A training and verification method for a risk determination model of construction scope intrusion

Abstract

The present invention relates to the technical field of image recognition model training, and particularly to a training and verification method for a construction scope intrusion risk determination model; the method includes the following steps: S1, obtaining images corresponding to a plurality of regions of interest, and dividing the images into a plurality of image sets according to different regions of interest; S2, performing manual contour recognition and model contour recognition on each image in each image set respectively to obtain manual contour recognition images and model contour recognition images, and obtaining a plurality of training sets; S3, obtaining a total evaluation factor and a construction machinery evaluation factor; S4, setting a first evaluation set value and a second evaluation set value, comparing the total evaluation factor, the construction machinery evaluation factor with the first evaluation set value and the second evaluation set value, and judging whether the recognition model for performing model contour recognition in step S2 is iteratively trained according to different situations; improving the accuracy of region of interest recognition in construction scope intrusion risk determination.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition model training, and in particular to a training and verification method for a construction scope intrusion risk determination model. Background Art

[0002] The rapid development of the construction industry has brought high risks and frequent accidents. Factors such as the dense flow of personnel on the construction site and complex cross-operation links have led to frequent safety accidents in construction projects. Among them, close-range dangerous area intrusion accidents, such as collisions between workers and equipment or equipment, are the main types of accidents. In particular, when workers and mobile machinery are in the same construction space, human-machine collisions caused by the intersection of work trajectories are an important cause of such accidents. Workers' unintentional blindness to nearby machinery or machine drivers' negligence in observation often cause construction entities with different predetermined trajectories to overlap their work trajectories, triggering dangerous intrusions and leading to safety accidents. Accurately supervising the interaction status between workers and mobile construction machinery and timely warning of the risk of intrusion into dangerous areas of construction machinery are crucial to reducing on-site safety accidents. However, traditional manual inspection methods are too one-sided and inefficient, and existing sensor monitoring methods require high costs and have poor on-site application effects.

[0003] In the article "Intelligent Early Warning Method for Construction Machinery Intrusion into Dangerous Areas Based on Deep Learning and Depth Estimation" published by Wu Han and Han Yu, a method consisting of three parts: an image dataset of on-site workers and construction machinery, a target detection network, and a depth estimation network is disclosed. In the above method, a target detection network is used to identify areas of interest such as construction workers and construction machinery photographed by a camera, and the pixel coordinates of the area of ​​interest are obtained according to the identified contours. The pixel coordinates are converted into world coordinates through a monocular depth estimation network, and the distance between the construction workers and the construction machinery is obtained by calculation. Judgment is made based on the distance between the construction workers and the construction machinery, thereby performing intelligent early warning of danger; the above method also discloses training and verifying the target detection network through the small / small-seg model in Yolov8 and the average pixel accuracy method, but the accuracy of the target detection network trained and verified using the above method needs to be improved.

[0004] Therefore, there is an urgent need to provide a training and verification method for the construction scope intrusion risk determination model, so as to improve the accuracy of identifying the area of ​​interest in the construction scope intrusion risk determination compared with the existing technology. Summary of the invention

[0005] The present invention solves the technical problems existing in the prior art and provides a training and verification method for a construction scope intrusion risk determination model.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A training and verification method for a risk determination model of construction scope intrusion, comprising the following steps:

[0008] S1. Obtain images corresponding to multiple regions of interest, and divide the images into multiple image sets according to different regions of interest. The regions of interest include construction workers and multiple construction machines;

[0009] S2. For each image in each image set, perform manual contour recognition and model contour recognition respectively to obtain a manual contour recognition image and a model contour recognition image. Set multiple training sets, and place all the processed manual contour recognition images and model contour recognition images of one image set in one training set. Each training set includes multiple training pairs, and each training pair includes a manual contour recognition image and a model contour recognition image corresponding to one image;

[0010] S3. Obtain a total evaluation factor and a construction machine evaluation factor according to the multiple training sets obtained in step S2;

[0011] S4. Set a first evaluation set value and a second evaluation set value, compare the total evaluation factor, the construction machine evaluation factor with the first evaluation set value and the second evaluation set value to form different situations, and judge whether the recognition model for model contour recognition in step S2 is iteratively trained according to different situations.

[0012] Further, S3 specifically includes the following steps:

[0013] S31. Analyze each training set to obtain the evaluation factor of each training set;

[0014] S32. Obtain the total evaluation factor according to the evaluation factor of each training set, which is specifically calculated by the following formula:

[0015] ;

[0016] ;

[0017] In the above formula, E represents the total evaluation factor, represents the evaluation factor corresponding to the i-th training set, represents the corresponding precision weight value, represents the i-th set threshold;

[0018] S33. Set , , , to be the evaluation factors of the training sets corresponding to four different construction machines respectively, and take , , , The average value of is the construction machinery evaluation factor.

[0019] Furthermore, S31 specifically includes the following steps:

[0020] S311. Obtain the evaluation factors of each training pair in the i-th training set. The method for obtaining the evaluation factor of the j-th training pair in the i-th training set specifically includes the following steps:

[0021] S3111. For the j-th training pair in the i-th training set, obtain the pixel coordinates of all pixel points within the contour in the artificial contour recognition image to form a first pixel coordinate set; at the same time, obtain the pixels of all pixel points within the contour in the model contour recognition image to form a second pixel coordinate set;

[0022] S3112. Perform error value screening on the first pixel coordinate set to obtain a first screened pixel coordinate set after error screening. The first screened pixel coordinate set includes multiple first screened pixel coordinates;

[0023] S3113. Perform error value screening on the second pixel coordinate set to obtain a second screened pixel coordinate set after error screening. The second screened pixel coordinate set includes multiple second screened pixel coordinates;

[0024] S3114. According to the first screened pixel coordinate set and the second screened pixel coordinate set, obtain the evaluation factor of the j-th training pair in the i-th training set, where j takes values from 1 to Q, and Q represents the total number of training pairs in the i-th training set;

[0025] S312. Take the average value of the evaluation factors of all training pairs in the i-th training set obtained in step S311 to obtain the evaluation factor corresponding to the i-th training set.

[0026] Furthermore, in step S3114, the specific method for obtaining the evaluation factor of the j-th training pair in the i-th training set is as follows: For each first screened pixel coordinate, calculate the Euclidean distance with all second screened pixel coordinates to obtain the second screened pixel coordinate with the smallest Euclidean distance value from the first screened pixel coordinate. Combine the corresponding first screened pixel coordinate and the second screened pixel coordinate to form a sample pair. Each sample pair corresponds to a sample pair distance value, and the sample pair distance value is the Euclidean distance value between the first screened pixel coordinate and the second screened pixel coordinate in the sample pair;

[0027] Set a first distance judgment value and a second distance judgment value. The first distance judgment value is 0, and the second distance judgment value is greater than the first distance judgment value; Screen the number of sample pair distance values equal to the first distance judgment value, denoted as ;Filter the number of sample pair distance values that are greater than the first distance judgment value and less than or equal to the second distance judgment value, denoted as ;Filter the number of sample pair distance values that are greater than the second distance judgment value, denoted as ;

[0028] The evaluation factor of the j-th training pair in the i-th training set is calculated by the following formula:

[0029] ;

[0030] In the above formula, represents the evaluation factor of the j-th training pair in the i-th training set, represents the first weight value, represents the second weight value, represents the third weight value.

[0031] Furthermore, 、 、 are respectively calculated by the following formulas:

[0032] ;

[0033] ;

[0034] ;

[0035] In the above formula, represents the distance value of the l-th sample pair that is greater than the first distance judgment value and less than or equal to the second distance judgment value, represents the average value of the distance values of all sample pairs that are greater than the first distance judgment value and less than or equal to the second distance judgment value.

[0036] Furthermore, the second distance judgment value is calculated by the following formula:

[0037] ;

[0038] In the above formula, represents the second distance judgment value, A represents the first correction value, B represents the second correction value, and A and B respectively take different constants between 2% and 5%.

[0039] Further, in step S3112, the specific method for obtaining the first screened pixel coordinate set is as follows: Calculate the Euclidean distance between each first pixel coordinate in the first pixel coordinate set and all other first pixel coordinates except itself, obtaining a plurality of first Euclidean distances. All the first Euclidean distances form a first Euclidean distance set. Set a distance threshold for each training pair, where the distance threshold is the longest distance of the region of interest corresponding to the j-th training pair on the artificial contour recognition image. Compare each first Euclidean distance in the first Euclidean distance set with the distance threshold, and screen out the first Euclidean distances greater than the distance threshold. After screening out the first pixel coordinates corresponding to all the screened first Euclidean distances, the first screened pixel coordinate set is obtained.

[0040] Further, in step S3113, the specific method for obtaining the second screened pixel coordinate set is as follows: Calculate the Euclidean distance between each second pixel coordinate in the second pixel coordinate set and all other second pixel coordinates except itself, obtaining a plurality of second Euclidean distances. All the second Euclidean distances form a second Euclidean distance set. Compare each second Euclidean distance in the second Euclidean distance set with the distance threshold, and screen out the second Euclidean distances greater than the distance threshold. After screening out the second pixel coordinates corresponding to all the screened second Euclidean distances, the second screened pixel coordinate set is obtained.

[0041] Further, in step S4, the specific method for determining whether the recognition model for model contour recognition in step S2 is iteratively trained according to different situations is as follows:

[0042] (1) When , and , the recognition model is not iteratively trained;

[0043] (2) When , and , the recognition model is iteratively trained F times;

[0044] (3) When , and , the recognition model is iteratively trained F times;

[0045] (4) When , and , the recognition model is iteratively trained F times;

[0046] In the above formula, represents the total evaluation factor, represents the mechanical evaluation factor, represents the first evaluation set value, represents the second evaluation set value, , , All represent integers.

[0047] Furthermore, F, n, and m satisfy the following formula:

[0048] ;

[0049] .

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] The present invention obtains images of construction workers and construction machinery at the construction site, performs artificial contour recognition and model contour recognition, obtains multiple training sets, obtains corresponding evaluation factors for different training sets, thereby obtaining a total evaluation factor and a construction machinery evaluation factor, and sets the number of iterative training times of the recognition model according to different situations, so that the recognition model performs corresponding numbers of iterative training in different situations, improves the recognition accuracy of the recognition model, and thus improves the accuracy of judging the risk of intrusion into the construction scope. Description of the Drawings

[0052] Figure 1 is a flowchart of the present invention. Detailed Embodiments

[0053] The technical solutions of the present invention will be clearly described below in conjunction with the description of the drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0054] As Figure 1 shown, the present invention provides a method for training and verifying a risk determination model for intrusion into a construction scope, including the following steps:

[0055] S1. Obtain images collected on-site. The images include at least one region of interest. According to different regions of interest, the regions of interest include construction workers, excavators, trucks, drilling machines, and pile drivers. The images are divided into multiple image sets according to different regions of interest. In this embodiment, they are divided into a first image set, a second image set, a third image set, a fourth image set, and a fifth image set. The first image set includes multiple images of construction workers, the second image set includes multiple images of excavators, the third image set includes multiple images of trucks, the fourth image set includes multiple images of drilling machines, and the fifth image set includes multiple images of pile drivers.

[0056] S2. For each image in each image set, perform manual contour recognition and model contour recognition respectively to obtain a manually recognized contour image and a model recognized contour image. In both the manually recognized contour image and the model recognized contour image, the corresponding region of interest is framed with a contour. Set up multiple training sets. Place all the manually recognized contour images and model recognized contour images processed from one image set in one training set. Each training set includes multiple training pairs, and each training pair includes the manually recognized contour image and the model recognized contour image corresponding to one image.

[0057] S3. Based on the multiple training sets obtained in step S2, obtain the total evaluation factor and the construction machinery evaluation factor, which specifically includes the following steps:

[0058] S31. Analyze each training set to obtain the evaluation factor corresponding to each training set; the evaluation factor corresponding to the i-th training set is the i-th evaluation factor, denoted as , where i ranges from 1 to 5, The method for obtaining

[0059] specifically includes the following steps: , where j ranges from 1 to Q, and Q represents the total number of training pairs in the i-th training set, The method for obtaining

[0060] S311. For the j-th training pair in the i-th training set, obtain the pixel coordinates of all the pixels inside the contour in the manually recognized contour image to form a first pixel coordinate set, and at the same time, obtain the pixel coordinates of all the pixels inside the contour in the model recognized contour image to form a second pixel coordinate set.

[0061] The first pixel coordinate set includes multiple first pixel coordinates, and each first pixel coordinate is the pixel coordinate of a pixel in the manually recognized contour image of the j-th training pair in the i-th training set; the second pixel coordinate set includes multiple second pixel coordinates, and each second pixel coordinate is the pixel coordinate of a pixel in the model recognized contour image of the j-th training pair in the i-th training set.

[0062] S3112. Perform error value screening on the first pixel coordinate set to obtain a first screened pixel coordinate set , where, represents the first first-screened pixel coordinate, represents the t-th first-screened pixel coordinate, t represents the total number of first-screened pixel coordinates, , and g represents the total number of first pixel coordinates in the first pixel coordinate set.

[0063] The specific method for screening the error values of the first set of pixel coordinates to obtain the first screened set of pixel coordinates after error value screening is as follows: Calculate the Euclidean distance between each first pixel coordinate in the first set of pixel coordinates and all other first pixel coordinates except itself to obtain multiple first Euclidean distances. All the first Euclidean distances form a first Euclidean distance set. Set a distance threshold for each training pair. The distance threshold is the longest distance of the region of interest corresponding to the jth training pair in the artificial contour recognition image. Compare each first Euclidean distance in the first Euclidean distance set with the distance threshold, and screen out the first Euclidean distances greater than the distance threshold. After screening out the first pixel coordinates corresponding to all the screened first Euclidean distances, the first screened set of pixel coordinates is obtained.

[0064] S3113. Screen the error values of the second set of pixel coordinates to obtain the second screened set of pixel coordinates after error value screening. , ; among them, represents the first second screened pixel coordinate, represents the Tth second screened pixel coordinate, and T represents the total number of second screened pixel coordinates. , and G represents the total number of second pixel coordinates in the second set of pixel coordinates.

[0065] The specific method for screening the error values of the second set of pixel coordinates to obtain the second screened set of pixel coordinates after error value screening is as follows: Calculate the Euclidean distance between each second pixel coordinate in the second set of pixel coordinates and all other second pixel coordinates except itself to obtain multiple second Euclidean distances. All the second Euclidean distances form a second Euclidean distance set. Compare each second Euclidean distance in the second Euclidean distance set with the distance threshold, and screen out the second Euclidean distances greater than the distance threshold. After screening out the second pixel coordinates corresponding to all the screened second Euclidean distances, the second screened set of pixel coordinates is obtained.

[0066] S3114. Calculate the Euclidean distance between each first screened pixel coordinate and all the second screened pixel coordinates to obtain the second screened pixel coordinate with the smallest Euclidean distance value for each first screened pixel coordinate, and obtain multiple sample pairs. Each sample pair includes a first screened pixel coordinate and the second screened pixel coordinate with the smallest Euclidean distance value to it. Each sample pair has a Euclidean distance value, and the Euclidean distance value of each sample pair is set as the sample pair distance value; Set the first distance judgment value and the second distance judgment value . The second distance judgment value is greater than the first distance judgment value. Compare all the sample pair distance values with the first distance judgment value and the second distance judgment value. The number of sample pair distance values equal to the first distance judgment value is denoted as ; The number of sample pair distance values greater than the first distance judgment value and less than or equal to the second distance judgment value is denoted as , and the number of sample pair distance values greater than the second distance judgment value is denoted as .

[0067] The first distance judgment value and the second distance judgment value are calculated according to the following formula:

[0068] ;

[0069] ;

[0070] In the above formula, A represents the first correction value, B represents the second correction value, and A and B respectively take different constants between 2% and 5%.

[0071] According to , , , the evaluation factor of the j-th training pair in the i-th training set is obtained, and is specifically expressed as:

[0072] ;

[0073] In the above formula, represents the evaluation factor of the j-th training pair in the i-th training set, represents the first weight value, represents the second weight value, represents the third weight value.

[0074] , , are respectively calculated according to the following formula:

[0075] ;

[0076] ;

[0077] ;

[0078] In the above formula, represents the l-th sample pair distance value greater than the first distance judgment value and less than or equal to the second distance judgment value, represents the average value of all sample pair distance values greater than the first distance judgment value and less than or equal to the second distance judgment value.

[0079] S312. According to the evaluation factors of each training pair in the i-th training set, the evaluation factor of the i-th training set is obtained, and is specifically calculated according to the following formula:

[0080] .

[0081] S32. Obtain the total evaluation factor based on the evaluation factors of each training set, which is specifically calculated by the following formula:

[0082] ;

[0083] ;

[0084] In the above formula, E represents the total evaluation factor, represents the corresponding precision weight value, represents the i-th set threshold, which is obtained by taking the average of the distance thresholds of all training pairs in the i-th training set.

[0085] S33. Set , , , , to be the evaluation factors of the training sets corresponding to construction workers, excavators, trucks, drilling machines, and pile drivers respectively. The construction machinery evaluation factor is obtained by the following formula:

[0086] ;

[0087] In the above formula, represents the construction machinery evaluation factor.

[0088] S4. According to the total evaluation factor and the construction machinery evaluation factor obtained in step S3, determine whether the recognition model for model contour recognition in step S2 needs to be iteratively trained. The specific method is as follows:

[0089] Set the first evaluation setting value and the second evaluation setting value, which are set to and respectively, and .

[0090] (1) When , and , the recognition model does not perform iterative training.

[0091] (2) When , and , the recognition model performs iterative training F times.

[0092] (3) When , and , the recognition model performs iterative training F times.

[0093] (4) When , and When, the recognition model performs iterative training F times.

[0094] F, n, and m satisfy the following formula:

[0095] ;

[0096] ;

[0097] , n, and m are all integers.

[0098] The present invention obtains images of construction workers and construction machinery at the construction site, performs manual contour recognition and model contour recognition to obtain multiple training sets, obtains corresponding evaluation factors for different training sets, thereby obtaining the total evaluation factor and the construction machinery evaluation factor, and sets the number of iterative training times of the recognition model according to the total evaluation factor and the construction machinery evaluation factor, so that when in different situations, the recognition model performs corresponding numbers of iterative training, improves the recognition accuracy of the recognition model, and thus improves the accuracy of the determination of the risk of intrusion into the construction scope.

[0099] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than a limitation on the protection scope of the present invention. Any simple modification or equivalent replacement of the technical solution of the present invention by those of ordinary skill in the art does not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A training and verification method for a construction scope intrusion risk determination model, characterized in that: The following steps are involved: S1, acquiring images corresponding to a plurality of regions of interest, and dividing the images into a plurality of image sets according to different regions of interest, where the regions of interest include construction workers and a plurality of construction machines; S2. For each image in each image set, artificial contour recognition and model contour recognition are performed respectively to obtain artificial contour recognition images and model contour recognition images, and multiple training sets are set. All artificial contour recognition images and model contour recognition images processed by one image set are placed in one training set. Each training set includes multiple training pairs, and each training pair includes an artificial contour recognition image and a model contour recognition image corresponding to one image. S3, obtaining a total evaluation factor and a construction machinery evaluation factor according to the multiple training sets obtained in step S2; specifically comprising the following steps: S31, analyzing each training set to obtain an evaluation factor for each training set; specifically comprising the following steps: S311, obtaining the evaluation factor of each training pair in the ith training set, wherein the method for obtaining the evaluation factor of the jth training pair in the ith training set specifically comprises the following steps: S3111, for the j-th training pair in the i-th training set, obtaining pixel coordinates of all pixel points in the contour in the artificial contour recognition image to form a first pixel coordinate set; and simultaneously obtaining pixel coordinates of all pixel points in the contour in the model contour recognition image to form a second pixel coordinate set; S3112, performing error value screening on the first pixel coordinate set to obtain a first screened pixel coordinate set that has undergone error screening, the first screened pixel coordinate set including a plurality of first screened pixel coordinates; S3113, performing error value screening on the second pixel coordinate set to obtain a second screened pixel coordinate set that has undergone error screening, the second screened pixel coordinate set including a plurality of second screened pixel coordinates; S3114, obtaining an evaluation factor of the jth training pair in the ith training set according to the first screening pixel coordinate set and the second screening pixel coordinate set, where j ranges from 1 to Q, and Q represents the total number of training pairs in the ith training set; In step S3114, the specific method for obtaining the evaluation factor of the j-th training pair in the i-th training set is as follows: for each first screening pixel coordinate, the Euclidean distance calculation is performed with all the second screening pixel coordinates to obtain the second screening pixel coordinate with the smallest Euclidean distance value with the first screening pixel coordinate, and the corresponding first screening pixel coordinates and the second screening pixel coordinates are combined into a sample pair, each sample pair corresponds to a sample pair distance value, and the sample pair distance value is the Euclidean distance value between the first screening pixel coordinate and the second screening pixel coordinate in the sample pair; Set a first distance judgment value and a second distance judgment value, the first distance judgment value is 0, and the second distance judgment value is greater than the first distance judgment value; select the number of sample pairs with distance values ​​equal to the first distance judgment value, recorded as ; Screen the number of sample distance values ​​greater than the first distance judgment value and less than or equal to the second distance judgment value, recorded as ; Screen the number of sample distance values ​​greater than the second distance judgment value, recorded as ; The evaluation factor of the jth training pair in the i-th training set is calculated by the following formula: ； In the above formula, represents the evaluation factor of the jth training pair in the i-th training set, represents the first weight value, represents the second weight value, represents the third weight value; S312, taking the average of the evaluation factors of all training pairs in the i-th training set obtained in step S311, to obtain the evaluation factor corresponding to the i-th training set; S32. According to the evaluation factor of each training set, the total evaluation factor is obtained, which is specifically calculated by the following formula: ； ； In the above formula, E represents the total evaluation factor, represents the evaluation factor corresponding to the i-th training set, express The corresponding precision weight value, represents the i-th set threshold; S33, Settings , , , are the evaluation factors of the training sets corresponding to four different construction machines. , , , The average value of the construction machinery evaluation factor ; S4. Set a first evaluation setting value and a second evaluation setting value, compare the total evaluation factor and the construction machinery evaluation factor with the first evaluation setting value and the second evaluation setting value, form different situations, and determine whether the recognition model for model contour recognition in step S2 should be iteratively trained according to the different situations.

2. The training and verification method of the construction scope intrusion risk determination model according to claim 1 is characterized in that: , , They are calculated by the following formulas: ； ； ； In the above formula, represents the distance value of the lth sample pair that is greater than the first distance judgment value and less than or equal to the second distance judgment value, Represents the average value of all sample distance values ​​that are greater than the first distance judgment value and less than or equal to the second distance judgment value.

3. The training and verification method of a construction scope intrusion risk determination model according to claim 1 is characterized in that: The second distance judgment value is calculated according to the following formula: ； In the above formula, represents the second distance judgment value, A represents the first correction value, B represents the second correction value, and A and B respectively take different constants between 2% and 5%.

4. The training and verification method of a construction scope intrusion risk determination model according to claim 1 is characterized in that: In step S3112, the specific method for obtaining the first filtered pixel coordinate set is: for each first pixel coordinate in the first pixel coordinate set, the Euclidean distance is calculated with all other first pixel coordinates except itself, to obtain multiple first Euclidean distances, all first Euclidean distances form a first Euclidean distance set, a distance threshold is set for each training pair, and the distance threshold is the longest distance of the region of interest corresponding to the jth training pair on the artificial contour recognition image; each first Euclidean distance in the first Euclidean distance set is compared with the distance threshold, and the first Euclidean distance greater than the distance threshold is screened out, and the first pixel coordinates corresponding to all the screened first Euclidean distances are screened out to obtain the first filtered pixel coordinate set.

5. The training and verification method of the construction scope intrusion risk determination model according to claim 4 is characterized in that: In step S3113, the specific method for obtaining the second filtered pixel coordinate set is: calculating the Euclidean distance between each second pixel coordinate in the second pixel coordinate set and all other second pixel coordinates except itself, obtaining multiple second Euclidean distances, all second Euclidean distances form a second Euclidean distance set, comparing each second Euclidean distance in the second Euclidean distance set with a distance threshold, filtering out second Euclidean distances greater than the distance threshold, filtering out the second pixel coordinates corresponding to all the filtered second Euclidean distances, and obtaining the second filtered pixel coordinate set.

6. The training and verification method of a construction scope intrusion risk determination model according to claim 1 is characterized in that: In step S4, the specific method for judging whether the recognition model for model contour recognition in step S2 should be iteratively trained according to different situations is: (1) When ,and When , the recognition model does not perform iterative training; (2) When ,and When , the recognition model is iteratively trained F times; (3) When ,and When F Second-rate; (4) When ,and When F Second-rate; In the above formula, represents the overall evaluation factor, represents the mechanical evaluation factor, represents the first evaluation setting value, represents the second evaluation setting value, , , All represent integers.

7. The training and verification method of the construction scope intrusion risk determination model according to claim 6 is characterized in that: F, n, and m satisfy the following formula: ； 。

Images