A method for detecting mounting device data based on deep learning

Abstract

The application relates to the technical field of light supplement lamp fault detection, and discloses a method for detecting mounting equipment data based on deep learning, which comprises the following steps: acquiring an actual current change characteristic curve of a target mounting equipment, comparing the actual current change characteristic curve with a preset standard current change characteristic curve, and obtaining a fault judgment result one; acquiring an actual light supplement image of the target mounting equipment, inputting the actual light supplement image into a trained light supplement lamp fault recognition model, and obtaining a fault judgment result two; judging whether the traffic flow and license plate recognition of each lane are abnormal, and obtaining a fault judgment result three; comprehensively judging the light supplement lamp fault by integrating the current change of the light supplement lamp, the image features during light supplement, and the detection of the traffic flow and license plate recognition of the lane obtained by the camera assisted by the light supplement lamp, so that the accuracy of light supplement lamp fault detection is improved.

Description

Technical Field

[0001] This invention relates to the field of supplementary lighting fault technology, and specifically to a method for detecting data of mounted equipment based on deep learning. Background Technology

[0002] High-definition cameras and traffic auxiliary lights are typically mounted on outdoor road monitoring poles. During the use of these devices, the auxiliary lights are used to enhance image quality, especially in nighttime scenes. When using fast shutter speeds to capture moving objects such as vehicles, the auxiliary lights are crucial to image quality. Typically, a strobe light is used to provide supplemental illumination at the moment of capture, ensuring that vehicles, road surfaces, and traffic signs are clearly imaged in the captured image. If the auxiliary lights malfunction and cannot provide supplemental illumination, the nighttime images captured by the high-definition cameras will have poor image quality due to the fast shutter speed and low ambient light, rendering them unusable as evidence.

[0003] With the widespread use of these mounted devices and the automated operation of the entire system without human supervision, relying on manual troubleshooting of mounted device faults, such as those of supplemental lighting, is time-consuming, labor-intensive, slow in response, and prone to missed detections. Therefore, there is an urgent need for a solution for detecting supplemental lighting faults. Summary of the Invention

[0004] The purpose of this invention is to provide a method for detecting mounted device data based on deep learning, in order to solve the following technical problems:

[0005] How can we provide a more accurate method for detecting faulty supplementary lighting?

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for detecting mounted device data based on deep learning, comprising:

[0008] Obtain the actual current variation characteristic curve of the target mounted device, and compare the actual current variation characteristic curve with the preset standard current variation characteristic curve to obtain fault judgment result one;

[0009] Acquire the actual supplementary lighting image of the target mounted device, and input the actual supplementary lighting image into the trained supplementary lighting fault identification model to obtain fault judgment result two;

[0010] Determine whether the traffic flow and license plate recognition of each lane are abnormal, obtain fault judgment result three, integrate the fault judgment result one, fault judgment result two and fault judgment result three, and obtain the final fault judgment result of the target mounted equipment;

[0011] The target mounting device is a target supplementary light mounted on a traffic monitoring pole in the electronic police system to provide supplementary lighting for the camera, and the supplementary light fault identification model is a trained deep learning model.

[0012] Preferably, the process of obtaining the actual current variation characteristic curve of the target supplementary light is as follows:

[0013] The current signal flowing through the target supplementary light is collected within a preset time period to obtain the current value corresponding to the collection time.

[0014] Based on the current value, a corresponding curve showing the actual current change characteristic of the target supplementary light over time is generated.

[0015] Preferably, the process of comparing the actual current change characteristic curve with a preset standard current change characteristic curve to obtain fault judgment result one is as follows:

[0016] Obtain multiple sets of the actual current variation characteristic curves;

[0017] Calculate the similarity between the multiple sets of actual current change characteristic curves and the preset standard current change characteristic curves within the same time period;

[0018] The similarity scores are arranged from largest to smallest, and the average value of the first preset number of similarity scores is calculated.

[0019] The average similarity Compare with the preset similarity error ranges [SIMlow, SIMup]:

[0020] like Then it is determined that the actual current change characteristic curve matches the standard current change characteristic curve, that is, the fault judgment result is a match.

[0021] like If the actual current change characteristic curve does not match the standard current change characteristic curve, then the fault judgment result is that they do not match.

[0022] Preferably, the process of calculating the similarity between multiple sets of actual current change characteristic curves and preset standard current change characteristic curves within the same time period is as follows:

[0023] The length L of the curves that coincide with the preset standard current change characteristic curves within the same time period is obtained from multiple sets of actual current change characteristic curves. 重合 The number of overlapping pixels N 重合 and the sum of the differences in the areas under the curves ΔS 总 ;

[0024] The similarity SIM:

[0025]

[0026] Among them, L 实际 L is the total length of the actual current variation characteristic curve. 标准 N is the total length of the standard current variation characteristic curve. 标准 S represents the total number of pixels in the standard current variation characteristic curve. 实时 S is the total area under the curve of the actual current variation characteristic curve within the same time period. 标准 σ1, σ2, and σ3 are the total area under the curve of the standard current change characteristic curve within the same time period, and are preset weighting coefficients.

[0027] Preferably, the process of acquiring the actual supplementary lighting image of the target supplementary lighting lamp and inputting the actual supplementary lighting image into the trained supplementary lighting lamp fault recognition model to obtain fault judgment result two is as follows:

[0028] Acquire actual supplemental lighting images of the target supplemental lighting during a preset time period;

[0029] Extract the image features of the actual supplementary lighting image, and input the image features of the actual supplementary lighting image into the trained supplementary lighting fault recognition model to output the corresponding fault judgment result 2;

[0030] The second fault judgment result includes qualified and unqualified.

[0031] Preferably, the process of determining whether the traffic flow and license plate recognition of each lane are abnormal, and obtaining fault judgment result three, is as follows:

[0032] A preset nighttime period is selected as the discrimination period, and the traffic flow T of the lanes within the discrimination period is obtained. ab and license plate recognition rate R ab ;

[0033] Obtain the average traffic flow of the lanes during the same time period as the electronic police system under normal conditions and the discrimination time period. and license plate recognition rate

[0034] Obtain the traffic flow anomaly determination value V and the license plate recognition rate anomaly determination value N, and compare the traffic flow anomaly determination value V and the license plate recognition rate anomaly determination value N with preset anomaly thresholds, that is, compare the traffic flow anomaly determination threshold V0 and the license plate recognition rate anomaly determination threshold N0:

[0035] If the current lane has V≥V0 and N≥N0 within the specified time period, then calculate V and N for the adjacent lanes in the same time period:

[0036] If V < V0 and N < N0 in the same time period for adjacent lanes, it is preliminarily determined that the target supplementary light in the current lane is faulty, that is, the third fault judgment result is that the target supplementary light is faulty;

[0037] If V ≥ V0 and N ≥ N0 in the same time period for adjacent lanes, it is necessary to determine whether the V and N values in the same lane during the daytime are normal:

[0038] If V < V0 and N < N0 during the daytime, it is determined that the target supplementary lights in the current lane and the adjacent lanes are faulty, that is, the third fault judgment result is that the target supplementary light is faulty;

[0039] If V ≥ V0 and N ≥ N0 during the daytime, it is determined that the camera corresponding to the target supplementary light in the current lane may be faulty, that is, the third fault judgment result is that the camera is faulty;

[0040] If V < V0 and N < N0 in the current lane during the discrimination time period, the third fault judgment result is that the target supplementary light is not faulty.

[0041] Preferably, obtain the average traffic flow of the lanes in the same time period as the discrimination time period when the electronic police system is in normal condition and the license plate recognition rate The process is as follows:

[0042] ' Obtain the vehicle passing data of each lane in the X days before when the electronic police system is in normal condition, and take 1 hour as the time period unit to count the vehicle flow and license plate recognition rate of each lane at each time period every day. Then count the data of the previous X days and calculate the average vehicle flow and license plate recognition rate at the same time period every day, that is and

[0043] where the license plate recognition rate is the ratio of the number of vehicles with correctly recognized license plates to the vehicle flow.

[0044] Preferably, the traffic flow anomaly determination value V and the license plate recognition rate anomaly determination value N:

[0045]

[0046]

[0047] where T ab is the vehicle flow of the b-th lane in the a-th time period of the discrimination time period, R ab is the license plate recognition rate of the b-th lane in the a-th time period of the discrimination time period, is the average vehicle flow of the b-th lane in the a-th time period when the electronic police system is in normal condition, σ4 and σ5 are preset weighting coefficients, where σ4 represents the average license plate recognition rate of the b-th lane during time period a when the electronic police system is in normal operation.

[0048] Preferably, the process of integrating the fault judgment result one, fault judgment result two, and fault judgment result three to obtain the final fault judgment result of the target mounted equipment is as follows:

[0049] Determine whether the fault diagnosis result one is normal:

[0050] If the fault judgment result is consistent, the final fault judgment result of the target mounted equipment is normal, and no fault warning is issued.

[0051] If the first fault judgment result is inconsistent, then determine whether the second fault judgment result is normal:

[0052] If the second fault judgment result is qualified, the final fault judgment result of the target mounted equipment is normal, and no fault warning is issued.

[0053] If the second fault judgment result is unqualified, then determine whether the third fault judgment result is normal:

[0054] If the third fault judgment result is that the target supplementary light is not faulty, then the final fault judgment result of the target mounted equipment is normal, and no fault warning is issued;

[0055] If the third fault judgment result is a target supplementary light fault, then the final fault judgment result of the target mounted equipment is supplementary light abnormality, and a fault warning is issued.

[0056] If the third fault judgment result is a camera fault, then the final fault judgment result of the target mounted equipment is a camera abnormality, and a fault warning is issued.

[0057] The beneficial effects of this invention are:

[0058] This deep learning-based method for detecting attached equipment data improves the accuracy of supplementary lighting fault detection by integrating the current changes of the supplementary lighting, image features during supplementary lighting, and lane traffic flow and license plate recognition obtained by the camera assisted by the supplementary lighting to make a comprehensive judgment on supplementary lighting faults. Attached Figure Description

[0059] The invention will now be further described with reference to the accompanying drawings.

[0060] Figure 1 This is a flowchart of the method steps for detecting faults in mounted equipment according to the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Please see Figure 1 As shown, the present invention is a method for detecting mounted device data based on deep learning, comprising:

[0063] Obtain the actual current variation characteristic curve of the target mounted equipment, and compare the actual current variation characteristic curve with the preset standard current variation characteristic curve to obtain the fault judgment result 1.

[0064] Acquire the actual supplementary lighting image of the target mounted device, and input the actual supplementary lighting image into the trained supplementary lighting fault recognition model to obtain fault judgment result two;

[0065] Determine whether the traffic flow and license plate recognition of each lane are abnormal, obtain fault judgment result three, integrate fault judgment result one, fault judgment result two and fault judgment result three to obtain the final fault judgment result of the target mounted equipment;

[0066] Among them, the target mounting device is a target supplementary light mounted on a traffic monitoring pole in the electronic police system to provide supplementary lighting for the camera, and the supplementary light fault identification model is a trained deep learning model.

[0067] The above technical solution first detects the fault of the supplementary light by comparing the actual current change characteristic curve with the preset standard current change characteristic curve.

[0068] Secondly, by inputting actual supplementary lighting images into the trained supplementary lighting fault identification model, the images during supplementary lighting can be monitored. The light source of the supplementary lighting includes multiple matrix-arranged light-emitting units. Faults in different light-emitting units can provide a large number of training samples during the training of the deep learning model, which helps to train the deep learning model efficiently. Therefore, the identification accuracy of the supplementary lighting fault identification model can be guaranteed, and the fault of the supplementary lighting can be detected by identifying the images during supplementary lighting.

[0069] Then, the fault of the auxiliary lights is detected by judging the traffic flow and license plate recognition of each lane;

[0070] By combining the above-mentioned current changes of the supplementary light, image features during supplementary lighting, and detection of lane traffic flow and license plate recognition obtained by the camera assisted by the supplementary light, the accuracy of supplementary light fault detection is improved.

[0071] The deep learning algorithm used in this invention is based on Convolutional Neural Networks (CNNs). CNNs are a type of feedforward neural network that includes convolutional computation and has a deep structure. They are commonly used to analyze visual images. In practical applications, they are often used for image classification and retrieval. Their advantages include the ability to share convolutional kernels, ease of processing high-dimensional data, no need to manually select feature values, and good feature classification results. In this invention, the training samples of the deep learning model are obtained in the same way as the actual supplementary lighting images.

[0072] The process of obtaining the actual current change characteristic curve of the target supplementary light is as follows:

[0073] The current signal flowing through the target supplementary light is collected within a preset time period to obtain the current value corresponding to the collection time.

[0074] The actual current change characteristic curve of the target supplementary light is generated based on the current value over time.

[0075] The standard current variation characteristic curve described above is used to identify the variation characteristics of the standard current flowing through the target supplementary light when it is working normally. Specifically, this curve is plotted based on multiple standard current values ​​flowing through the target supplementary light at different times during normal operation. Therefore, when setting the standard current variation characteristic curve, it is necessary to collect the standard current value of the target supplementary light multiple times. The more times the standard current value is collected, the more accurate the plotted standard current variation characteristic curve will be.

[0076] The process of comparing the actual current change characteristic curve with the preset standard current change characteristic curve to obtain fault judgment result one is as follows:

[0077] Obtain multiple sets of actual current variation characteristic curves;

[0078] Calculate the similarity between multiple sets of actual current change characteristic curves and preset standard current change characteristic curves within the same time period;

[0079] Sort the similarity scores from highest to lowest, and calculate the average of the first preset number of similarity scores.

[0080] Average similarity Compare with the preset similarity error ranges [SIMlow, SIMup]:

[0081] like If the actual current change characteristic curve matches the standard current change characteristic curve, then the fault judgment result is a match.

[0082] like If the actual current change characteristic curve does not match the standard current change characteristic curve, then the fault judgment result is that they do not match.

[0083] Through the above technical solution, multiple sets of actual current change characteristic curves are obtained, and the similarity between the multiple sets of actual current change characteristic curves and the preset standard current change characteristic curves in the same time period is calculated. The similarity is arranged from largest to smallest, and the average value of the preset number of similarity curves at the top of the list is calculated. This allows for the selection of more accurate actual current change characteristic curves, avoiding errors caused by unstable current changes due to factors such as temperature. The preset similarity error range is an acceptable error interval specifically set according to actual needs.

[0084] The process of calculating the similarity between multiple sets of actual current change characteristic curves and preset standard current change characteristic curves within the same time period is as follows:

[0085] The length L of the curves that coincide with the preset standard current change characteristic curves within the same time period is obtained from multiple sets of actual current change characteristic curves. 重合 The number of overlapping pixels N 重合 and the sum of the differences in the areas under the curves ΔS 总 ;

[0086] Similarity SIM:

[0087]

[0088] Among them, L 实际 L is the total length of the actual current variation characteristic curve. 标准 N represents the total length of the standard current variation characteristic curve. 标准 S represents the total number of pixels in the standard current variation characteristic curve. 实时 S represents the total area under the curve of the actual current variation characteristic curve within the same time period. 标准 σ1, σ2, and σ3 are the total area under the curve of the standard current change characteristic curve within the same time period, and are preset weighting coefficients.

[0089] The above technical solution provides a similarity calculation method, using the formula...

[0090] The similarity score (SIM) is used to comprehensively judge the similarity between the actual current change characteristic curve and the preset standard current change characteristic curve. Based on the similarity score, the degree of agreement between the actual current change characteristic curve and the standard current change characteristic curve is determined, thereby identifying the fault of the supplementary light.

[0091] It should be noted that the overlap length L between the actual current change characteristic curve and the preset standard current change characteristic curve is... 重合 The number of overlapping pixels N 重合 and the sum of the differences in the areas under the curves ΔS 总 The curve is obtained by shifting it up and down within the same time period; the preset weighting coefficients σ1, σ2, and σ3 are selectively set based on empirical data.

[0092] The process of acquiring the actual supplementary lighting image of the target supplementary lighting lamp and inputting the actual supplementary lighting image into the trained supplementary lighting lamp fault recognition model to obtain fault judgment result two is as follows:

[0093] Acquire actual supplemental lighting images of the target supplemental lighting during a preset time period;

[0094] Extract the image features of the actual supplementary lighting image, and input the image features of the actual supplementary lighting image into the trained supplementary lighting fault recognition model to output the corresponding fault judgment result 2;

[0095] The second fault diagnosis result includes qualified and unqualified.

[0096] Using the above technical solution, the actual supplementary lighting image is input into the trained supplementary lighting fault recognition model. If the actual supplementary lighting image is identified as a faulty image, the fault judgment result 2 (unqualified) is output; if the actual supplementary lighting image is identified as a normal image, the fault judgment result 2 (qualified) is output.

[0097] It should be noted that the actual supplemental lighting image is obtained by a pre-installed image acquisition device, and this image acquisition device does not affect the normal operation of the electronic police system.

[0098] The process of determining whether the traffic flow and license plate recognition of each lane are abnormal, and obtaining the third fault diagnosis result, is as follows:

[0099] A preset nighttime period is selected as the discrimination period, and the traffic flow T of the lanes within the discrimination period is obtained. ab and license plate recognition rate R ab ;

[0100] Obtain the average traffic flow of the lanes during the same time period when the electronic traffic enforcement system is operating normally and during the judgment period. and license plate recognition rate

[0101] Obtain the abnormal traffic volume determination value V and the abnormal license plate recognition rate determination value N, and compare the abnormal traffic volume determination value V and the abnormal license plate recognition rate determination value N with the preset abnormal situation threshold, that is, compare with the traffic volume abnormal situation determination threshold V0 and the license plate recognition rate abnormal situation determination threshold N0:

[0102] If V≥V0 and N≥N0 in the current lane during the discrimination time period, calculate V and N of the adjacent lane in the same time period:

[0103] If V<V0 and N<N0 in the adjacent lane in the same time period, initially determine that the target supplementary light in the current lane is faulty, that is, the fault judgment result three is that the target supplementary light is faulty;

[0104] If V≥V0 and N≥N0 in the adjacent lane in the same time period, determine whether the V and N values in the same lane during the daytime are normal:

[0105] If V<V0 and N<N0 during the daytime, determine that the target supplementary lights in the current lane and the adjacent lane are faulty, that is, the fault judgment result three is that the target supplementary light is faulty;

[0106] If V≥V0 and N≥N0 during the daytime, determine that the camera corresponding to the target supplementary light in the current lane may be faulty, that is, the fault judgment result three is that the camera is faulty;

[0107] If V<V0 and N<N0 in the current lane during the discrimination time period, the fault judgment result three is that the target supplementary light is not faulty.

[0108] The process of obtaining the average traffic volume and the license plate recognition rate of the lane in the same time period when the electronic police system is in a normal situation is as follows:

[0109] Obtain the vehicle passing data of each lane in the previous X days when the electronic police system is in a normal situation, and take 1 hour as the time period unit to count the vehicle flow and license plate recognition rate of each lane at each time period of each day, and then count the data of the previous X days, and count the average vehicle flow and license plate recognition rate at the same time period of each day, that is and

[0110] Among them, the license plate recognition rate is the ratio of the number of vehicles with correctly recognized license plates to the vehicle flow.

[0111] The above technical solution addresses the issue that a malfunction in the supplementary lighting reduces the brightness of nighttime vehicle images, leading to a rapid decrease in vehicle detection and license plate recognition rates. By analyzing the percentage drop in these rates, it's possible to determine if the supplementary lighting is faulty. Furthermore, by using high-definition cameras from the electronic police system to capture vehicle images and recognize license plates, the system calculates traffic flow and license plate recognition rates within a given time period. These rates are then compared to normal traffic flow and recognition rates for the same time period to determine if vehicle detection and license plate recognition for that lane are abnormal. Finally, a comparison with a preset abnormality threshold is made to determine if the supplementary lighting is malfunctioning, thus enabling the detection of supplementary lighting failures.

[0112] Traffic flow anomaly judgment value V and license plate recognition rate anomaly judgment value N:

[0113]

[0114]

[0115] Among them, T ab To determine the traffic flow of the b-th lane during time period a, R ab To determine the license plate recognition rate of the b-th lane during time period a, This represents the average traffic flow of lane b during time period a when the electronic traffic enforcement system is operating normally. σ4 and σ5 are preset weighting coefficients, representing the average license plate recognition rate of the b-th lane during time period a when the electronic police system is operating normally.

[0116] The above technical solution provides a method for calculating the traffic flow anomaly determination value V and the license plate recognition rate anomaly determination value N, based on... The data is obtained and compared with a preset abnormal situation threshold to determine the fault of the supplementary lighting.

[0117] It should be noted that the camera simultaneously covers the current lane and the adjacent lane, and captures images of vehicles in both lanes. Each lane has a supplementary light. If one of the supplementary lights is not lit or its brightness is reduced, it will affect the vehicle detection rate and license plate recognition rate of that lane at night, thus determining whether there is a problem with the supplementary light. The preset weighting coefficients σ4 and σ5 are selectively set based on empirical data.

[0118] The process of integrating fault diagnosis results one, two, and three to obtain the final fault diagnosis result for the target mounted equipment is as follows:

[0119] Determine if fault diagnosis result one is normal:

[0120] If the fault diagnosis result is consistent, the final fault diagnosis result of the target mounted equipment is normal, and no fault warning is issued.

[0121] If fault diagnosis result one is inconsistent, then determine whether fault diagnosis result two is normal:

[0122] If the second fault judgment result is qualified, the final fault judgment result of the target mounted equipment is normal, and no fault warning is issued.

[0123] If fault diagnosis result two is unqualified, then determine whether fault diagnosis result three is normal:

[0124] If the third fault judgment result is that the target supplementary light is not faulty, then the final fault judgment result of the target mounted equipment is normal, and no fault warning will be issued.

[0125] If the third fault diagnosis result is a fault in the target supplementary lighting, then the final fault diagnosis result for the target mounted equipment is that the supplementary lighting is abnormal, and a fault warning will be issued.

[0126] If the third fault diagnosis result is a camera fault, then the final fault diagnosis result for the target mounted equipment is a camera malfunction, and a fault warning will be issued.

[0127] By integrating the current changes of the supplementary light, the image features during supplementary lighting, and the detection of lane traffic flow and license plate recognition obtained by the camera assisted by the supplementary light, the accuracy of supplementary light fault detection is improved.

[0128] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for detecting mounted device data based on deep learning, characterized in that, include: Obtain the actual current variation characteristic curve of the target mounted device, and compare the actual current variation characteristic curve with the preset standard current variation characteristic curve to obtain fault judgment result one; Acquire the actual supplementary lighting image of the target mounted device, and input the actual supplementary lighting image into the trained supplementary lighting fault identification model to obtain fault judgment result two; Determine whether the traffic flow and license plate recognition of each lane are abnormal, obtain fault judgment result three, integrate the fault judgment result one, fault judgment result two and fault judgment result three, and obtain the final fault judgment result of the target mounted equipment; The target mounting device is a target supplementary light mounted on a traffic monitoring pole in the electronic police system to provide supplementary lighting for the camera, and the supplementary light fault identification model is a trained deep learning model. The process of acquiring the actual supplementary lighting image of the target supplementary lighting lamp and inputting the actual supplementary lighting image into the trained supplementary lighting lamp fault recognition model to obtain fault judgment result two is as follows: Acquire actual supplemental lighting images of the target supplemental lighting during a preset time period; Extract the image features of the actual supplementary lighting image, and input the image features of the actual supplementary lighting image into the trained supplementary lighting fault recognition model to output the corresponding fault judgment result 2; The second fault judgment result includes qualified and unqualified; The process of integrating fault judgment result one, fault judgment result two, and fault judgment result three to obtain the final fault judgment result of the target mounted equipment is as follows: Determine whether the fault diagnosis result one is normal: If the fault judgment result is consistent, the final fault judgment result of the target mounted equipment is normal, and no fault warning is issued. If the first fault judgment result is inconsistent, then determine whether the second fault judgment result is normal: If the second fault judgment result is qualified, the final fault judgment result of the target mounted equipment is normal, and no fault warning is issued. If the second fault judgment result is unqualified, then determine whether the third fault judgment result is normal: If the third fault judgment result is that the target supplementary light is not faulty, then the final fault judgment result of the target mounted equipment is normal, and no fault warning is issued; If the third fault judgment result is a target supplementary light fault, then the final fault judgment result of the target mounted equipment is supplementary light abnormality, and a fault warning is issued. If the third fault judgment result is a camera fault, then the final fault judgment result of the target mounted equipment is a camera abnormality, and a fault warning is issued.

2. The method for detecting mounted device data based on deep learning according to claim 1, characterized in that, The process of obtaining the actual current change characteristic curve of the target supplementary light is as follows: The current signal flowing through the target supplementary light is collected within a preset time period to obtain the current value corresponding to the collection time. Based on the current value, a corresponding curve showing the actual current change characteristic of the target supplementary light over time is generated.

3. The method for detecting mounted device data based on deep learning according to claim 2, characterized in that, The process of comparing the actual current change characteristic curve with the preset standard current change characteristic curve to obtain fault judgment result one is as follows: Obtain multiple sets of the actual current variation characteristic curves; Calculate the similarity between the multiple sets of actual current change characteristic curves and the preset standard current change characteristic curves within the same time period; The similarity scores are arranged from largest to smallest, and the average value of the first preset number of similarity scores is calculated. The average similarity Similarity error range [ , Comparison: like [ , If the actual current change characteristic curve matches the standard current change characteristic curve, then the fault judgment result is a match. like [ , If the actual current change characteristic curve does not match the standard current change characteristic curve, then the fault judgment result is that they do not match.

4. The method for detecting mounted device data based on deep learning according to claim 3, characterized in that, The process of calculating the similarity between multiple sets of actual current change characteristic curves and preset standard current change characteristic curves within the same time period is as follows: Obtain the lengths of the actual current change characteristic curves that coincide with the preset standard current change characteristic curves within the same time period. Number of overlapping pixels and the sum of the differences in the areas under the curves ; The similarity : ； in, The total length of the actual current variation characteristic curve is given. For the , The total number of pixels in the standard current variation characteristic curve. The total area under the curve of the actual current variation characteristic curve within the same time period. for The total area under the curve within the same time period , , These are preset weighting coefficients.

5. The method for detecting mounted device data based on deep learning according to claim 1, characterized in that, The process of determining whether the traffic flow and license plate recognition of each lane are abnormal, and obtaining the third fault diagnosis result, is as follows: A preset nighttime period is selected as the discrimination period, and the traffic flow of the lanes within the discrimination period is obtained. and license plate recognition rate ; Obtain the average traffic flow of the lanes during the same time period as the electronic police system under normal conditions and the discrimination time period. and license plate recognition rate ; Obtain the abnormal traffic flow judgment value V and the abnormal license plate recognition rate judgment value. And the abnormal traffic flow judgment value V and the abnormal license plate recognition rate judgment value are used to determine the abnormal traffic flow value V and the abnormal license plate recognition rate value V. The data is compared with a preset threshold for abnormal conditions, i.e., the threshold for determining abnormal traffic flow conditions. Threshold for judging abnormal license plate recognition rates Compare: If the current lane is judged within the time period, the V and Then calculate V and V of adjacent lanes in the same time period. : If adjacent lanes have V at the same time period and If the target auxiliary light in the current lane is determined to be faulty, then fault judgment result three is that the target auxiliary light is faulty. If adjacent lanes have V at the same time period and Then determine the V, V, of the same lane during the daytime. Are the values ​​normal? If V during the daytime and If the target supplementary lights in the current lane and the adjacent lane are faulty, then the fault judgment result three is that the target supplementary lights are faulty. If V during the daytime and If the camera corresponding to the target supplement light in the current lane is determined to be faulty, then fault judgment result three indicates that the camera is faulty. If the current lane is judged within the time period, the V and If the fault judgment result is three, then the target supplementary light is not faulty.

6. The method for detecting mounted device data based on deep learning according to claim 5, characterized in that, Obtain the average traffic flow of the lanes during the same time period as the electronic police system under normal conditions and the discrimination time period. and license plate recognition rate The process is as follows: The system acquires vehicle passage data for each lane over the previous X days when the electronic police system is operating normally. Then, it calculates the vehicle flow and license plate recognition rate for each lane at each time period, using one-hour intervals. Finally, it calculates the average vehicle flow and license plate recognition rate for the same time period each day, based on the data from the previous X days. and ; The license plate recognition rate is the ratio of the number of vehicles whose license plates are correctly recognized to the traffic flow.

7. The method for detecting mounted device data based on deep learning according to claim 5, characterized in that, The abnormal traffic flow judgment value V and the abnormal license plate recognition rate judgment value : V ； ； in, To determine the traffic flow of the b-th lane during time period a, To determine the license plate recognition rate of the b-th lane during time period a, The average traffic flow of the b-th lane during time period a when the electronic traffic enforcement system is operating normally. The average license plate recognition rate of the b-th lane during time period a when the electronic police system is operating normally. and These are preset weighting coefficients.

Images