Building night scene lighting anomaly detection method based on RetinaXNet

Through the RetinaXNet network model, the construction night view lights are automatically detected, which solves the problems of high manual inspection costs, low real-time and strong subjectivity, and achieves efficient and accurate lighting abnormality detection.

CN113762084BActive Publication Date: 2024-11-29JIANGXI WANGKAI CONSTR CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202110909371.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-08-09
Publication Date
2024-11-29
Estimated Expiration
2041-08-09

AI Technical Summary

Technical Problem

The light inspection of night view of existing buildings mainly relies on manual inspection, which has problems such as high cost, low real-time and strong subjectivity, making it difficult to effectively detect lighting failures.

Method used

The image processing method based on RetinaXNet is adopted to collect video data through the camera, build a collection of night view light images, perform data preprocessing and network model training, and use the RetinaXNet network model to perform abnormal detection to achieve automated detection.

Benefits of technology

It reduces detection costs, reduces artificial subjectivity, improves detection accuracy, and maintains good detection results under changes in lighting, angle and clarity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113762084B_ABST
    Figure CN113762084B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for detecting abnormal lighting of building night scenes based on RetinaXNet. The data set adopts equalization processing to retain the texture information of the image and reduce the complexity of the image. The input module of the RetinaXNet network reduces the video frame to a 224*224 image, the backbone module adopts an improved residual structure to extract the contour information of the image, the detection head module adopts the XNet network to strengthen the integration of information, and performs classification and regression, and the output module restores the image to its original size according to the reduction ratio. The RetinaXNet network proposed in the present invention can be used to detect the position of the fault light in the image and the fault classification, realize automatic detection of abnormalities, improve the accuracy of detection, reduce the situation of false detection, and provide a reliable method for detecting abnormal lighting of building night scenes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image processing and anomaly detection, and in particular to a method for detecting anomalies of building night scene lights based on RetinaXNet. Background Art

[0002] With the application of modern urban science and technology and the rapid development of economic strength, urban lighting projects have played a significant role in improving the urban environment and building livable cities, enhancing the overall functions of cities, stimulating domestic demand, promoting the development of urban economy, and enhancing the image of corresponding enterprises. However, due to the fact that the layout of building night lights is exposed to the outdoors all year round, due to problems such as aging of lamps, installation environment and heat dissipation, building night lights frequently fail. The existing detection methods are mainly based on manual inspection and visual inspection, which has the disadvantages of high cost, low real-time performance and strong subjectivity. With the development of artificial intelligence technology, deep learning-based detection methods can replace traditional manual methods in some image-related fields, and use pre-trained networks to detect abnormalities in building night lights, which improves the accuracy of detection and reduces human subjectivity, and realizes the automation of detection. Summary of the invention

[0003] Aiming at the shortcomings of current manual inspection, such as high cost, low real-time performance and strong subjectivity, a building night light anomaly detection method based on RetinaXNet is proposed. Through cameras and network models, the night light anomaly detection is automated and the detection accuracy is improved.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows: a method for detecting abnormal lighting of night scenes of buildings based on RetinaXNet, comprising the following steps:

[0005] 1) Construct an initial night scene light image set C and send it to the GPU computing server for storage;

[0006] 2) Process the image set C to obtain a data set, and divide the data set into a training set E and a test set T;

[0007] 3) Constructing a RetinaXNet network model; the RetinaXNet network model includes an input module, a backbone module, and a detection head module;

[0008] 4) Use the UFL function to optimize the weights of the RetinaXNet network model;

[0009] 5) Train the RetinaXNet network model;

[0010] 6) Night scene light anomaly detection, that is, obtaining the frame to be detected through the camera and sending it to the RetinaXNet network, mapping the network output back to the original image, and judging whether the night scene light is abnormal.

[0011] Furthermore, the above step 1) includes the following steps:

[0012] 1.1) Using a camera to collect video data V of night scene lights, the camera is pre-fixedly installed at a location where night scene lights need to be detected;

[0013] 1.2) Extract a frame of image from the video data V at every time Δt to construct an initial night scene light image set C, denoted as C = {I1, I2...I n}, I i is the i-th frame image, n is the number of night light images;

[0014] 1.3) Send the initial night light image set C to the GPU computing server for storage.

[0015] Furthermore, the above step 2) includes the following steps:

[0016] 2.1) For each frame of the image set C, calculate the probability p of a pixel with a value less than j i (j), the calculation formula is as follows:

[0017]

[0018] p i (j) represents the probability of the grayscale in the i-th frame being greater than 0 and less than j, n t is the number of pixels with gray level less than j, n I is the total number of pixels in each frame;

[0019] 2.2) Calculate the histogram result G(i) of each frame image in the set C. The calculation formula is as follows:

[0020]

[0021] G(i) is the grayscale histogram processing result of the i-th frame, where 0≤i<256, p i (j) represents the probability of the pixel in the i-th frame being greater than 0 and less than j;

[0022] 2.3) Calculate the pixel equalization result H(v), and equalize the image set C, where C = {I1, I2...I n}, the processed image set is recorded as C′, C′={I1′,I2′...I n '}, the calculation formula is as follows:

[0023]

[0024] Where v is the pixel value of a single image I in the image set C, H(v) is the result of equalization of v, G(v) is the histogram processing result of the current v, and G min is the minimum value of histogram processing, G max is the maximum value of the histogram processing, L is the grayscale level, and round represents the rounding of the pixel value result. After all pixels are calculated, a single image I' is obtained, and the set is recorded as C';

[0025] 2.4) Calculate the average pixel value a for the images in the image set C′, where C′={I1′,I2′...I n '}, the calculation formula is as follows:

[0026]

[0027] Where M is the length pixel value of the image, N is the width pixel value of the image, and I t '(r,c) is the coordinate of the image pixel in the image set C', and t is the number of the selected image;

[0028] 2.5) Perform missing and filling processing on the images in the image set C′, the missing and filled values ​​are g(i,j), and the data set C' is obtained. The calculation formula is as follows:

[0029]

[0030] In the formula, g(i,j) is the missing and filled value, I'(i,j) is the pixel value of the image I' with coordinates (i,j) in the image set C', and Th is the set threshold;

[0031] 2.6) Divide the dataset C' into training set E and test set T in the ratio m:n.

[0032] Furthermore, the above step 3) constructing the RetinaXNet network model includes the following steps:

[0033] 3.1) Using the input module, the images in the training set E are uniformly reduced to images of size r:l, where r is the reduced length pixel value, l is the reduced width pixel value, and the transformed set is recorded as X = {x1, x2...x3};

[0034] 3.2) Use the backbone module to extract the contour features of each frame in the set X through the residual structure. The residual structure formula is as follows:

[0035] F(x)=f(x)+f(f(x))+f(f(f(x)))

[0036] Where f(x) = δ(W*x) + c

[0037] In the formula, each frame image x is used as the input of the convolution layer, W is the parameter to be learned for convolution, δ is the activation function, and F(x) is the output result of the residual structure;

[0038] 3.3) Set the input parameter structure F of classification fusion in the detection head module CLS , the formula is as follows:

[0039] F CLS =δ[W CLS *F(x)]+c

[0040] Where F(x) is the output result of the residual structure, W CLS is the training parameter, δ is the activation function, and c is the constant term;

[0041] 3.4) Structure for setting nonlinear regression connection parameters in the detection head module The formula is as follows:

[0042]

[0043] In the formula, F CLS is the input part of parameter fusion, W1 and W2 are training parameters, δ is the activation function, c is the constant term, and LCR connection is a nonlinear connection, which is used to strengthen the connection between classification and regression;

[0044] 3.5) Calculate the enhanced parameters of regression The calculation formula is as follows:

[0045]

[0046] In the formula, F reg is the original regression parameter of the network, is the enhanced regression parameter;

[0047] 3.6) Structure for setting nonlinear classification connection parameters in the detection head module The formula is as follows:

[0048]

[0049] Where W3 is the training parameter, is the enhanced regression parameter, The structure of the connection parameters for nonlinear classification;

[0050] 3.7) Calculate the enhanced parameters of classification The calculation formula is as follows:

[0051]

[0052] In the formula, F cls is the original classification parameter of the network, is an enhanced classification parameter.

[0053] Furthermore, in step 4), the UFL function is used to optimize the weights of the RetinaXNet network model. The optimization formula is as follows:

[0054]

[0055] In the formula, y is the true label, which takes the value of 0 or 1. is the dynamic adjustment factor parameter of UFL, γ is the rate of adjusting sample weight, and α is the weight parameter.

[0056] Furthermore, the above step 5) training the RetinaXNet network model includes the following steps:

[0057] 5.1) Calculate the accuracy. The calculation formula is as follows:

[0058]

[0059] Where, accuracy is the accuracy rate, TP represents the number of positive samples whose network output is in the test set T and whose reference standard is also positive samples; TN represents the number of negative samples whose network output is in the test set T and whose reference standard is also negative samples; FP represents the number of positive samples whose network output is in the test set T and whose reference standard is negative samples; FN represents the number of positive samples whose network output is in the test set T and whose reference standard is positive samples.

[0060] 5.2) Calculate the recall rate recall, the calculation formula is as follows:

[0061]

[0062] In the formula, recall is the recall rate. TP represents the number of samples that are judged as positive samples, that is, the number of positive samples; FN represents the number of samples that are judged as negative samples but are actually positive samples;

[0063] 5.3) Calculate the F1 value using the following formula:

[0064]

[0065] In the formula, F1 is the balance between accuracy and recall, and is the calculation result of the comprehensive accuracy and recall rate;

[0066] 5.4) Determine whether F1 is less than t. If so, go to step 5.1) and retrain. Otherwise, go to step 6).

[0067] Furthermore, the above step 6) determining whether the night view light is abnormal includes the following steps:

[0068] 6.1) Map the network output back to the original image. The coordinate, length and height formulas mapped to the original image are as follows:

[0069]

[0070]

[0071]

[0072]

[0073] In the formula, is the length and height of the original image, x, y, W, and H are the coordinates of the upper left corner of the detection frame and the length and height of the detection frame respectively, x N ,y N , W N , H N The coordinates, length and height mapped to the original image;

[0074] 6.2) Anomaly detection, the detection formula is as follows:

[0075]

[0076] Wherein, Warn=0 means no abnormality, and Warn=1 means abnormality occurs and the alarm is triggered.

[0077] Compared with the traditional method of detecting abnormal lighting in night scenes of buildings, the present invention can save costs, reduce the subjectivity of human judgment, and improve detection efficiency. It can also achieve good detection effects under conditions of lighting changes, angle changes, and impaired clarity. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 The figure is a flow chart of the method for detecting abnormal lighting of a building night scene in the present invention. DETAILED DESCRIPTION

[0079] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the technical solution and design principle of the present invention are described in detail below, but the protection scope of the present invention is not limited thereto. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essential content of the present invention belong to the protection scope of the present invention.

[0080] The RetinaXNet-based building night scene lighting anomaly detection method described in the present invention uses devices including an Internet of Things camera, a GPU computing server, and an alarm.

[0081] The process of the building night scene lighting anomaly detection method based on RetinaXNet of the present invention is as follows: Figure 1 As shown, the following steps are included:

[0082] 1) Constructing an initial night scene light image set C and sending it to a GPU computing server for storage, as a preferred embodiment of the present invention, includes the following steps:

[0083] 1.1) Using a camera to collect video data V of night scene lights, wherein the camera is pre-fixedly installed at a location where night scene lights need to be detected;

[0084] 1.2) Extract a frame of image from the video data V at every time Δt to construct an initial night scene light image set C, denoted as C = {I1, I2...I n}, I i is the i-th frame image, n is the number of night light images;

[0085] 1.3) Sending the initial night scene light image set C to the GPU computing server for storage;

[0086] 2) Processing the image set C to obtain the image set C′, performing missing and filling processing on the images in the image set C′ to obtain a data set C″, and dividing C″ into a training set E and a test set T; as a preferred embodiment of the present invention, the following steps are included:

[0087] 2.1) For each frame of the image set C, calculate the probability p of a pixel with a value less than j i (j), the calculation formula is as follows:

[0088]

[0089] p i (j) represents the probability of the grayscale in the i-th frame being greater than 0 and less than j, n t is the number of pixels with gray level less than j, n I is the total number of pixels in each frame;

[0090] 2.2) Calculate the histogram result G(i) of each frame image in the set C. The calculation formula is as follows:

[0091]

[0092] G(i) is the grayscale histogram processing result of the i-th frame, where 0≤i<256, p i (j) represents the probability of occurrence of pixels greater than 0 and less than j in the i-th frame image.

[0093] 2.3) Calculate the pixel equalization result H(v), and equalize the image set C, where C = {I1, I2...I n}, the processed image set is recorded as C′, C′={I1′,I2′...I n '}, the calculation formula is as follows:

[0094]

[0095] Where v is the pixel value of a single image I in the image set C, H(v) is the result of equalization of v, G(v) is the histogram processing result of the current v, and G min is the minimum value of histogram processing, G max is the maximum value of the histogram processing, L is the grayscale level, and round represents the rounding of the pixel value result. After all pixels are calculated, a single image I' is obtained, and the set is recorded as C'.

[0096] 2.4) Calculate the average pixel value a for the images in the image set C′, where C′={I1′,I2′...I n '}, the formula is as follows:

[0097]

[0098] Where M is the length pixel value of the image, N is the width pixel value of the image, and I t '(r,c) is the coordinate of the image pixel in the image set C', and t is the number of the selected image;

[0099] 2.5) Perform missing and filling processing on the images in the image set C′, the missing and filled values ​​are g(i,j), and the data set C' is obtained. The formula is as follows:

[0100]

[0101] Wherein, g(i,j) is the missing and filled value, I'(i,j) is the pixel value of the image I' with coordinates (i,j) in the image set C', Th is the set threshold value, and in the specific embodiment of the present invention, Th=180;

[0102] 2.6) Dividing the data set C' into a training set E and a test set T according to a ratio m:n. In a specific embodiment of the present invention, m:n=9:1;

[0103] 3) Constructing a RetinaXNet network model; wherein the RetinaXNet network model includes an input module, a backbone module, and a detection head module; as a preferred embodiment of the present invention, the following steps are included:

[0104] 3.1) Using the input module, the images in the training set E are uniformly reduced to images of size r:l, where r is the reduced length pixel value, l is the reduced width pixel value, and the transformed set is recorded as X = {x1, x2...x3}; in the specific embodiment of the present invention, r = 224, l = 224;

[0105] 3.2) Use the backbone module to extract the contour features of each frame in the set X through the residual structure. The residual structure formula is as follows:

[0106] F(x)=f(x)+f(f(x))+f(f(f(x)))

[0107] Where f(x) = δ(W*x) + c

[0108] In the formula, each frame image x is used as the input of the convolution layer, W is the parameter to be learned for the convolution, δ is the activation function, and F(x) is the output result of the residual structure. The method uses n repeated calculations, and in a specific embodiment of the present invention, n=50 is set.

[0109] 3.3) Set the input parameter structure F of classification fusion in the detection head module CLS , the formula is as follows:

[0110] F CLS =δ[W CLS *F(x)]+c

[0111] Where F(x) is the output result of the residual structure, W CLS is the training parameter, δ is the activation function, and c is the constant term.

[0112] 3.4) Structure for setting nonlinear regression connection parameters in the detection head module The formula is as follows:

[0113]

[0114] In the formula, F CLS It is the input part of parameter fusion, W1 and W2 are training parameters, δ is the activation function, c is the constant term, and LCR connection is a nonlinear connection, which serves to strengthen the connection between classification and regression.

[0115] 3.5) Calculate the enhanced parameters of regression The formula is as follows:

[0116]

[0117] In the formula, F reg is the original regression parameter of the network, are the enhanced regression parameters.

[0118] 3.6) Structure for setting nonlinear classification connection parameters in the detection head module The formula is as follows:

[0119]

[0120] Where W3 is the training parameter, is the enhanced regression parameter, The structure of the connection parameters for nonlinear classification.

[0121] 3.7) Calculate the enhanced parameters of classification The formula is as follows:

[0122]

[0123] In the formula, F cls is the original classification parameter of the network, is an enhanced classification parameter.

[0124] 4) Use the UFL function to optimize the weights of the RetinaXNet network model; the formula is as follows:

[0125]

[0126] In the formula, y is the true label, which takes the value of 0 or 1. is the dynamic adjustment factor parameter of UFL, γ is the rate of adjusting the sample weight, α is the weight parameter, in the specific embodiment of the present invention, γ=2, =0.25;

[0127] 5) Training the network model, as a preferred embodiment of the present invention, includes the following steps:

[0128] 5.1) Calculate the accuracy. The calculation formula is as follows:

[0129]

[0130] Where, accuracy is the accuracy rate, TP represents the number of positive samples whose network output is in the test set T and whose reference standard is also positive samples; TN represents the number of negative samples whose network output is in the test set T and whose reference standard is also negative samples; FP represents the number of positive samples whose network output is in the test set T and whose reference standard is negative samples; FN represents the number of positive samples whose network output is in the test set T and whose reference standard is positive samples.

[0131] 5.2) Calculate the recall rate recall, the calculation formula is as follows:

[0132]

[0133] In the formula, recall is the recall rate. TP represents the number of samples that are judged as positive samples, that is, the number of positive samples; FN represents the number of samples that are judged as negative samples but are actually positive samples;

[0134] 5.3) Calculate the F1 value using the following formula:

[0135]

[0136] In the formula, F1 is the balance between accuracy and recall, and is the calculation result of the comprehensive accuracy and recall.

[0137] 5.4) Determine whether F1 is less than t, if so, go to step 5.1) for retraining, otherwise go to step 6), in a specific embodiment of the present invention, t = 0.6;

[0138] 6) The frame to be detected is obtained through the camera and sent to the network for detection, and the output result of the network is mapped back to the original image to determine whether a fault light appears. As a preferred embodiment of the present invention, the following steps are included:

[0139] 6.1) Map the network output back to the original image. The coordinate, length and height formulas mapped to the original image are as follows:

[0140]

[0141]

[0142]

[0143]

[0144] In the formula, is the length and height of the original image, x, y, W, and H are the coordinates of the upper left corner of the detection frame and the length and height of the detection frame respectively, N ,y N , W N , H N The coordinates, length and height mapped to the original image.

[0145] 6.2) Anomaly detection, the detection formula is as follows:

[0146]

[0147] Wherein, Warn=0 means no abnormality, and Warn=1 means abnormality occurs and the alarm is triggered.

Claims

1. A method for detecting abnormal lighting in night scenes of buildings based on RetinaXNet, characterized in that: The steps include: 1) Construct an initial night scene light image set C and send it to the GPU computing server for storage; 2) Process the image set C to obtain a data set, and divide the data set into a training set E and a test set T; 3) Constructing a RetinaXNet network model; the RetinaXNet network model includes an input module, a backbone module, and a detection head module; 4) Use the UFL function to optimize the weights of the RetinaXNet network model; 5) Train the RetinaXNet network model; 6) Night scene light anomaly detection, that is, obtaining the frame to be detected through the camera and sending it to the RetinaXNet network, mapping the network output back to the original image, and judging whether the night scene light is abnormal; The step 3) constructing the RetinaXNet network model includes the following steps: 3.1) Using the input module, the images in the training set E are uniformly reduced to images of size r:l, where r is the reduced length pixel value, l is the reduced width pixel value, and the transformed set is recorded as X = {x1, x2...x3}; 3.2) Use the backbone module to extract the contour features of each frame in the set X through the residual structure. The residual structure formula is as follows: F(x)=f(x)+f(f(x))+f(f(f(x))) Where f(x) = δ(W*x) + c In the formula, each frame image x is used as the input of the convolution layer, W is the parameter to be learned for convolution, δ is the activation function, and F(x) is the output result of the residual structure; 3.3) Set the input parameter structure F of classification fusion in the detection head module CLS , the formula is as follows: F CLS =δ[W CLS *F(x)]+c Where F(x) is the output result of the residual structure, W CLS is the training parameter, δ is the activation function, and c is the constant term; 3.4) Structure for setting nonlinear regression connection parameters in the detection head module The formula is as follows: In the formula, F CLS is the input part of parameter fusion, W1 and W2 are training parameters, δ is the activation function, c is the constant term, and LCR connection is a nonlinear connection, which is used to strengthen the connection between classification and regression; 3.5) Calculate the enhanced parameters of regression The calculation formula is as follows: In the formula, F reg is the original regression parameter of the network, is the enhanced regression parameter; 3.6) Structure for setting nonlinear classification connection parameters in the detection head module The formula is as follows: Where W3 is the training parameter, is the enhanced regression parameter, The structure of the connection parameters for nonlinear classification; 3.7) Calculate the enhanced parameters of classification The calculation formula is as follows: In the formula, F cls is the original classification parameter of the network, is an enhanced classification parameter.

2. The method for detecting abnormal lighting of night scenes of buildings based on RetinaXNet as claimed in claim 1, characterized in that: The step 1) comprises the following steps: 1.1) Using a camera to collect video data V of night scene lights, the camera is pre-fixedly installed at a location where night scene lights need to be detected; 1.2) Extract one frame of image from the video data V at every time Δt to construct an initial night scene light image set C, denoted as C = {I1, I2…I n }, I i is the i-th frame image, n is the number of night light images; 1.3) Send the initial night light image set C to the GPU computing server for storage.

3. The method for detecting abnormal lighting of night scenes of buildings based on RetinaXNet as claimed in claim 1, characterized in that: The step 2) comprises the following steps: 2.1) For each frame of the image set C, calculate the probability p of a pixel with a value less than j i (j), the calculation formula is as follows: p i (j) represents the probability of the grayscale in the i-th frame being greater than 0 and less than j, n t is the number of pixels with gray level less than j, n I is the total number of pixels in each frame; 2.2) Calculate the histogram result G(i) of each frame image in the set C. The calculation formula is as follows: G(i) is the grayscale histogram processing result of the i-th frame, where 0≤i<256, p i (j) represents the probability of the pixel in the i-th frame being greater than 0 and less than j; 2.3) Calculate the pixel equalization result H(v), and equalize the image set C, where C = {I1, I2...I n }, the processed image set is recorded as C′, C′={I1′,I2′...I n '}, the calculation formula is as follows: Where v is the pixel value of a single image I in the image set C, H(v) is the result of equalization of v, G(v) is the histogram processing result of the current v, and G min is the minimum value of histogram processing, G max is the maximum value of the histogram processing, L is the grayscale level, and round represents the rounding of the pixel value result. After all pixels are calculated, a single image I' is obtained, and the set is recorded as C'; 2.4) Calculate the average pixel value a for the images in the image set C′, where C′={I1′,I2′...I n '}, the calculation formula is as follows: Where M is the length pixel value of the image, N is the width pixel value of the image, and I t '(r,c) is the coordinate of the image pixel in the image set C', and t is the number of the selected image; 2.5) Perform missing and filling processing on the images in the image set C′, the missing and filled values ​​are g(i,j), and the data set C' is obtained. The calculation formula is as follows: In the formula, g(i,j) is the missing and filled value, I'(i,j) is the pixel value of the image I' with coordinates (i,j) in the image set C', and Th is the set threshold; 2.6) Divide the dataset C' into training set E and test set T in the ratio m:n.

4. The method for detecting abnormal lighting of night scenes of buildings based on RetinaXNet as claimed in claim 1, characterized in that: In step 4), the UFL function is used to optimize the weights of the RetinaXNet network model. The optimization formula is as follows: In the formula, y is the true label, which takes the value of 0 or 1. is the dynamic adjustment factor parameter of UFL, γ is the rate of adjusting sample weight, and α is the weight parameter.

5. The method for detecting abnormal lighting of night scenes of buildings based on RetinaXNet as claimed in claim 1, characterized in that: The step 5) training the RetinaXNet network model includes the following steps: 5.1) Calculate the accuracy. The calculation formula is as follows: Where, accuracy is the accuracy rate, TP represents the number of positive samples whose network output is in the test set T and whose reference standard is also positive samples; TN represents the number of negative samples whose network output is in the test set T and whose reference standard is also negative samples; FP represents the number of positive samples whose network output is in the test set T and whose reference standard is negative samples; FN represents the number of positive samples whose network output is in the test set T and whose reference standard is positive samples. 5.2) Calculate the recall rate recall, the calculation formula is as follows: In the formula, recall is the recall rate TP, which represents the number of samples that are judged to be positive samples, that is, the number of positive samples; FN represents the number of samples that are judged to be negative samples but are actually positive samples; 5.3) Calculate the F1 value using the following formula: In the formula, F1 is the balance between accuracy and recall, and is the calculation result of the comprehensive accuracy and recall rate; 5.4) Determine whether F1 is less than t. If so, go to step 5.1) and retrain. Otherwise, go to step 6).

6. The method for detecting abnormal lighting of night scenes of buildings based on RetinaXNet as claimed in claim 1, characterized in that: The step 6) of determining whether the night view light is abnormal comprises the following steps: 6.1) Map the network output back to the original image. The coordinate, length and height formulas mapped to the original image are as follows: In the formula, is the length and height of the original image, x, y, W, and H are the coordinates of the upper left corner of the detection frame and the length and height of the detection frame respectively, N ,y N , W N , H N The coordinates, length and height mapped to the original image; 6.2) Anomaly detection, the detection formula is as follows: Wherein, Warn=0 means no abnormality, and Warn=1 means abnormality occurs and the alarm is triggered.

Citation Information

Patent Citations

  • Urban rail transit panoramic monitoring video fault detection method based on depth learning

    CN106709511A

  • Rapid building night scene lighting light fault detection method

    CN112200019A