Disclosure of Invention
The invention provides a substation equipment state monitoring method and system based on accurate alignment camera shooting, which are used for solving the technical problems that the existing substation equipment state monitoring method is high in cost, is easily influenced by severe weather of GPS positioning deviation planned by a preset air line, is low in stability, is easily influenced by robot positioning and advancing attitude deviation, and is influenced by the fact that images need to be subjected to preset partitioning in advance to influence the positioning accuracy of target equipment.
In view of this, a first aspect of the present invention provides a substation equipment state monitoring method based on precise alignment camera shooting, including:
the method comprises the following steps of building a camera monitoring system for monitoring the state of the transformer substation equipment, wherein the camera monitoring system comprises a three-axis stability-increasing cradle head, a high-definition night vision network camera mounted on the three-axis stability-increasing cradle head and a background processing terminal in communication connection with the high-definition night vision network camera;
the method comprises the steps that a background processing terminal acquires images of transformer substation equipment through a high-definition night vision network camera to construct a first image data set and a second image data set, wherein image samples in the first image data set are image samples which are acquired by the high-definition night vision network camera and contain the transformer substation equipment after image preprocessing, and image samples in the second image data set are image samples obtained by cutting transformer substation equipment areas in the image samples in the first image data set;
the background processing terminal respectively trains a first Efficientdet network model and a second Efficientdet network model by using a first image data set and a second image data set to obtain an equipment body detection Efficientdet model for detecting a target position and an equipment state detection Efficientdet model for detecting a target state;
the background processing terminal controls the high-definition night vision network camera to collect images of preset points according to a preset patrol strategy;
the background processing terminal detects whether target equipment exists in the image of the preset point in the image input equipment body detection Efficientdet model of the preset point, if not, the image of the preset point is collected again, and if yes, the central point of the target equipment is positioned according to the number of the target equipment existing in the image of the preset point;
the background processing terminal calculates the x-axis position adjustment amount and the y-axis position adjustment amount of the target equipment according to the pixels of the central point of the target equipment;
the background processing terminal adjusts the target equipment to the central position of the image of the preset point according to the x-axis position adjustment amount and the y-axis position adjustment amount to obtain a first target image;
the background processing terminal calculates the ratio of the target equipment in the first target image according to the width and the height of the target equipment in the first target image;
the background processing terminal calculates the focal length variation of the camera according to the corresponding relation coefficient of the ratio and the pixel zoom coefficient, adjusts the focal length of the camera according to the focal length variation of the camera and automatically focuses the first target image to obtain a second target image;
and the background processing terminal detects the state of the second target image input equipment in the Efficientdet model to identify the state of the target equipment.
Optionally, after the background processing terminal detects the state of the second target image input device in the Efficientdet model to identify the state of the target device, the method further includes:
and judging whether the target equipment has abnormal state, if so, sending out abnormal state warning information of the target equipment, wherein the abnormal state warning information of the target equipment comprises the name and the state of the target equipment.
Optionally, the detecting, by the background processing terminal, whether the target device exists in the image of the preset point in the image input device body detection Efficientdet model of the preset point, if not, acquiring the image of the preset point again, and if yes, positioning the center point of the target device according to the number of the target devices existing in the image of the preset point, including:
the background processing terminal inputs the image of the preset point into the device body detection Efficientdet model to detect whether target devices exist in the image of the preset point or not, if not, the image of the preset point is collected again, if yes, whether and only one target device exists in the image of the preset point is judged, if yes, the central point of the target device is directly positioned, otherwise, the confidence coefficient of each target device is calculated, when the confidence coefficient is larger than 0.8 and only one target device exists, the central point of the target device is directly positioned, when the confidence coefficient is larger than 0.8 and more than one target device exists, the area of each target device is calculated, and the target device with the largest area is selected to position the central point of the target device.
Optionally, the calculation formula for calculating the distance between the stationing position and the monitored equipment according to the stationing position of the camera and the position of the monitored equipment is as follows:
wherein,
in order to monitor the location of the equipment being monitored,
dthe distance between the stationing position and the monitored equipment.
Optionally, the calculating, by the background processing terminal, the x-axis position adjustment amount and the y-axis position adjustment amount of the target device according to the pixel of the central point of the target device includes:
the background processing terminal calculates the central pixel deviation according to the pixels of the central point of the target equipment, and the calculation formula is as follows:
wherein,
is the x-axis center pixel deviation,
is the y-axis center pixel offset and,
the x-axis pixel that is the center point of the target device,
is the y-axis pixel of the center point of the target device,
the x-axis pixel being the center position of the image of the preset point,
a y-axis pixel being a center position of the image of the preset point;
the background processing terminal performs the processing according to the deviation of the central pixel of the x axis
Calculating the x-axis position adjustment amount of the target equipment according to a first corresponding relation coefficient of the x-axis position adjustment amount, wherein the first corresponding relation coefficient is as follows:
wherein,
the amount of x-axis position adjustment for the target device,
is a coefficient of the first correspondence relationship,
,
is the focal length of the camera during shooting;
the background processing terminal performs processing according to the deviation of the central pixel of the y axis
And calculating the y-axis position adjustment quantity of the target equipment according to a second corresponding relation coefficient of the y-axis position adjustment quantity, wherein the second corresponding relation coefficient is as follows:
wherein,
the amount of y-axis position adjustment for the target device,
in order to be the second correspondence coefficient,
。
optionally, the calculating, by the background processing terminal, a ratio of the target device in the first target image according to the width and the height of the target device in the first target image includes:
the background processing terminal detects a first target image in the Efficientdet model by inputting the first target image into the equipment body, and a detection target is setWidth of the devicewAnd heighth;
Background processing terminal according to widthwAnd heighthCalculating a height ratio and a width ratio, taking the maximum value of the height ratio and the width ratio as the ratio of the target equipment in the image, wherein the calculation formula of the height ratio and the width ratio is as follows:
wherein,
in order to make the ratio of the height to the height,
bin order to have a width ratio,
His the height of the first target image,
Wis the width of the first target image.
Optionally, the calculation formula of the focal length variation of the camera is as follows:
wherein,
is the variation of the focal length of the camera,
as a correspondence coefficient of the duty ratio and the pixel zoom factor,
。
the invention provides a substation equipment state monitoring system based on accurate alignment camera shooting, which comprises a triaxial stability augmentation holder, a high-definition night vision network camera mounted on the triaxial stability augmentation holder and a background processing terminal in communication connection with the high-definition night vision network camera;
the triaxial stability augmentation holder is used for carrying a high-definition night vision network camera and adjusting the visual angle of the high-definition night vision network camera;
the high-definition night vision network camera is used for inspecting and collecting images of the transformer substation equipment;
the background processing terminal is used for:
acquiring an image of the transformer substation equipment acquired by a high-definition night vision network camera, and constructing a first image data set and a second image data set, wherein an image sample in the first image data set is an image sample which is acquired by the high-definition night vision network camera and contains the transformer substation equipment after image preprocessing, and an image sample in the second image data set is an image sample obtained by cutting a transformer substation equipment area in the image sample in the first image data set;
respectively training a first Efficientdet network model and a second Efficientdet network model by using a first image data set and a second image data set to obtain an equipment body detection Efficientdet model for detecting a target position and an equipment state detection Efficientdet model for detecting a target state;
controlling a high-definition night vision network camera to acquire images of preset points according to a preset patrol strategy;
judging whether target equipment exists in the image of the preset point, if not, re-collecting the image of the preset point, and if so, positioning the central point of the target equipment according to the number of the target equipment existing in the image of the preset point;
calculating the x-axis position adjustment amount and the y-axis position adjustment amount of the target equipment according to the pixels of the central point of the target equipment;
adjusting the target equipment to the central position of the image of the preset point according to the x-axis position adjustment amount and the y-axis position adjustment amount to obtain a first target image;
inputting a first target image into an Efficientdet model for detecting the width and the height of target equipment to obtain the proportion of the target equipment in an image;
calculating the focal length variation of the camera according to the corresponding relation coefficient of the ratio and the pixel zoom coefficient, adjusting the focal length of the camera according to the focal length variation of the camera, and automatically focusing the first target image to obtain a second target image;
and inputting the second target image into the state detection Efficientdet model to identify the state of the target equipment.
Optionally, the background processing terminal is further configured to:
after the state of the target equipment is identified in the second target image input equipment state detection Efficientdet model, judging whether the target equipment is abnormal in state or not, if so, sending out abnormal state alarm information of the target equipment, wherein the abnormal state alarm information of the target equipment comprises the name and the state of the target equipment.
Optionally, determining whether target equipment exists in the image of the preset point, if not, re-acquiring the image of the preset point, and if so, positioning a central point of the target equipment according to the number of the target equipment existing in the image of the preset point, including:
judging whether target equipment exists in the image of the preset point, if not, re-acquiring the image of the preset point, if so, judging whether the image of the preset point has only one target equipment, if so, directly positioning the central point of the target equipment, otherwise, calculating the confidence coefficient of each target equipment, directly positioning the central point of the target equipment when the confidence coefficient is greater than 0.8 and only one target equipment exists, and when more than one target equipment with the confidence coefficient greater than 0.8 is available, calculating the area of each target equipment, and taking the target equipment with the largest area to position the central point of the target equipment.
Optionally, calculating an x-axis position adjustment amount and a y-axis position adjustment amount of the target device according to the pixels of the center point of the target device includes:
calculating the central pixel deviation according to the pixels of the central point of the target equipment, wherein the calculation formula is as follows:
wherein,
is the x-axis center pixel deviation,
is the y-axis center pixel offset and,
the x-axis pixel that is the center point of the target device,
the y-axis pixel being the center point of the target device,
the x-axis pixel being the center position of the image of the preset point,
a y-axis pixel being a center position of the image of the preset point;
the background processing terminal performs the processing according to the central pixel deviation of the x axis
Calculating the x-axis position adjustment amount of the target device according to a first corresponding relation coefficient of the x-axis position adjustment amount, wherein the first corresponding relation coefficient is as follows:
wherein,
the amount of x-axis position adjustment for the target device,
is a coefficient of the first correspondence relationship,
,
is the focal length of the camera during shooting;
the background processing terminal performs processing according to the deviation of the central pixel of the y axis
And calculating the y-axis position adjustment quantity of the target equipment according to a second corresponding relation coefficient of the y-axis position adjustment quantity, wherein the second corresponding relation coefficient is as follows:
wherein,
the amount of y-axis position adjustment for the target device,
is a coefficient of the second correspondence relationship,
。
according to the technical scheme, the transformer substation equipment state monitoring method based on accurate alignment camera shooting has the following advantages:
the transformer substation equipment state monitoring method based on accurate alignment camera shooting provided by the invention comprises the steps of constructing two image data sets, respectively training two Efficientdet network models to obtain an equipment body detection Efficientdet model and an equipment state detection Efficientdet model, detecting the position of target equipment under a wide angle condition through the equipment body detection Efficientdet model for carrying out angle adjustment and focal length adjustment on an alignment target, identifying the state of the target equipment through the equipment state detection Efficientdet model, not partitioning a scene image, adapting to a new scene, having good mobility and high accuracy, reducing the cost by using a high-definition night vision network camera fixed by three-axis stabilization compared with a mode of carrying the camera and a patrol robot by an unmanned aerial vehicle, having no deviation problem caused by positioning and traveling postures, solving the problems that the existing transformer substation equipment state monitoring method is high in cost, easy to be influenced by positioning deviation of a preset course and extreme weather such as storm and rain, and the problem that the positioning deviation caused by positioning and traveling postures of a tripod head is easily influenced by the high positioning and the problem of the positioning and traveling posture of a robot.
The substation equipment state monitoring system based on the accurate alignment camera shooting is used for executing the substation equipment state monitoring method based on the accurate alignment camera shooting, the principle and the achieved technical effect of the substation equipment state monitoring system based on the accurate alignment camera shooting are the same as those of the substation equipment state monitoring method based on the accurate alignment camera shooting, and the details are not repeated herein.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
For easy understanding, referring to fig. 1, an embodiment of a substation device status monitoring method based on accurate alignment camera shooting is provided in the present invention, and includes:
101, a camera monitoring system for monitoring the state of the substation equipment is built, wherein the camera monitoring system comprises a triaxial stability-increasing cradle head, a high-definition night vision network camera installed on the triaxial stability-increasing cradle head and a background processing terminal in communication connection with the high-definition night vision network camera.
It should be noted that, in the embodiment of the present invention, a camera monitoring system for monitoring the state of a substation device is first built, where the camera monitoring system includes a triaxial stability enhancement cradle head, a high-definition night-vision network camera mounted on the triaxial stability enhancement cradle head, and a background processing terminal in communication connection with the high-definition night-vision network camera, and the triaxial stability enhancement cradle head carries the high-definition night-vision network camera and supports adjustment of a viewing angle of the high-definition night-vision network camera. The high-definition night vision network camera is arranged at a preset shooting position in the transformer substation through the triaxial stability-increasing cradle head and is used for inspecting and collecting images of transformer substation equipment. The background processing terminal is used for acquiring images acquired by the high-definition night vision network camera and performing preset processing analysis on the images.
102, the background processing terminal acquires images of the transformer substation equipment through the high-definition night vision network camera and constructs a first image data set and a second image data set, wherein image samples in the first image data set are image samples which are acquired by the high-definition night vision network camera and contain the transformer substation equipment after image preprocessing, and image samples in the second image data set are image samples obtained by cutting transformer substation equipment areas in the image samples in the first image data set.
The method includes the steps that after the background processing terminal acquires an image of the substation equipment acquired by the high-definition night vision network camera, a first image data set and a second image data set are constructed, an image sample in the first image data set is an image sample which is acquired by the high-definition night vision network camera and contains the substation equipment after image preprocessing, and an image sample in the second image data set is an image sample obtained by cutting a substation equipment area in the image sample in the first image data set. The high-definition night vision network camera collects images of the transformer substation equipment, after data cleaning, clearly identifiable images with correct focal length are screened out for marking, and the position, name and state of the equipment in the images are marked out to be used as image samples in the first image data set. And cutting out the equipment area in the image sample in the first image data set to obtain a new image sample as the image sample in the second image data set. The image samples in the second image dataset are annotated with device names and states.
103, the background processing terminal respectively trains a first Efficientdet network model and a second Efficientdet network model by using the first image data set and the second image data set to obtain an equipment body detection Efficientdet model for detecting the target position and an equipment state detection Efficientdet model for detecting the target state.
It should be noted that the Efficientdet network model belongs to a deep learning target detection one-stage algorithm, and compared with the two-stage algorithm, the detection speed is faster by adopting the one-stage algorithm, and the identification accuracy effect is good. The image samples in the first image data set and the image samples in the second image data set are divided into 80% of training sets, 10% of testing sets and 10% of verifying sets respectively, the first Efficientdet network model and the second Efficientdet network model are trained respectively, and the equipment body detection Efficientdet model for detecting the target position and the equipment state detection Efficientdet model for detecting the target state can be obtained after training is completed.
Specifically, a first offset network model is trained by using a first image data set, a second offset network model is trained by using a second image data set, the sizes of input pictures of the two models are set to be 600 x 600, the maximum training round is 50 rounds, the test is performed once in each round of training, the learning rate is adjusted in a mode that the existing learning rate is multiplied by 0.1 to be used as a new learning rate when the preset iteration number is reached, and the new learning rate is adjusted after 10 rounds of training and 15 rounds of training respectively. And stopping training when the loss value of the los tends to be flat, and finally obtaining an equipment body detection Efficientdet model and an equipment state detection Efficientdet model.
The trained models are used to perform tests on the test set. The method comprises the steps of testing an equipment body detection Efficientdet model by using a test set divided by a first image data set, testing an equipment state detection Efficientdet model by using a test set divided by a second image data set, analyzing evaluation indexes such as model recall rate and accuracy, carrying out processing modes such as parameter adjustment aiming at the condition of false missing detection and error report in the test set, and carrying out repeated related training until the evaluation indexes achieve a better effect.
Because the target devices to be detected have different sizes and dimensions in different task scenarios, before training an Efficientdet Network model, 9 anchors with different sizes and aspect ratios are clustered according to different training data sets (in a target detection task, an input image is extracted through a backbone Network to obtain a feature map, each pixel point on the feature map is an anchor point, or in an operation mode that a sliding window is used at a feature map, a mapping point of a current sliding window in an original pixel space is called an anchor, and simply speaking, an RPN (Region pro-polar Network) generates 9 target frames with preset and preset areas for each position (i.e. the anchor) by means of a window sliding on a shared feature map, wherein the 9 initial anchors comprise three areas (128 × 128, 256, 512 × 512), and each area comprises three aspect ratios (1, 1 aspect ratio 2, 1), namely.
And step 104, the background processing terminal controls the high-definition night vision network camera to collect images of preset points according to a preset patrol strategy.
It should be noted that the high-definition night vision network camera acquires an image of a preset point according to a preset patrol strategy (for example, rotates at an appointed angle of 30 °), and transmits the acquired image of the preset point to the background processing terminal.
And 105, the background processing terminal inputs the image of the preset point into the equipment body detection Efficientdet model to detect whether target equipment exists in the image of the preset point or not, if not, the image of the preset point is collected again, and if yes, the central point of the target equipment is positioned according to the number of the target equipment existing in the image of the preset point.
It should be noted that the background processing terminal may perform target device identification on the image of the preset point through the device body detection Efficientdet model, as shown in fig. 2, if the target device does not exist in the image of the preset point, the high-definition night vision network camera is controlled to continue to collect the image of the preset point according to the preset patrol strategy, and if the target device exists, the central point of the target device is located according to the number of the target devices existing in the image of the preset point. Specifically, whether there is only one target device in the image of the preset point is judged, if yes, the center point of the target device is directly positioned, otherwise, the confidence coefficient of each target device is calculated, when there is only one target device with the confidence coefficient larger than 0.8, the center point of the target device is directly positioned, and when there is more than one target device with the confidence coefficient larger than 0.8, the area of each target device is calculated, wherein the calculation mode is as follows:
wherein,Sis the area of the target device and,wis the width of the target device and,his the height of the target device.
And positioning the central point of the target equipment by the target equipment with the largest area.
And 106, calculating the x-axis position adjustment amount and the y-axis position adjustment amount of the target equipment by the background processing terminal according to the pixel of the central point of the target equipment.
After determining the target device of the image of the preset point and locating the center point of the target device, the target device needs to be adjusted to the center position of the image of the preset point. Therefore, the x-axis position adjustment amount and the y-axis position adjustment amount of the target device need to be calculated, and the target device is adjusted to the image center position of the preset point according to the x-axis position adjustment amount and the y-axis position adjustment amount of the target device.
Specifically, the computing device body detects the central pixel deviation between the pixel at the central point position of the target device detected by the Efficientdet model and the pixel at the central point of the image of the preset point, and the computing formula is as follows:
wherein,
is the x-axis center pixel deviation,
is the y-axis center pixel deviation,
the x-axis pixel that is the center point of the target device,
the y-axis pixel being the center point of the target device,
the x-axis pixel being the center position of the image of the preset point,
to presetA y-axis pixel of a center position of the image of points;
the background processing terminal performs the processing according to the central pixel deviation of the x axis
Calculating the x-axis position adjustment amount of the target equipment according to a first corresponding relation coefficient of the x-axis position adjustment amount, wherein the first corresponding relation coefficient is as follows:
wherein,
the amount of x-axis position adjustment for the target device,
is a coefficient of the first correspondence relationship,
,
is the focal length of the camera during shooting;
the background processing terminal performs processing according to the deviation of the central pixel of the y axis
And calculating the y-axis position adjustment quantity of the target equipment according to a second corresponding relation coefficient of the y-axis position adjustment quantity, wherein the second corresponding relation coefficient is as follows:
wherein,
the amount of y-axis position adjustment for the target device,
in order to be the second correspondence coefficient,
,
is the focal length of the camera during shooting.
And step 107, the background processing terminal adjusts the target device to the central position of the image of the preset point according to the x-axis position adjustment amount and the y-axis position adjustment amount to obtain a first target image.
It should be noted that, by adjusting the target device to the center position of the image of the preset point according to the x-axis position adjustment amount and the y-axis position adjustment amount, the first target image and the center point position information and the size information (the height and the width of the target device in pixels) of the target device in the first target image can be obtained.
And step 108, the background processing terminal calculates the ratio of the target equipment in the first target image according to the width and the height of the target equipment in the first target image.
It should be noted that, according to the width of the target device in the first target imagewAnd heighthAnd calculating the height ratio and the width ratio, and taking the maximum value of the height ratio and the width ratio as the ratio of the target equipment in the image. The calculation formula of the height ratio and the width ratio is as follows:
wherein,
in order to make the ratio of the height to the height,
bin order to have a width ratio,
His the height of the first target image,
Wis the width of the first target image.
Taking the maximum value of the height ratio and the width ratio as the ratio of the target device in the image, namely:
wherein,cis the occupation ratio of the target device in the image.
And step 109, the background processing terminal calculates the focal length variation of the camera according to the corresponding relation coefficient of the duty ratio and the pixel zoom factor, adjusts the focal length of the camera according to the focal length variation of the camera and automatically focuses the first target image to obtain a second target image.
It should be noted that the occupation ratio of the target device in the image is based on
cCoefficient of correspondence with pixel zoom factor
Calculating the variation of focal length of the camera
. The calculation formula is as follows:
wherein,
is the variation of the focal length of the camera,
as a correspondence coefficient of the duty ratio and the pixel zoom factor,
。
focal length variation of camera
And adjusting the focal length of the camera and automatically focusing the first target image to obtain a clear second target image.
And 110, the background processing terminal detects the state of the second target image input device in the Efficientdet model to identify the state of the target device.
It should be noted that the second target image is input into the equipment state detection Efficientdet model, so that the state information in the target equipment in the second target image can be obtained, and the state monitoring of the target equipment is realized.
In one embodiment, after the state of the target device is identified, whether the target device has an abnormal state may be determined, and if so, abnormal state warning information of the target device is sent out to notify a manager to perform review and maintenance, where the abnormal state warning information of the target device includes the name and the state of the target device.
The substation equipment state monitoring method based on accurate alignment camera shooting provided by the embodiment of the invention comprises the steps of constructing two image data sets, respectively training two Efficientdet network models to obtain an equipment body detection Efficientdet model and an equipment state detection Efficientdet model, detecting the position of target equipment under a wide-angle condition through the equipment body detection Efficientdet model for carrying out angle adjustment and focal length adjustment on an alignment target, identifying the state of the target equipment through the equipment state detection Efficientdet model without zoning a scene image, adapting to a new scene, having good mobility and high accuracy, reducing the cost by using a three-axis augmented stability cradle head fixed night vision network camera compared with a mode of carrying a camera and an unmanned aerial vehicle inspection robot, having high stability, being not easily influenced by GPS positioning deviation planned by a preset air route and extreme weather such as storm, having no deviation problem caused by positioning and traveling posture, solving the problems of the existing substation equipment state monitoring method, being easily influenced by the preset air route positioning deviation planned by high cost, being easily influenced by the positioning deviation planned by the GPS positioning deviation and the GPS positioning deviation, being easily influenced by the GPS positioning deviation and the traveling posture of the preset air and the preset target equipment, and the problem of the robot positioning and the high accuracy of the equipment.
For easy understanding, please refer to fig. 3, an embodiment of a substation equipment state monitoring system based on accurate alignment camera shooting is further provided in the present invention, including a triaxial stability increasing pan-tilt, a high-definition night vision network camera mounted on the triaxial stability increasing pan-tilt, and a background processing terminal in communication connection with the high-definition night vision network camera;
the three-axis stability augmentation holder is used for carrying a high-definition night vision network camera and adjusting the visual angle of the high-definition night vision network camera;
the high-definition night vision network camera is used for inspecting and collecting images of the transformer substation equipment;
the background processing terminal is used for:
acquiring an image of the transformer substation equipment acquired by a high-definition night vision network camera, and constructing a first image data set and a second image data set, wherein an image sample in the first image data set is an image sample which is acquired by the high-definition night vision network camera and contains the transformer substation equipment after image preprocessing, and an image sample in the second image data set is an image sample obtained by cutting a transformer substation equipment area in the image sample in the first image data set;
respectively training a first Efficientdet network model and a second Efficientdet network model by using a first image data set and a second image data set to obtain an equipment body detection Efficientdet model for detecting a target position and an equipment state detection Efficientdet model for detecting a target state;
controlling a high-definition night vision network camera to acquire images of preset points according to a preset patrol strategy;
judging whether target equipment exists in the image of the preset point, if not, re-acquiring the image of the preset point, and if so, positioning the central point of the target equipment according to the number of the target equipment existing in the image of the preset point;
calculating the x-axis position adjustment amount and the y-axis position adjustment amount of the target equipment according to the pixels of the central point of the target equipment;
adjusting the target equipment to the central position of the image of the preset point according to the x-axis position adjustment amount and the y-axis position adjustment amount to obtain a first target image;
inputting a first target image into an Efficientdet model for detecting the width and the height of target equipment to obtain the proportion of the target equipment in an image;
calculating the focal length variation of the camera according to the corresponding relation coefficient of the ratio and the pixel zoom coefficient, adjusting the focal length of the camera according to the focal length variation of the camera, and automatically focusing the first target image to obtain a second target image;
and inputting the second target image into the state detection Efficientdet model to identify the state of the target equipment.
The background processing terminal is further used for:
after the state of the target equipment is identified in the second target image input equipment state detection Efficientdet model, judging whether the target equipment is abnormal in state or not, if so, sending out abnormal state alarm information of the target equipment, wherein the abnormal state alarm information of the target equipment comprises the name and the state of the target equipment.
Judging whether target equipment exists in the image of the preset point, if not, re-acquiring the image of the preset point, and if so, positioning a central point of the target equipment according to the number of the target equipment existing in the image of the preset point, wherein the method comprises the following steps:
judging whether target equipment exists in the image of the preset point, if not, re-acquiring the image of the preset point, if so, judging whether the image of the preset point has only one target equipment, if so, directly positioning the central point of the target equipment, otherwise, calculating the confidence coefficient of each target equipment, directly positioning the central point of the target equipment when the confidence coefficient is greater than 0.8 and only one target equipment exists, and when more than one target equipment with the confidence coefficient greater than 0.8 is available, calculating the area of each target equipment, and taking the target equipment with the largest area to position the central point of the target equipment.
Calculating the x-axis position adjustment amount and the y-axis position adjustment amount of the target equipment according to the pixels of the central point of the target equipment, wherein the method comprises the following steps:
calculating the central pixel deviation according to the pixel of the central point of the target equipment, wherein the calculation formula is as follows:
wherein,
is the x-axis center pixel deviation,
is the y-axis center pixel deviation,
the x-axis pixel that is the center point of the target device,
is the y-axis pixel of the center point of the target device,
the x-axis pixel being the center position of the image of the preset point,
a y-axis pixel being a center position of the image of the preset point;
the background processing terminal performs the processing according to the central pixel deviation of the x axis
Calculating the x-axis position adjustment amount of the target device according to a first corresponding relation coefficient of the x-axis position adjustment amount, wherein the first corresponding relation coefficient is as follows:
wherein,
the amount of x-axis position adjustment for the target device,
is a coefficient of the first correspondence relationship,
,
is the focal length of the camera during shooting;
the background processing terminal performs processing according to the deviation of the central pixel of the y axis
And calculating the y-axis position adjustment quantity of the target equipment according to a second corresponding relation coefficient of the y-axis position adjustment quantity, wherein the second corresponding relation coefficient is as follows:
wherein,
the amount of y-axis position adjustment for the target device,
is a coefficient of the second correspondence relationship,
,
is the focal length of the camera during shooting.
Calculating the occupation ratio of the target device in the first target image according to the width and the height of the target device in the first target image, and the method comprises the following steps:
the background processing terminal inputs the first target image into the equipment body detection Efficientdet model to detect the width of the target equipmentwAnd heighth;
Background processing terminal according to widthwAnd heighthCalculating a height ratio and a width ratio, taking the maximum value of the height ratio and the width ratio as the ratio of the target equipment in the image, wherein the calculation formula of the height ratio and the width ratio is as follows:
wherein,
in order to make the ratio of the height to the height,
bin order to have a width ratio,
His the height of the first target image,
Wis the width of the first target image.
The calculation formula of the focal length variation of the camera is as follows:
wherein,
is the variation of the focal length of the camera,
as a correspondence coefficient of the duty ratio and the pixel zoom factor,
。
the substation equipment state monitoring system based on accurate alignment camera shooting provided by the embodiment of the invention constructs two image data sets, respectively trains two Efficientdet network models to obtain an equipment body detection Efficientdet model and an equipment state detection Efficientdet model, detects the position of target equipment under a wide angle condition through the equipment body detection Efficientdet model for carrying out angle adjustment and focal length adjustment on an alignment target, identifies the state of the target equipment through the equipment state detection Efficientdet model, does not need to partition a scene image, can adapt to a new scene, has good mobility and high accuracy, reduces the cost by using a three-axis stability-increasing fixed high-definition network camera compared with a mode that an unmanned aerial vehicle carries the camera and a patrol robot, does not have the problem of deviation caused by the positioning and the traveling posture, solves the problems that the existing substation equipment state monitoring method is easily influenced by high cost, is easily influenced by the weather positioning deviation planned by a high air route and a severe storm rain, and the problem that the positioning deviation caused by a positioning and the GPS positioning posture are easily influenced by a high positioning technology of a tripod head in advance.
The substation equipment state monitoring system based on the accurate alignment camera shooting is used for executing the substation equipment state monitoring method based on the accurate alignment camera shooting, the principle and the achieved technical effect of the substation equipment state monitoring system based on the accurate alignment camera shooting are the same as those of the substation equipment state monitoring method based on the accurate alignment camera shooting, and the details are not repeated herein.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.