Method for evaluating future generation images based on environmental multi-element outdoor scenes

Abstract

This invention discloses an evaluation method for future generated images of outdoor scenes based on multiple environmental factors. To explore the relationships between various environmental factors in an outdoor scene and evaluate the effect of generated images over a future period, firstly, the actual and predicted values ​​of meteorological data, as well as the predicted value of solar position information, are obtained; then, the prediction probability of the effect of illumination on the generated image is calculated using the total probability formula; finally, the aerosol transfer function (MTF) is used... a Prediction probability and turbulence transfer function (MTF) e The prediction probability calculation calculates the predicted probability of atmospheric effects on the generated image; finally, the overall predicted probability of both factors on the generated image is used to evaluate the effect of the generated image in the outdoor scene over a long period of time. This invention adds an evaluation index for the effect of image generation over a long period of time. Based on the causal relationship of dynamic factors affecting the image, it uses the predicted and actual values ​​of environmental element data in the outdoor scene to calculate the predicted probability of the generated image. The parameter information is simple to obtain, applicable to different outdoor scenes, and highly portable.

Description

Technical Field

[0001] This invention relates to a method for evaluating generated images, and more particularly to a method for evaluating the accuracy of generated images of outdoor scenes over a future period of time based on environmental factors. Background Technology

[0002] Image generation technology is a multidisciplinary information technology that generates images under different environmental conditions based on the relationship and development patterns between images and time, as well as external influencing factors. Currently, outdoor scene image generation has broad application prospects in fields such as scene matching, virtual reality, autonomous driving, and aircraft simulation. In image generation technology, to evaluate the quality of generated images, it is necessary to introduce evaluation methods to ensure the accuracy of the generated images.

[0003] Currently, domestic and international patents have proposed different methods for evaluating the image generation effect, each with its own advantages and disadvantages.

[0004] The patent document with patent number US9338348B2 and publication date of January 12, 2021 proposes a method for evaluating image quality in real time. It extracts regions of interest from images captured by multiple sensors, obtains the pixel percentage of each image, and evaluates the image quality by calculating the confidence probability through sensor parameters. However, this method can only evaluate the quality of real-time images and is not suitable for evaluating the accuracy of images generated in the future.

[0005] Patent document US20190012581A1, published on January 10, 2019, proposes a machine learning-based method for evaluating generated images. It establishes an input sample set including real and generated images, trains a neural network to determine the features of the input samples, and compares the corresponding statistical information of the generated and real images. Distance vector correlation metrics IS (Inception Score), FID (Fréchet Inception Distance), and SOS (Supervised Omni-Score) are used to evaluate the accuracy of image generation. However, these metrics only judge based on image features, and the spatial relationships of these features are not one-to-one. Furthermore, the evaluation of generated images for different outdoor scenes requires retraining, resulting in poor portability.

[0006] A patent document with publication number CN109598299A and publication date of April 9, 2009, discloses an image similarity determination method, apparatus, and electronic device. This method calculates the peak signal-to-noise ratio (PSNR) based on the mean square error between the generated image and a standard image. If the PSNR is less than a threshold, the similarity of the image to be tested is determined based on at least one of the standard image's brightness similarity, contrast similarity, and structural similarity, thus achieving an evaluation of the quality of the generated image. However, this similarity evaluation index cannot accurately evaluate any single factor among image noise, brightness, contrast, and structure. The accuracy of the evaluation needs improvement, and it cannot accurately evaluate future generated images of outdoor scenes.

[0007] A patent document with publication number CN101853504B and publication date of April 25, 2012, proposes an image quality evaluation method based on visual characteristics and structural similarity. This method reads pixel information from a reference image and the generated image to be tested, divides the image into blocks, and comprehensively considers the different levels of human visual attention and sensitivity to different regions of the image. It then applies a weighted average to the structural similarity index of each block, obtaining a comprehensive image quality evaluation index based on visual characteristics and structural similarity, thus completing the quality evaluation method based on visual characteristics and structural similarity. However, this method does not consider the impact of dynamic changes in outdoor scenes such as lighting, temperature, humidity, and particulate matter concentration on the image, and therefore cannot provide a comprehensive and objective evaluation of the accuracy of images generated in outdoor scenes.

[0008] A patent document with publication number CN114972812A and publication date of August 30, 2022, proposes a nonlocal attention learning method based on structural similarity. This method utilizes structural similarity to measure the similarity between signals at different spatial locations, fully considering human visual perception of signal brightness, contrast, and structure. It also applies the Multi-scale Structural Similarity (MSSIM) metric to evaluate the accuracy of generated images at different scales. While this design achieves the evaluation of generated image features, it does not consider the probability of changes in the outdoor environment over time affecting the generated image, making it unsuitable for evaluating the accuracy of generated images under different future environmental conditions in outdoor scenes.

[0009] Lee C et al. proposed a full-reference image evaluation index based on information theory, Visual Information Fidelity (VIF), in their paper "Objective video quality assessment" (Lee C, Cho S, Choe J, et al. Objective video quality assessment[J]. Optical engineering, 2006, 45(1):017004-017004-11.). This method evaluates the accuracy of generated images by linking visual quality with the mutual information between the test image and the reference image based on the concepts of natural scene statistics and image signal extraction by the human visual system. However, it can only evaluate grayscale images and cannot reflect the structural information of the image, so it is difficult to apply to the accuracy evaluation of generated images in actual outdoor scenes.

[0010] In summary, existing research on image generation quality evaluation methods fails to consider the impact of the probability of changes in outdoor environmental factors over time on the image, making it impossible to accurately evaluate the quality of generated images over a future period and thus failing to meet the needs of evaluating the accuracy of future generated images in outdoor scenes. This invention innovatively proposes a predictive probability evaluation index suitable for evaluating images generated in outdoor scenes over a future period. By combining the relationships between various environmental factors in outdoor scenes and applying statistical methods, a mathematical model for calculating the predicted probability of generated images is established, enabling the evaluation of the quality of generated images in outdoor scenes over a future period. This approach has advantages such as simple data acquisition, strong portability, and wide applicability. Summary of the Invention

[0011] To explore the relationships between various environmental elements in outdoor scenes and evaluate the quality of images generated over a future period, this invention provides an evaluation method for future generated images of outdoor scenes based on multiple environmental factors. The method calculates the confidence probability between generated images and real images over a future period based on the temporal relationship between illumination and image generation, as well as the causal relationship between atmospheric effects and image generation. This quantitatively evaluates the predicted probability of generated images, thereby achieving an accurate evaluation of future generated images of outdoor scenes.

[0012] To address the aforementioned technical problems, this invention proposes an evaluation method for future generated images of outdoor scenes based on multiple environmental factors. First, it acquires the actual and predicted values ​​of meteorological data, as well as the predicted value of solar position information. Then, it calculates the prediction probability of illumination effects on the generated image using a full probability calculation method. Next, it calculates the prediction probability of atmospheric effects on the generated image using the aerosol transfer function and turbulence transfer function prediction probabilities. Finally, it evaluates the effectiveness of the future generated image of the outdoor scene based on the total prediction probability of illumination and atmospheric effects on the generated image. The specific steps are as follows:

[0013] Step 1: Obtain the actual values ​​of meteorological element data from the established meteorological station, and obtain the predicted values ​​of meteorological element data from the weather forecast issued by the local meteorological bureau. Use the current solar position information as the predicted value of the solar position information at the same time in the next 3 days. The solar position information includes the solar angle of incidence, the solar altitude angle, and the solar azimuth angle. The meteorological element data includes temperature, humidity, wind speed, PM2.5, and PM10.

[0014] Step 2, the predicted probability P(M) of the generated image due to illumination effect, includes:

[0015] 2-1) Determine the date and time of the prediction, the longitude and latitude of the target scene location, and thus obtain the geographical latitude φ, solar declination δ, and solar hour angle t of the target scene location at the prediction time;

[0016] 2-2) Using equations (1), (2) and (3), the solar altitude angle H at the predicted time of the target scene location is obtained. S The azimuth of the sun, A S The angle of incidence θ of the sun i :

[0017] sinH S =sinφ×sinδ+cosφ×cosδ×cost (1)

[0018] cosA S =(sinH S ×sinφ-sinδ)÷(cosH S ×cosφ) (2)

[0019] In equations (1) and (2), φ represents the geographical latitude, δ represents the solar declination, and t represents the solar hour angle.

[0020] cosθ i =cosH S cosA S (3)

[0021] 2-3) The solar altitude angle H at the predicted time of the target scene location.S The azimuth of the sun, A S The angle of incidence θ of the sun i As the true value;

[0022] 2-4) Calculate the prediction probability P(M) of the generated image by the illumination effect using equation (4).

[0023] P(M)=P(M|A1)P(A1)+P(M|A2)P(A2) (4)

[0024] In equation (4), P(A1) and P(A2) are the solar altitude angles H with a precision of seconds and centered at the stated time point within 5 minutes. S The azimuth of the sun, A S The probabilities that the actual value and the predicted value are the same: P(M|A1) and P(M|A2) are the angles of incidence of the sun within 5 minutes with a precision of seconds and the stated time point as the midpoint. i When the actual value and the predicted value are the same, the solar altitude angle H S The azimuth of the sun, A S The probability that the actual value and the predicted value are the same;

[0025] Step 3, the predicted probability P(N) of atmospheric effects on the generated image, includes:

[0026] 3-1) Calculate the aerosol transfer function (MTF) a The predicted probability P(B)

[0027]

[0028] Based on the actual and predicted values ​​of meteorological elements over the past N years (N = 1-10 years), the visibility R for each day is obtained using the visibility calculation formula for the target location. m The actual and predicted values; taking three consecutive days as a group, calculate the visibility R for each group of three days. m The difference R between the actual value and the predicted value m_i , i = 1, 2, 3; for all differences R m_i Statistical fitting was performed to obtain the probability density function f(b) of the visibility difference over the next 3 days. i The threshold σ1 is determined based on the visibility of the target scene location;

[0029] 3-2) Calculate the turbulent transfer function MTF e The predicted probability P(C),

[0030]

[0031] Based on the actual and predicted values ​​of meteorological elements from the past N years, the refractive index structure constant for each day is obtained using the formula for calculating the refractive index structure constant of the target scene location. The actual and predicted values; taking three consecutive days as a group, calculate the refractive index structure constant for each group of three days. The difference between the actual value and the predicted value For all differences Statistical fitting was performed to obtain the probability density function f(c) of the difference in refractive index structure constants over the next 3 days. i The threshold σ2 is determined based on the refractive index structure constant of the target scene location.

[0032] 3-3) Calculate the prediction probability P(N) of atmospheric effects on the generated image.

[0033] P(N)=P(B)×P(C) (7)

[0034] Step 4: Consider the combined effects of illumination and atmosphere on the total prediction probability P of the generated image. total

[0035] P total =P(M)×P(N) (8)

[0036] Through the total predicted probability P total It enables the evaluation of the generated image effect in outdoor scenes.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] 1) Existing image generation evaluation methods often do not consider the relationship between the generated images of outdoor scenes and the changes over time. They generally focus on evaluating the images at a certain time point. This invention adds an evaluation index for the image generation effect over a long period of time. It evaluates the accuracy of the generated images by calculating the prediction probability of the generated images, thus making up for the shortcomings of existing evaluation methods.

[0039] 2) Existing methods for image quality assessment are mostly limited to extracting and comparing one or more features such as edges, contrast, brightness, and structure of generated and real images to determine evaluation indicators for accuracy assessment. However, they do not consider the impact of dynamic factors such as illumination, temperature, humidity, and particulate matter concentration on images, lack an understanding of the probability of dynamic changes in images, and the evaluation information is incomplete. This invention establishes a mathematical model between environmental elements and predicted probabilities in outdoor scenes based on the causal relationship of the impact of dynamic factors on images, enabling the evaluation of the probability of generating images of outdoor scenes over a future period.

[0040] 3) Some methods that use deep learning and neural networks to evaluate the accuracy of generated images require training with a large number of scene images to assess the accuracy of the generated images. Furthermore, they need to be relearned for different outdoor scenes, resulting in poor universality. This invention uses environmental element data predicted from outdoor scenes and corresponding real data to calculate the predicted probability of the generated images. The parameter information is simple to obtain, applicable to different outdoor scenes, and highly portable. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the evaluation method for future generated images of outdoor scenes based on multiple environmental factors according to the present invention.

[0042] Figure 2 This is a schematic diagram of the prediction probability results for generating outdoor scene images according to the present invention;

[0043] Figure 3 The graph shows the probability density function results of the difference between visibility and refractive index structural constant in the target scene of this invention. Among them, (a) is the probability density function of the difference between visibility over three consecutive days; and (b) is the probability density function of the difference between refractive index structural constant over three consecutive days. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the following embodiments are by no means intended to limit the present invention.

[0045] This invention proposes an evaluation method for future generated images of outdoor scenes based on multiple environmental factors. The design idea is as follows: First, obtain the actual and predicted values ​​of meteorological data, as well as the predicted value of solar position information; then, calculate the predicted probability of illumination effects on the generated image using the total probability formula; finally, utilize the aerosol transfer function (MTF). a Prediction probability and turbulence transfer function (MTF) e The prediction probability is calculated to determine the impact of atmospheric effects on the generated image. Finally, the overall prediction probability of the generated image based on the combined effects of illumination and atmospheric effects is used to evaluate the quality of the future generated image for the outdoor scene. Figure 1 As shown, the specific steps are as follows:

[0046] Step 1: Obtain the actual values ​​of meteorological element data from the established meteorological station, and obtain the predicted values ​​of meteorological element data from the weather forecast issued by the local meteorological bureau. Use the current solar position information as the predicted value of the solar position information at the same time in the next 3 days. The solar position information includes the solar angle of incidence, the solar altitude angle, and the solar azimuth angle. The meteorological element data includes temperature, humidity, wind speed, PM2.5, and PM10.

[0047] Step 2: The prediction probability P(M) of the generated image based on illumination. The illumination part mainly considers the influence of changes in the sun's position on the image. The main influencing factor of P(M) is the sun's altitude angle H. S The azimuth of the sun, A S The angle of incidence θ of the sun i .

[0048] First, determine the date and time of the prediction, and the longitude and latitude of the target scene location. From this, obtain the geographical latitude φ, solar declination δ, and solar hour angle t of the target scene location at the prediction time. Then, use equations (1), (2), and (3) to obtain the solar altitude angle H of the target scene location at the prediction time. S The azimuth of the sun, A S The angle of incidence θ of the sun i :

[0049] sinH S =sinφ×sinδ+cosφ×cosδ×cost (1)

[0050] cosA S =(sinH S ×sinφ-sinδ)÷(cosH S ×cosφ) (2)

[0051] In equations (1) and (2), φ represents the geographical latitude, δ represents the solar declination, and t represents the solar hour angle.

[0052] cosθ i =cosH S cosA S (3)

[0053] In this invention, the solar altitude angle H at the predicted time of the target scene location is used. S The azimuth of the sun, A S The angle of incidence θ of the sun i As the true value, the predicted probability P(M) of the illumination effect on the generated image is calculated using equation (4) according to the law of total probability.

[0054] P(M)=P(M|A1)P(A1)+P(M|A2)P(A2) (4)

[0055] In equation (4), considering the accuracy and precision of camera time during the generation of the target scene image, P(A1) and P(A2) are the solar altitude angles H with a precision of seconds and the midpoint of the time point within 5 minutes, respectively. S The azimuth of the sun, A SThe probability that the actual value and the predicted value are the same. P(M|A1) and P(M|A2) are the angles of incidence of the sun within 5 minutes with a precision of seconds and the stated time point as the midpoint. i When the actual value and the predicted value are the same, the solar altitude angle H S The azimuth of the sun, A S The probability that the actual value and the predicted value are the same.

[0056] Step 3: The predicted probability P(N) of atmospheric effects on the generated image. During atmospheric transmission, the imaging light is affected by absorption and scattering from atmospheric particles, as well as atmospheric turbulence. Absorption and scattering mainly result from aerosol particles in the atmosphere, leading to a decrease in image brightness and contrast. Turbulence causes image blurring and distortion due to random variations in atmospheric temperature. The influence of atmospheric effects on the image can be obtained by applying the total atmospheric transfer function (MTF) estimated from meteorological data (temperature, humidity, particulate matter concentration, etc.) to a known image. The total atmospheric MTF is the aerosol transfer function MTF. a and turbulent transfer function MTF e The product of P(N) and P(B) can be assumed to have independent effects on the image. Therefore, the prediction probability of the atmospheric effect can be expressed as: P(N) = P(B) × P(C). Where P(B) and P(C) are the aerosol transfer functions MTF, respectively. a and turbulent transfer function MTF e Predicted probability, by MTF a The calculation formula can obtain accurate values ​​for parameters such as transmission distance R and radiation wavelength λ, without affecting the aerosol transfer function MTF. a The predicted visibility R m P(B) is calculated based on meteorological data forecasts, representing visibility R. m The confidence probability of a predicted value relative to the actual value. Calculate the predicted and actual visibility values ​​using meteorological data obtained from weather forecasts over a period of time, and simultaneously calculate the difference R between the predicted and actual values ​​for adjacent i-day periods. m_i And statistically fit to obtain the probability density function f(b) of the visibility difference between adjacent i days. i The threshold σ1 is determined based on different application requirements. Visibility differences within the threshold range are considered to be within the aerosol transfer function (MTF). a Predicted and true values ​​have the same effect on the image; the MTF of the weather transfer function for adjacent i-th elements is... a The predicted probability P(B) is

[0057]

[0058] Based on the actual and predicted values ​​of meteorological elements over the past N years (N = 1-10 years), the visibility R for each day is obtained using the visibility calculation formula for the target location.m The actual and predicted values; taking three consecutive days as a group, calculate the visibility R for each group of three days. m The difference R between the actual value and the predicted value m_i , i = 1, 2, 3; for all differences R m_i Statistical fitting was performed to obtain the probability density function f(b) of the visibility difference over the next 3 days. i The threshold σ1 is determined based on the visibility of the target scene location.

[0059] Similarly, from the calculation process of the turbulent transfer function MTFe, it can be seen that P(C) is the refractive index structure constant. The confidence probability of the predicted value relative to the actual value. The predicted and actual visibility values ​​calculated from weather forecast data over a period of time are statistically analyzed, and the difference between the predicted and actual values ​​of the refractive index structure constant within adjacent i-days is also calculated. And statistical fitting is used to obtain the probability density function f(c) of the visibility difference between adjacent i days. i The threshold σ² is determined based on different application requirements. The difference in refractive index structural constants within the threshold range is considered to be the turbulent transfer function (MTF). e Predicted and true values ​​have the same effect on the image; the turbulence transfer function (MTF) of adjacent i days... e The formula for calculating the prediction probability P(C) is:

[0060]

[0061] Based on the actual and predicted values ​​of meteorological elements from the past N years, the refractive index structure constant for each day is obtained using the formula for calculating the refractive index structure constant of the target scene location. The actual and predicted values; taking three consecutive days as a group, calculate the refractive index structure constant for each group of three days. The difference between the actual value and the predicted value For all differences Statistical fitting was performed to obtain the probability density function f(c) of the difference in refractive index structure constants over the next 3 days. i The threshold σ2 is determined based on the refractive index structure constant of the target scene location.

[0062] Step 4: Since the effects of illumination and atmosphere on outdoor scene images are independent of each other, we consider the total predicted probability P of the generated image within a future time period based on the effects of illumination and atmosphere. total =P(M)×P(N).

[0063] As described above, the method of the present invention is based on simple data information and takes into account the influence of multiple environmental factors on the image, and establishes a predictive probability model for the future generated image, thereby realizing the evaluation of the generated image effect in outdoor scenes.

[0064] Research materials:

[0065] Evaluation methods for future generated images of outdoor scenes based on multiple environmental factors, such as Figure 1 As shown, the process is as follows:

[0066] First, based on the actual values ​​of meteorological element data for 332 days in 2021 obtained from the established meteorological stations, the predicted values ​​of the corresponding meteorological data were obtained through the weather forecasts issued by the Tianjin Meteorological Bureau. The solar position information at the current moment was used as the predicted value of the solar position information at the same moment in the next 3 days. The solar position information includes the solar angle of incidence, the solar altitude angle, and the solar azimuth angle. The meteorological element data includes temperature, humidity, wind speed, PM2.5, and PM10.

[0067] In this study, taking 12:00 on March 27, 2021 as an example, a location in Nankai District, Tianjin, is used as the target scene location. The latitude and longitude of the target scene location are 39.1° East and 117.1° North. Based on this, the geographical latitude φ, solar declination δ, and solar hour angle t of the target scene location at the predicted time are calculated. Using equations (1) to (3), the solar altitude angle H within 5 minutes with a precision of seconds and the time point as the midpoint is calculated respectively. S The azimuth of the sun, A S The same probabilities P(A1) and P(A2) and the same angle of incidence θ of the sun. i When the actual value and the predicted value are the same, the solar altitude angle H S The azimuth of the sun, A S Using the same probabilities P(M|A1) and P(M|A2), and combining the relationship between the sun's incident angle, altitude angle, and azimuth angle, the confidence probability of the sun's position is calculated. Then, the predicted probability P(M) of the future three-day illumination effect on the generated image is calculated using equation (4). The results are as follows: Figure 2 As shown.

[0068] Based on the actual and predicted values ​​of the obtained meteorological element data, the visibility R for each day is obtained using the meteorological element data of the current target scene location (39.1°E, 117.1°N) and the visibility calculation formula. m The actual and predicted values; taking three consecutive days as a group, calculate the visibility R for each group of three days. m The difference R between the actual value and the predicted value m_i , i = 1, 2, 3; for all differences R m_i Statistical fitting was performed to obtain the probability density function f(b) of the visibility difference over the next 3 days. i ),like Figure 3As shown in (a). Based on the visibility of the target scene location, the threshold σ1 = 3km is determined, and the aerosol transfer function MTF for the next 3 days is calculated using equation (5). a The predicted probability P(B) is calculated. Similarly, based on the actual and predicted values ​​of the obtained meteorological element data, the refractive index structure constant for each day is obtained using the meteorological element data of the current target scene location (39.1°E, 117.1°N) and the refractive index structure constant. The actual and predicted values; taking three consecutive days as a group, calculate the refractive index structure constant for each group of three days. The difference between the actual value and the predicted value For all differences Statistical fitting was performed to obtain the probability density function f(c) of the visibility difference over the next 3 days. i ),like Figure 3 As shown in (b), the threshold σ² = 5 × 10⁻⁶ is determined based on the refractive index structure constant of the target scene location. -15 m -2 / 3 The turbulence transfer function (MTF) for the next 3 days is calculated using equation (6). e The predicted probability P(C) is calculated. Based on the predicted probability P(N) = P(B) × P(C) for the atmospheric effect, the predicted probability P(N) of the atmospheric effect on the generated image over the next 3 days is calculated, as shown below. Figure 2 As shown.

[0069] Finally, the prediction probability P(M) of the generated image caused by illumination and the prediction probability P(N) of the generated image caused by atmospheric effects are multiplied together to obtain the total prediction probability P of the generated outdoor scene images for the next 3 days. total =P(M)×P(N), which evaluates the generated images of the target scene over a future period of time, and the results are as follows. Figure 2 As shown.

[0070] This invention's evaluation method comprehensively considers environmental factors in outdoor scenes. It applies statistical methods to establish a mathematical model predicting the probability of future generated images of outdoor scenes, calculating the likelihood of such images occurring in the future, and thus evaluating the quality of future generated images of outdoor scenes. The data required for this invention is readily available, applicable to various outdoor scenes, and enables a more comprehensive and accurate evaluation of the quality of future generated images of outdoor scenes.

[0071] Although the present invention has been described above in conjunction with the accompanying drawings, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many modifications under the guidance of the present invention without departing from the spirit of the present invention, and these modifications are all within the protection scope of the present invention.

Claims

1. An evaluation method for future generated images of outdoor scenes based on multiple environmental factors, characterized in that, First, obtain the actual and predicted values ​​of meteorological data, as well as the predicted value of solar position information. Then, calculate the predicted probability of illumination effects on the generated image using a full probability calculation method. Calculate the predicted probability of atmospheric effects on the generated image using the aerosol transfer function and turbulence transfer function. Finally, evaluate the effect of the future generated image of the outdoor scene based on the total predicted probability of illumination and atmospheric effects on the generated image. This includes the following steps: Step 1: Obtain the actual values ​​of meteorological element data from the established meteorological station, and obtain the predicted values ​​of meteorological element data from the weather forecast issued by the local meteorological bureau. Use the current solar position information as the predicted value of the solar position information at the same time in the next 3 days. The solar position information includes the solar angle of incidence, the solar altitude angle, and the solar azimuth angle. The meteorological element data includes temperature, humidity, wind speed, PM2.5, and PM10. Step 2: The predictive probability of the generated image based on illumination ,include: Step 2-1) Determine the date and time of the prediction, and the longitude and latitude of the target scene location, thereby obtaining the geographical latitude of the target scene location at the predicted time. The declination of the sun and solar hour angle ; Step 2-2) Use equations (1), (2) and (3) to obtain the solar altitude angle at the predicted time of the target scene location. Sun's azimuth Angle of incidence of the sun : （1）； （2）； In equations (1) and (2), Indicates geographical latitude, Indicates the declination of the sun. Indicates solar hour angle, （3）； Steps 2-3) Predict the solar altitude angle at the target scene location at the predicted time. Sun's azimuth Angle of incidence of the sun As the true value; Steps 2-4) Calculate the prediction probability of the generated image by the illumination effect using equation (4). ; （4）； In equation (4), and These are the solar altitude angles, accurate to the second, within 5 minutes with the stated time point as the midpoint. Sun's azimuth The probability that the actual value and the predicted value are the same: and These are the angles of solar incidence within 5 minutes, with a precision of seconds and the stated time point as the midpoint. When the actual value and the predicted value are the same, the solar altitude angle Sun's azimuth The probability that the actual value and the predicted value are the same; Step 3: Predicting the probability of atmospheric effects on the generated image ,include: Step 3-1) Calculate the aerosol transfer function The predicted probability P(B) （5）； Based on the actual and predicted values ​​of meteorological elements over the past N years (N=1-10 years), the visibility for each day is obtained using the visibility calculation formula for the target location. The actual and predicted values; taking three consecutive days as a group, calculate the visibility for each of the three days in the group. The difference between the actual value and the predicted value , i=1,2,3; for all differences Statistical fitting was performed to obtain the probability density function of the visibility difference over the next 3 days. Determine the threshold based on the visibility of the target scene location. ; Step 3-2) Calculate the turbulent transfer function Predicted probability , （6）； Based on the actual and predicted values ​​of meteorological elements from the past N years, the refractive index structure constant for each day is obtained using the formula for calculating the refractive index structure constant of the target scene location. The actual and predicted values; taking three consecutive days as a group, calculate the refractive index structure constant for each group of three days. The difference between the actual value and the predicted value , i=1,2,3; for all differences Statistical fitting was performed to obtain the probability density function of the difference in refractive index structural constants for the next 3 days. The threshold is determined based on the refractive index structure constant of the target scene location. ; Step 3-3) Calculate the prediction probability of atmospheric effects on the generated image. , （7）； Step 4: Consider the combined effects of illumination and atmosphere on the overall prediction probability of the generated image. ; （8）； Through total predicted probability It enables the evaluation of the generated image effect in outdoor scenes.

Images