration errors. (3) The AI ​​vision module’s method for detecting the accuracy of oil fume recognition can automatically detect the accuracy of oil fume recognition. By relying on automated data comparison and accuracy statistics mechanism, it avoids the difference in results caused by subjective human judgment and reduces the cost of manual operation. (4) The AI ​​vision module's method for detecting the accuracy of oil fume identification quantifies and calibrates the amount of oil fume in the cooking scene, reducing human sensory errors and reducing a lot of manual labor, thus improving detection efficiency. (5) The accuracy detection method of this AI vision module for identifying cooking fumes relies on the simultaneous verification of two core indicators: effective recognition rate and false recognition rate. The judgment criteria are unified and fixed, avoiding the bias caused by subjective human judgment, and making the evaluation process standardized and rigorous. This effectively ensures the overall operational stability and practicality of the cooking fume recognition system. (6) The AI ​​vision module’s method for detecting the accuracy of oil fume recognition adopts quantitative calculation to statistically analyze the recognition indicators. The calculation rules are uniform and standardized, and the resulting data is objective and true, which can intuitively reflect the actual effectiveness of image recognition work.

[0065] In summary, the accuracy detection method for identifying cooking fumes using an AI vision module according to embodiments of the present invention first acquires n cooking fume sample images collected by the AI ​​vision module during cooking on a stove. It then determines the number of negative cooking fume sample images, the number of misidentified negative cooking fume sample images, and the number of valid cooking fume sample images among the n images. Next, it calculates the misidentification rate of the corresponding negative cooking fume sample images based on the number of negative cooking fume sample images and the number of misidentified negative images. Finally, it calculates the effective recognition rate based on the number of valid cooking fume sample images. Finally, it determines whether the AI ​​vision module's identification of cooking fumes is qualified based on the misidentification rate and the effective recognition rate. This method processes and judges multiple collected cooking fume sample images in this way. By obtaining multiple invalid and valid sample counts, the false recognition rate and valid recognition rate are further calculated, and the accuracy of the AI ​​vision module in recognizing cooking fumes is judged based on the calculated values. Compared with traditional manual judgment methods, this AI vision recognition method for cooking fumes quantifies and calibrates the amount of cooking fumes in the cooking scene, reducing errors from human perception and significantly reducing manual labor, thus improving detection efficiency. Furthermore, by recognizing and summarizing multiple cooking fume sample images, the recognition ratio is calculated. Unlike instantaneous judgment based on a single image, this method avoids accidental interference from lighting and fumes in a single frame, reducing random errors in single recognition and effectively enhancing the reliability and accuracy of recognition. Simultaneously, through automated data comparison and accuracy statistics mechanisms, it avoids the result differences caused by subjective human judgment, reduces manual operation costs, overcomes the pain points of low testing efficiency and difficulty in data reproduction, and comprehensively improves detection accuracy and testing efficiency.

[0066] In one embodiment of the present invention, an AI vision module accuracy detection device 1000 for identifying cooking fumes is also provided for use in a range hood 2000.

[0067] Figure 4 This is a schematic diagram of the structure of an AI vision module for detecting the accuracy of oil fume recognition according to an embodiment of the present invention. Figure 4 As shown, in one embodiment of the present invention, the AI ​​vision module's accuracy detection device 1000 for identifying cooking fumes includes: an acquisition module 100, used to acquire n cooking fume sample images collected by the AI ​​vision module during the cooking process on a stove; a first determination module 200, used to determine the number of negative cooking fume sample images and the number of misidentified negative cooking fume sample images among the n cooking fume sample images, and to determine the number of valid cooking fume sample images among the n cooking fume sample images; a second determination module 300, used to determine the misidentification rate of negative cooking fume sample images based on the number of negative cooking fume sample images and the number of misidentified negative cooking fume sample images, and to determine the effective recognition rate based on the number of valid cooking fume sample images; and a judgment module 400, used to judge whether the AI ​​vision module's identification of cooking fumes is qualified based on the misidentification rate and the effective recognition rate.

[0068] In one embodiment of the present invention, the first determining module 200 determines the number of valid oil fume sample images among n oil fume sample images, including: determining the oil fume level calibration value and oil fume level detection value of each oil fume sample image; calculating the absolute value of the difference between the oil fume level calibration value and the oil fume level detection value of each oil fume sample image; oil fume sample images whose absolute value of the difference meets the preset conditions are regarded as valid oil fume sample images, and the number of valid oil fume sample images is counted.

[0069] In one embodiment of the present invention, the first determining module 200 determines the number of negative oil fume sample images among n oil fume sample images, including: taking images that do not contain stoves among the n oil fume sample images as negative oil fume sample images, and counting the number of negative oil fume sample images.

[0070] In one embodiment of the present invention, the preset condition includes: the absolute value of the difference is less than or equal to a preset value.

[0071] In one embodiment of the present invention, the first determining module 200 determines the oil fume level calibration value of each oil fume sample image, including: obtaining the oil fume level calibration value corresponding to each oil fume sample image based on the two-dimensional mapping relationship between the pre-calibrated different oil fume sample images and their corresponding oil fume level calibration values.

[0072] In one embodiment of the present invention, the first determining module 200 determines the oil fume level detection value of each oil fume sample image, including: inputting each oil fume sample image into a pre-trained oil fume level detection model to obtain the corresponding oil fume level detection value, wherein the oil fume level detection model is trained based on multiple sets of oil fume sample images and their corresponding oil fume level detection values.

[0073] In one embodiment of the present invention, the second determining module 300 determines the effective recognition rate based on the number of effective oil fume sample images, including: taking the quotient of the number of effective oil fume sample images and the number of n oil fume sample images as the effective recognition rate.

[0074] In one embodiment of the present invention, the judgment module 400 determines whether the AI ​​vision module's recognition of cooking fumes is qualified based on the false recognition rate and the effective recognition rate, including: when the effective recognition rate is greater than or equal to a first preset threshold and the false recognition rate is less than or equal to a second preset threshold, the AI ​​vision module's recognition of cooking fumes is determined to be qualified; otherwise, the AI ​​vision module's recognition of cooking fumes is determined to be unqualified.

[0075] It should be noted that the specific implementation of the AI ​​vision module for detecting the accuracy of oil fume recognition in this embodiment of the invention is similar to the specific implementation described in the AI ​​vision module for detecting the accuracy of oil fume recognition in the above embodiment of the invention. For details, please refer to the description of the AI ​​vision module for detecting the accuracy of oil fume recognition in the section on the AI ​​vision module for detecting the accuracy of oil fume recognition. To reduce redundancy, it will not be repeated here.

[0076] Specifically, the AI ​​vision module oil fume recognition accuracy detection device 1000 according to an embodiment of the present invention is used to implement the AI ​​vision module oil fume recognition accuracy detection method of the above embodiment of the present invention. First, it acquires n oil fume sample images collected by the AI ​​vision module during cooking on a stove. It determines the number of negative oil fume sample images, the number of misidentified negative oil fume sample images, and the number of valid oil fume sample images among the n oil fume sample images. Then, it calculates the misidentification rate of the corresponding negative oil fume sample images based on the number of negative oil fume sample images and the number of misidentified negative oil fume sample images. It calculates the effective recognition rate based on the number of valid oil fume sample images. Finally, it judges whether the AI ​​vision module's oil fume recognition is qualified based on the misidentification rate and the effective recognition rate. This method detects the oil fume collected by the AI ​​vision module during cooking on a stove. This method processes and judges multiple oil fume sample images to obtain the number of invalid and valid samples. It then calculates the false recognition rate and effective recognition rate, and based on these calculated values, determines the accuracy of the AI ​​vision module in recognizing oil fumes. Compared to traditional manual judgment methods, this AI visual oil fume recognition accuracy detection method quantifies and calibrates the amount of oil fumes in cooking scenarios, reducing errors from human perception and significantly decreasing manual labor, thus improving detection efficiency. Furthermore, by recognizing and summarizing multiple oil fume sample images and calculating the recognition ratio, unlike instantaneous judgment from a single image, it avoids accidental interference from lighting and oil fume morphology in a single frame, reducing random errors in single recognition and effectively enhancing the reliability and accuracy of recognition. Simultaneously, through automated data comparison and accuracy statistics mechanisms, it avoids the results differences caused by subjective human evaluation, reduces manual operation costs, overcomes the pain points of low testing efficiency and difficulty in data reproduction, and comprehensively improves detection accuracy and testing efficiency.

[0077] In one embodiment of the present invention, a range hood 2000 is also provided.

[0078] Figure 5 This is a schematic diagram of the structure of a range hood according to an embodiment of the present invention. Figure 5 As shown, in one embodiment of the present invention, the range hood 2000 includes: a range hood body 2100; an AI vision module 2200 disposed on the range hood body 2100 for collecting images of oil fume samples generated during the cooking process of the stove; and an accuracy detection device 1000 for identifying oil fumes using the AI ​​vision module as described in the above embodiment of the present invention.

[0079] Figure 6This is a schematic diagram of a range hood according to another embodiment of the present invention. Figure 6 As shown, in another embodiment of the present invention, the range hood 2000 includes: a range hood body 2100; an AI vision module 2200 disposed on the range hood body 2100 for acquiring images of oil fume samples generated during the cooking process of the stove; and a controller 2300 for executing the accuracy detection method for identifying oil fumes using the AI ​​vision module as described in the above embodiment of the present invention.

[0080] It should be noted that the specific implementation method of the range hood 2000 in this embodiment of the invention for detecting the accuracy of oil fume recognition by the AI ​​vision module is similar to the specific implementation method described in the above embodiment of the invention for detecting the accuracy of oil fume recognition by the AI ​​vision module. For details, please refer to the description of the method for detecting the accuracy of oil fume recognition by the AI ​​vision module. To reduce redundancy, it will not be repeated here.

[0081] Furthermore, other components and functions of the range hood 2000 according to the above embodiments of the present invention are known to those skilled in the art, and will not be described in detail in order to reduce redundancy.

[0082] Specifically, the range hood 2000 according to an embodiment of the present invention includes: a range hood body 2100; an AI vision module 2200 disposed on the range hood body 2100 for acquiring images of oil fume samples generated during cooking on a stove; and an AI vision module oil fume recognition accuracy detection device 1000 as described in the above embodiment of the present invention; or, a controller 2300, the controller 2300 being used to execute the AI ​​vision module oil fume recognition accuracy detection method as described in the above embodiment of the present invention, first acquiring n oil fume sample images acquired by the AI ​​vision module during cooking on a stove, determining the number of negative oil fume sample images, the number of misidentified negative oil fume sample images, and the number of valid oil fume sample images among the n oil fume sample images, then calculating the misidentification rate of the corresponding negative oil fume sample images based on the number of negative oil fume sample images and the number of misidentified negative oil fume sample images, and calculating the misidentification rate of the valid oil fume sample images. The method calculates the effective recognition rate and then determines whether the AI ​​vision module's recognition of cooking fumes is qualified based on the false recognition rate and the effective recognition rate. This is achieved by processing multiple collected cooking fume sample images to obtain the number of invalid and valid samples. The false recognition rate and effective recognition rate are then calculated, and the accuracy of the AI ​​vision module's cooking fume recognition is determined based on these calculated values. Compared to traditional manual judgment methods, this AI vision-based cooking fume recognition accuracy detection method quantifies and calibrates the amount of cooking fumes in a cooking scenario, reducing errors from human perception and significantly reducing manual labor, thus improving detection efficiency. Furthermore, by recognizing and summarizing multiple cooking fume sample images and calculating the recognition ratio, unlike instantaneous judgment based on a single image, it avoids accidental interference from lighting and fumes in a single frame, reducing the random error of a single recognition and effectively enhancing the reliability and accuracy of the recognition judgment. Simultaneously, through automated data comparison and accuracy statistics mechanisms, it avoids the result differences caused by subjective human evaluation, reduces manual operation costs, overcomes the pain points of low testing efficiency and difficulty in data reproduction, and comprehensively improves detection accuracy and testing efficiency.

[0083] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example.

[0084] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for detecting the accuracy of an AI vision module in recognizing cooking fumes, characterized in that, For use in a range hood, the method includes: Acquire n oil fume sample images collected by the AI ​​vision module during the cooking process on the stove; Determine the number of negative oil fume sample images and the number of misidentified negative oil fume sample images among the n oil fume sample images; and determine the number of valid oil fume sample images among the n oil fume sample images. The false recognition rate of the negative oil fume sample images is determined based on the number of negative oil fume sample images and the number of false recognitions, and the effective recognition rate is determined based on the number of effective oil fume sample images; The AI ​​vision module determines whether the identification of cooking fumes is qualified based on the false recognition rate and the effective recognition rate.

2. The method for detecting the accuracy of oil fume recognition using an AI vision module according to claim 1, characterized in that, Determining the number of valid oil fume sample images among the n oil fume sample images includes: Determine the fume level calibration value and fume level detection value for each of the aforementioned fume sample images; Calculate the absolute value of the difference between the oil fume level calibration value and the oil fume level detection value of each of the oil fume sample images; The oil fume sample images whose absolute values ​​of the differences satisfy the preset conditions are taken as the valid oil fume sample images, and the number of valid oil fume sample images is counted.

3. The method for detecting the accuracy of oil fume recognition by the AI ​​vision module according to claim 2, characterized in that, The preset condition includes: the absolute value of the difference is less than or equal to a preset value.

4. The method for detecting the accuracy of oil fume recognition by the AI ​​vision module according to claim 2, characterized in that, Determining the oil fume level calibration value for each of the oil fume sample images includes: Based on the two-dimensional mapping relationship between different pre-calibrated oil fume sample images and their corresponding oil fume level calibration values, the oil fume level calibration value corresponding to each oil fume sample image is obtained.

5. The method for detecting the accuracy of oil fume recognition by the AI ​​vision module according to claim 2, characterized in that, Determining the oil fume level detection value for each of the oil fume sample images includes: Each of the oil fume sample images is input into a pre-trained oil fume level detection model to obtain the corresponding oil fume level detection value. The oil fume level detection model is trained based on multiple sets of oil fume sample images and their corresponding oil fume level detection values.

6. The method for detecting the accuracy of oil fume recognition by the AI ​​vision module according to claim 1, characterized in that, The determination of the effective recognition rate based on the number of effective oil fume sample images includes: The quotient of the number of valid oil fume sample images and the number of n oil fume sample images is taken as the effective recognition rate.

7. The method for detecting the accuracy of oil fume recognition using an AI vision module according to claim 1, characterized in that, The step of determining whether the AI ​​vision module's identification of cooking fumes is qualified based on the false recognition rate and the effective recognition rate includes: When the effective recognition rate is greater than or equal to the first preset threshold and the false recognition rate is less than or equal to the second preset threshold, the AI ​​vision module is determined to be qualified in recognizing cooking fumes; otherwise, the AI ​​vision module is determined to be unqualified in recognizing cooking fumes.

8. The method for detecting the accuracy of oil fume recognition by the AI ​​vision module according to claim 1, characterized in that, Determining the number of negative oil fume sample images among the n oil fume sample images includes: The image that does not contain the stove among the n oil fume sample images is taken as the negative oil fume sample image, and the number of negative oil fume sample images is counted.

9. A device for detecting the accuracy of an AI vision module in recognizing cooking fumes, characterized in that, For use in a range hood, the device includes: The acquisition module is used to acquire n oil fume sample images collected by the AI ​​vision module during the cooking process on the stove. The first determining module determines the number of negative oil fume sample images and the number of misidentified negative oil fume sample images among the n oil fume sample images, and determines the number of valid oil fume sample images among the n oil fume sample images; The second determining module determines the false recognition rate of the negative oil fume sample images based on the number of negative oil fume sample images and the number of false recognitions, and determines the effective recognition rate based on the number of effective oil fume sample images; The judgment module is used to determine whether the AI ​​vision module's recognition of cooking fumes is qualified based on the false recognition rate and the effective recognition rate.

10. A range hood, characterized in that, include: Range hood body; An AI vision module is installed on the range hood body to collect images of oil fumes generated by the stove during cooking. The accuracy detection device for identifying cooking fumes using an AI vision module as described in claim 9; or, A controller, the controller being used to execute the accuracy detection method for identifying cooking fumes by an AI vision module as described in any one of claims 1-8.

Images