WO2022121130A1 - 电力目标检测方法、装置、计算机设备和存储介质 - Google Patents

电力目标检测方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2022121130A1
WO2022121130A1 PCT/CN2021/079280 CN2021079280W WO2022121130A1 WO 2022121130 A1 WO2022121130 A1 WO 2022121130A1 CN 2021079280 W CN2021079280 W CN 2021079280W WO 2022121130 A1 WO2022121130 A1 WO 2022121130A1
Authority
WO
WIPO (PCT)
Prior art keywords
power
target
image
identified
power target
Prior art date
Application number
PCT/CN2021/079280
Other languages
English (en)
French (fr)
Inventor
张豪
陈满
彭煜民
卢勇
刘涛
李建辉
岳鹏超
韩吉双
吕志鹏
林恺
韩玉麟
巩宇
郭海峰
王晓翼
王翰龙
Original Assignee
南方电网调峰调频发电有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南方电网调峰调频发电有限公司 filed Critical 南方电网调峰调频发电有限公司
Publication of WO2022121130A1 publication Critical patent/WO2022121130A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Definitions

  • the present application relates to the field of computer technology, and in particular, to a method, apparatus, computer device and storage medium for detecting a power target.
  • the power targets in the power production equipment area are often monitored, for example, the power targets in the collected monitoring images of the power production equipment area are identified and detected.
  • a power target detection method comprising:
  • the power monitoring image includes a power target to be identified
  • the image acquisition device is controlled at the second zoom factor to photograph the power target to be identified, and an image of the power target is obtained; in the power target image, the power target to be identified is located in the central main image area ; the second zoom factor is greater than the first zoom factor;
  • a target classification result of the power target to be identified is determined.
  • the method further includes:
  • the method before the step of detecting whether there is a human body in the power monitoring image, the method further includes:
  • the detecting whether there is a human body in the power monitoring image includes:
  • the pedestrian detection result includes at least one of presence of pedestrians and absence of pedestrians;
  • the pedestrian detection result is that there is a pedestrian and the change degree of the image content is greater than a preset threshold, it is determined that there is a human body in the power monitoring image.
  • the determining of the key position points of the power target to be identified in the power monitoring image further includes:
  • the key position points of the to-be-identified power target in the power monitoring image are determined.
  • the method further includes:
  • the step of controlling the image acquisition device to photograph the to-be-identified power target at a second zoom factor according to the key position points, and obtaining an image of the power target;
  • identifying the target classification result of the to-be-identified power target according to the power target image includes:
  • the power target image is input into a pre-trained target classification network to generate a target classification result of the to-be-identified power target.
  • a power target detection device the device includes:
  • an acquisition module configured to acquire a power monitoring image captured by the image acquisition device at a first zoom factor; the power monitoring image includes a power target to be identified;
  • a determination module configured to determine the key position points of the to-be-identified power target in the power monitoring image
  • control module configured to control the image acquisition device to photograph the to-be-identified power target at a second zoom factor according to the key position point to obtain a power target image; in the power target image, the to-be-identified power target in the central main image area; the second zoom factor is greater than the first zoom factor;
  • the identification module is configured to determine the target classification result of the power target to be identified according to the power target image.
  • a computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when the processor executes the computer program.
  • the above power target detection method, device, computer equipment and storage medium by acquiring the power monitoring image including the power target to be identified and captured by the image acquisition device at the first zoom factor, and determining the power target to be identified in the power monitoring image. key position point; then, according to the key position point, the image acquisition device is controlled at the second zoom factor to shoot the power target to be identified, and an image of the power target is obtained; in the power target image, the power target to be identified is in the central main image area; the second zoom The multiple is greater than the first zoom multiple; finally, the target classification result of the power target to be identified is determined according to the power target image; in this way, after the power monitoring image identifies the key position points of the power target to be identified in the power monitoring image, and based on the power monitoring image At this key position, the image acquisition device is controlled at the second zoom factor to shoot the power target to be identified, and a high-definition power target image is obtained, so that the power target image can contain as many features of the power target to be identified as possible, and then can
  • 1 is an application environment diagram of a power target detection method in one embodiment
  • FIG. 2 is a schematic flowchart of a method for detecting a power target in an embodiment
  • Fig. 3 is the training flow chart of a kind of key point detection network in one embodiment
  • FIG. 4 is a schematic diagram of the network structure of a key point detection network in one embodiment
  • FIG. 5 is a schematic structural diagram of a double downsampling module in one embodiment
  • FIG. 6a is a display diagram of a component target prediction process in one embodiment
  • FIG. 6b is a diagram showing an animal target prediction process in one embodiment
  • FIG. 7 is a schematic flowchart of a method for detecting a power target in another embodiment
  • FIG. 8 is a block diagram of a processing flow of a method for detecting a power target in one embodiment
  • FIG. 9 is a structural block diagram of a power target detection device in one embodiment.
  • Figure 10 is a diagram of the internal structure of a computer device in one embodiment.
  • the power target detection method provided in this application can be applied to the application environment shown in FIG. 1 .
  • the computer device 110 communicates with the image acquisition device 120 through a network.
  • the computer device 110 acquires the power monitoring image captured by the image acquisition device 120 at the first zoom factor; the power monitoring image includes the power target to be identified.
  • the computer device 110 determines the key position points of the power target to be identified in the power monitoring image.
  • the computer device 110 controls the image acquisition device to photograph the power target to be identified at the second zoom factor according to the key position points, and obtains an image of the power target; in the power target image, the power target to be identified is in the central main image area; the second zoom The magnification is greater than the first zoom magnification.
  • the computer device 110 determines the target classification result of the power target to be identified according to the power target image.
  • the computer device 110 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices.
  • a method for detecting a power target is provided, which is described by taking the method applied to the computer device 110 in FIG. 1 as an example, including the following steps:
  • Step S210 acquiring a power monitoring image captured by the image acquisition device at a first zoom factor; the power monitoring image includes a power target to be identified.
  • the power monitoring image may refer to an image collected when monitoring the power production area. Practical applications,
  • the power target may refer to various targets that need to be detected in power production. For example, in the production area of a power plant, it is necessary to detect small objects such as parts of equipment (screws, nuts, etc.) and small animals (snakes, mice, cats, etc.).
  • the image acquisition device may refer to a camera.
  • the camera has a zoom function.
  • digital zoom function optical zoom function.
  • the computer device acquires the power monitoring image captured by the image acquisition device at the first zoom factor.
  • Step S220 determining the key position points of the power target to be identified in the power monitoring image.
  • the computer device determines the key position points of the power target to be identified in the power monitoring image.
  • the computer equipment can input the power monitoring images to the pretrained keypoint recognition model.
  • the key position points of the power target to be identified in the power monitoring image are identified through the pre-trained key point identification model.
  • Step S230 according to the key position point, the image acquisition device is controlled to photograph the power target to be identified at the second zoom factor, and an image of the power target is obtained.
  • the power target to be identified is located in the central main image area.
  • the second zoom factor is greater than the first zoom factor.
  • the computer device controls the image acquisition device to photograph the power target to be identified at the second zoom factor to obtain the power target image.
  • the computer device sends the key position point to the image acquisition device, and then controls the image acquisition device to focus and zoom in, and shoots the to-be-identified power target so that the to-be-identified power target is in the central main image area, so as to obtain the power target image .
  • Step S240 Determine the target classification result of the power target to be identified according to the power target image.
  • the computer device determines the target classification result of the power target to be identified according to the power target image.
  • the computer equipment can input the power target image into a pre-trained target classification network, identify the power target image through the pre-trained target classification network, and obtain a target classification result for the power target to be identified.
  • the pre-trained target classification network can adopt the MobileNetV2 structure, in which the target classification network can use data augmentation to enhance the accuracy of multi-scale target recognition, that is, during training, by adjusting the size of the target to be recognized in the screen.
  • the proportion of data is enhanced to reduce the influence of the size of the object in the target frame on the accuracy of target recognition.
  • the power monitoring image including the power target to be identified and captured by the image acquisition device at the first zoom factor is obtained, and the key position points of the power target to be identified in the power monitoring image are determined;
  • the image acquisition device is controlled at the second zoom factor to shoot the power target to be identified, and an image of the power target is obtained; in the power target image, the power target to be identified is in the central main image area; the second zoom ratio is greater than the first zoom ratio;
  • the target classification result of the power target to be identified is determined according to the power target image; in this way, after the power monitoring image identifies the key position point of the power target to be identified in the power monitoring image, and based on the key position point, the image can be controlled
  • the acquisition device shoots the power target to be identified at the second zoom factor, and obtains a high-definition power target image, so that the power target image can contain as many features of the power target to be identified as possible, and then can accurately treat the power target based on the power target image.
  • the method further includes: detecting whether there is a human body in the power monitoring image; if there is no human body in the power monitoring image, Then execute the step of determining the key position points of the power target to be identified in the power monitoring image; if there is a human body in the power monitoring image, return to the step of acquiring the power monitoring image captured by the image acquisition device at the first zoom factor.
  • the computer device after the computer device acquires the power monitoring image captured by the image acquisition device at the first zoom factor, the computer device also needs to detect whether there is a human body in the power monitoring image; if there is no human body in the power monitoring image, then Triggering the subsequent target detection network is to perform the step of determining the key position points of the power target to be identified in the power monitoring image; if there is a human body in the power monitoring image, the subsequent target detection network is not triggered, and the image acquisition device is returned to obtain the first The step of the power monitoring image captured at the zoom factor is to input the next image frame sequence.
  • the detection algorithm can be pre-positioned with a trigger mechanism, and the detection algorithm is triggered for detection when certain conditions are met, which removes the interference of human targets, reduces the amount of calculation and saves energy. consumption.
  • the method before the step of detecting whether there is a human body in the power monitoring image, the method further includes: acquiring a difference between the power monitoring image and a preset historical background image; if the difference is greater than a preset difference threshold, returning to detecting Steps to monitor the presence or absence of a human body in an image.
  • detecting whether there is a human body in the power monitoring image includes: inputting the power monitoring image into a pre-trained pedestrian detection network to obtain a pedestrian detection result; the pedestrian detection result includes at least one of the presence of pedestrians and the absence of pedestrians If the pedestrian detection result is that there is a pedestrian, it is determined that there is a human body in the power monitoring image.
  • the computer device before the computer device detects whether there is a human body in the power monitoring image, the computer device specifically includes: the computer device can obtain the difference between the power monitoring image and the preset historical background image; if the difference is greater than the preset difference threshold, triggering again It is determined whether there is a human body in the power monitoring image.
  • the computer equipment can use the mixed Gaussian model for multiple frames of power monitoring images to establish a real-time model for the background of the moving target, and then take two consecutive frames of images for difference, so as to obtain the contour of the moving target. If the contour of the moving target exceeds the preset value When the target threshold, that is, the change degree of the image content is greater than the preset threshold, it is determined that there is a human body in the power monitoring image.
  • mixed Gaussian background modeling is a background representation method based on the statistical information of pixel samples, which uses statistical information such as the probability density of a large number of sample values of pixels over a long period of time (such as the number of patterns, the mean and standard deviation of each pattern) to represent the background , and then use statistical difference (such as the 3 ⁇ principle) to judge the target pixel, which can model the complex dynamic background.
  • the color information between pixels is considered to be irrelevant, and the processing of each pixel is mutually independant.
  • the change of its value in the sequence image can be regarded as a random process that continuously generates pixel values, that is, the Gaussian distribution is used to describe the color rendering law of each pixel.
  • Each new pixel value X t is compared with the current K models according to formula (1), until a distribution model matching the new pixel value is found, that is, the mean deviation from the model is within 2.5 ⁇ ;
  • X t is the RGB component of the pixel X at time t
  • ⁇ i, t-1 is the standard deviation of the ith Gaussian distribution at time t-1.
  • the pixel belongs to the background, otherwise it belongs to the foreground.
  • ⁇ represents the learning rate
  • w k,t and w k,t-1 are the weights.
  • ⁇ t (1- ⁇ )* ⁇ t-1 + ⁇ *X t (4);
  • ⁇ represents the learning rate and ⁇ is the parameter learning rate.
  • the pattern with the smallest weight is replaced, that is, the mean of the pattern is the current pixel value, the standard deviation is the initial larger value, and the weight is the smaller value;
  • the patterns are arranged in descending order according to w/ ⁇ 2 , and the patterns with large weight and small standard deviation are arranged in the front;
  • the first B modes are selected as the background, and the calculation of B is as in formula (6).
  • T is the weight threshold, and the value range is [0.5, 1].
  • the best distribution describing the background can be selected through the setting of T.
  • the image difference method is a simple and effective method to detect whether there is an abnormality in the current image by subtracting the currently acquired image and the background historical image. By subtracting images, the speed of image processing can be improved.
  • the present invention adopts the Gaussian background modeling algorithm to first perform background modeling on the input image, and then differentiate, so that it is more suitable for practical application scenarios.
  • the formula for calculating the difference is as follows:
  • F i represents the result of the image difference, where two frames of pictures are taken to make a difference for the corresponding pixels, the absolute value is taken, and then the average value of all the pixel values after the difference is taken, F i is the current frame image, F b represents The background image obtained by the mixture Gaussian background modeling algorithm.
  • the computer device determines that there is a human body in the power monitoring image according to the result of the image difference.
  • ⁇ F i represents the result of image difference
  • T is a threshold obtained according to the experiment
  • ⁇ F i is less than the threshold, it means that the difference between the current picture and the background is low
  • the flag bit flag1 is set to False
  • ⁇ F i is greater than or equal to the threshold value
  • the flag bit flag1 is set to True. In this way, it can be determined based on the flag bit flag1 that there is a human body in the power monitoring image.
  • the computer equipment inputs the power monitoring image to a pre-trained pedestrian detection network to obtain a pedestrian detection result; the pedestrian detection result includes at least one of the presence of pedestrians and the absence of pedestrians.
  • the pretrained pedestrian detection network can be the open source RFBNet300 network. If the pedestrian detection result output by the pre-trained pedestrian detection network is that there is a pedestrian, the flag bit flag2 is set to False. If the pedestrian detection result output by the pre-trained pedestrian detection network is that there is no pedestrian, the flag bit flag2 is set to True.
  • the computer device performs a summation operation on the flag bit flag1 and the flag bit flag2, and if the summation result is True (true), the step of determining the key position point of the power target to be identified in the power monitoring image is executed .
  • the technical solution of this embodiment is to obtain the difference between the power monitoring image and the preset historical background image; if the difference is greater than the preset difference threshold, return to the step of detecting whether there is a human body in the power monitoring image; When the difference between the output power monitoring image and the preset historical background image is greater than the preset difference threshold, the subsequent target detection processing is further performed, which reduces the redundant calculation amount and saves the power consumption.
  • determining the key position points of the power target to be identified in the power monitoring image includes: inputting the power monitoring image into a pre-trained key point detection network to obtain a heat map of key points for the power target to be identified; The key point heat map determines the key position points of the power target to be identified in the power monitoring image.
  • the computer equipment specifically includes: the computer equipment inputs the power monitoring image into a pre-trained key point detection network, and obtains a target for the power target to be identified.
  • the key point heat map of according to the key point heat map, determine the key position points of the power target to be identified in the power monitoring image.
  • Figure 3 exemplarily provides a training flow chart of a key point detection network; as shown in Figure 3, it is necessary to perform data preprocessing on the input samples, and then perform the forward operation of the neural network, calculate the loss function, and backpropagate Update the network parameters of the keypoint detection network.
  • the training data is the picture marked with the target frame.
  • the data is first enhanced, including random flip, random color change, random contrast change, random brightness change, random scale change, etc. Then standardize, subtract the mean and divide by the variance. Finally, a heat map of the target is generated, that is, a two-dimensional Gaussian distribution with different variances is generated at the center of each target according to the size of the target box.
  • Neural network forward operation input the processed data into the network to get the output, including the predicted heat map, and the offset of each prediction center relative to the target frame.
  • the loss function calculates the loss of the output heat map and the predicted heat map, use the Focal loss loss function (focus loss function), and calculate the loss of the predicted target frame offset, use the smooth L1 loss (smoothed L1 loss), the loss function is as follows :
  • FIG. 4 provides a schematic diagram of the network structure of a key point detection network; in which, the concatenate operation is to connect the feature maps by channel, and the bilinear interpolation algorithm is often used for double upsampling.
  • the key point detection network includes a double downsampling module.
  • FIG. 5 provides a schematic structural diagram of a double downsampling module.
  • Post-processing (1) perform threshold operation on the heat map, set the minimum threshold to A, then the response value is lower than the position of A; (2) perform the maximum pooling operation to extract the response extreme point; (3) take the extreme point The predicted offset of the position yields the target bounding box.
  • FIG. 6a provides a display diagram of a component target prediction process
  • FIG. 6b provides a display diagram of an animal target prediction process.
  • the method further includes: generating a target frame surrounding the power target to be identified in the power monitoring image according to the key position points; If the area of the target frame is within the preset area range, execute the step of controlling the image acquisition device at the second zoom factor to capture the power target to be identified according to the key position points, and obtain the power target image; if the area of the target frame is not within the preset area range, return to the step of acquiring the power monitoring image captured when the image capturing device is at the first zoom factor.
  • the method further includes: the computer device generates a target frame surrounding the power target to be identified in the power monitoring image according to the key position points in the power monitoring image Then, the computer equipment judges whether the area S of the target frame is in the preset area range; If the computer equipment determines that the area S of the target frame is in the preset area range, S satisfies the following formula:
  • S 1 and S 2 are the set area size limiting parameters.
  • secondary identification that is, to control the image acquisition device to be in the second zoom factor to shoot according to the key position points.
  • a target frame surrounding the power target to be identified is generated in the power monitoring image according to the key position points; if the area of the target frame is within the preset area range, the image acquisition device is controlled according to the key position points.
  • a method for detecting a power target is provided, and the method is applied to the computer device 110 in FIG. 1 as an example for illustration, including the following steps: Step S710 , acquiring an image acquisition device The power monitoring image captured at the first zoom factor; the power monitoring image includes the power target to be identified. Step S720, acquiring the difference between the power monitoring image and a preset historical background image. Step S730, if the difference is greater than a preset difference threshold, detect whether there is a human body in the power monitoring image. Step S740, if there is no human body in the power monitoring image, determine the key position points of the power target to be identified in the power monitoring image.
  • Step S750 control the image acquisition device to photograph the power target to be identified at a second zoom factor, and obtain a power target image; in the power target image, the power target to be identified is in the center the main image area; the second zoom factor is greater than the first zoom factor.
  • Step S760 Determine a target classification result of the power target to be identified according to the power target image.
  • FIG. 8 provides a block diagram of the processing flow of a power target detection method; wherein, firstly input the image frame sequence; the computer equipment judges whether the detection requirements are met; if the detection requirements are met, the image frame sequence is input Then, the computer equipment further judges whether secondary recognition is required; if secondary recognition is required, the secondary recognition is performed; the recognition result is output, and the input next image frame sequence is returned.
  • a power target detection device including:
  • an acquisition module 910 configured to acquire a power monitoring image captured when the image acquisition device is at a first zoom factor; the power monitoring image includes a power target to be identified;
  • a determination module 920 configured to determine the key position points of the power target to be identified in the power monitoring image
  • the control module 930 is configured to control the image acquisition device to photograph the to-be-identified power target at a second zoom factor according to the key position point, to obtain a power target image; in the power target image, the to-be-identified power target The target is in the central main image area; the second zoom factor is greater than the first zoom factor;
  • the identification module 940 is configured to determine a target classification result of the power target to be identified according to the power target image.
  • the power target detection device further includes: a human body detection module, configured to detect whether there is a human body in the power monitoring image; if there is no human body in the power monitoring image, perform the determination The step of the key position points of the power target to be identified in the power monitoring image; if there is a human body in the power monitoring image, returning the power monitoring image captured by the acquired image acquisition device at the first zoom factor A step of.
  • a human body detection module configured to detect whether there is a human body in the power monitoring image; if there is no human body in the power monitoring image, perform the determination The step of the key position points of the power target to be identified in the power monitoring image; if there is a human body in the power monitoring image, returning the power monitoring image captured by the acquired image acquisition device at the first zoom factor A step of.
  • the power target detection device further includes: a difference acquisition module, configured to acquire the difference between the power monitoring image and a preset historical background image; if the difference is greater than a preset difference threshold , then return to the step of detecting whether there is a human body in the power monitoring image.
  • a difference acquisition module configured to acquire the difference between the power monitoring image and a preset historical background image; if the difference is greater than a preset difference threshold , then return to the step of detecting whether there is a human body in the power monitoring image.
  • the human body detection module is specifically configured to input the power monitoring image into a pre-trained pedestrian detection network to obtain a pedestrian detection result; the pedestrian detection result includes the presence of pedestrians and the absence of pedestrians. At least one; if the pedestrian detection result is that there is a pedestrian and the change degree of the image content is greater than a preset threshold, it is determined that there is a human body in the power monitoring image.
  • the determining module 920 is specifically configured to input the power monitoring image into a pre-trained key point detection network to obtain a heat map of key points for the power target to be identified; according to the key points The heat map determines the key position points of the power target to be identified in the power monitoring image.
  • the power target detection device further includes: a judgment module, configured to generate a target frame surrounding the to-be-identified power target in the power monitoring image according to the key position points; If the area of the target frame is within the preset area range, then execute the step of controlling the image acquisition device to photograph the to-be-identified power target at the second zoom factor according to the key position points, and obtain an image of the power target; If the area of the target frame is not within the preset area range, return to the step of acquiring the power monitoring image captured when the image capturing device is at the first zoom factor.
  • a judgment module configured to generate a target frame surrounding the to-be-identified power target in the power monitoring image according to the key position points
  • the identification module 940 is specifically configured to input the power target image into a pre-trained target classification network to generate a target classification result of the power target to be identified.
  • All or part of the modules in the above-mentioned power target detection device can be implemented by software, hardware and combinations thereof.
  • the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
  • a computer device is provided, and the computer device may be a terminal, and its internal structure diagram may be as shown in FIG. 10 .
  • the computer equipment includes a processor, memory, a communication interface, a display screen, and an input device connected by a system bus.
  • the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium, an internal memory.
  • the nonvolatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium.
  • the communication interface of the computer device is used for wired or wireless communication with an external terminal, and the wireless communication can be realized by WIFI, operator network, NFC (Near Field Communication) or other technologies.
  • the computer program when executed by a processor, implements a method for detecting an electrical target.
  • the display screen of the computer equipment may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment may be a touch layer covered on the display screen, or a button, a trackball or a touchpad set on the shell of the computer equipment , or an external keyboard, trackpad, or mouse.
  • FIG. 10 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied. Include more or fewer components than shown in the figures, or combine certain components, or have a different arrangement of components.
  • a computer device including a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above-mentioned method for detecting a power target.
  • the steps of an electric power target detection method here may be the steps in the electric power target detection method of each of the above embodiments.
  • a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the processor causes the processor to perform the steps of the above-mentioned method for detecting a power target.
  • the steps of an electric power target detection method here may be the steps in the electric power target detection method of each of the above embodiments.
  • Non-volatile memory may include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, or optical memory, and the like.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • the RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

一种电力目标检测方法、装置、计算机设备和存储介质。所述方法包括：获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像；所述电力监控图像包括有待识别电力目标(S210)；确定所述待识别电力目标在所述电力监控图像中的关键位置点(S220)；根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像；在所述电力目标图像中，所述待识别电力目标处于中心主图区；所述第二变焦倍数大于所述第一变焦倍数(S230)；根据所述电力目标图像，确定所述待识别电力目标的目标分类结果(S240)。采用本方法能够提高电力目标检测准确度。

Description

电力目标检测方法、装置、计算机设备和存储介质 技术领域
本申请涉及计算机技术领域，特别是涉及一种电力目标检测方法、装置、计算机设备和存储介质。
背景技术
在工业界中为了确保生产过程的顺利，往往会对电力生产设备区域中的电力目标进行监测，如对采集到的电力生产设备区域的监控图像中的电力目标进行识别并检测。
然而，由于电力目标的自身真实的物理尺寸过小或与拍摄设备距离较远，这往往会导致电力目标在监控图像中占比小，使得电力目标通常在监控图像只含有几十个或更少的像素，这也使得现有技术无法准确地对监控图像中的电力目标进行分类检测。
发明内容
基于此，有必要针对上述技术问题，提供一种能够提高电力目标检测准确度的电力目标检测方法、装置、计算机设备和存储介质。
一种电力目标检测方法，包括：
获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像；所述电力监控图像包括有待识别电力目标；
确定所述待识别电力目标在所述电力监控图像中的关键位置点；
根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像；在所述电力目标图像中，所述待识别电力目标处于中心主图区；所述第二变焦倍数大于所述第一变焦倍数；
根据所述电力目标图像，确定所述待识别电力目标的目标分类结果。
在其中一个实施例中，在所述获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤之后，所述方法还包括：
检测所述电力监控图像中是否存在人体；
若所述电力监控图像中不存在人体，则执行所述确定所述待识别电力目标在所述电力监控图像中的关键位置点的步骤；
若所述电力监控图像中存在人体，则返回所述获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
在其中一个实施例中，在所述检测所述电力监控图像中是否存在人体的步骤之前，还包括：
获取所述电力监控图像与预设的历史背景图像之间的差异；
若所述差异大于预设差异阈值，则返回所述检测所述电力监控图像中是否存在人体的步骤。
在其中一个实施例中，所述检测所述电力监控图像中是否存在人体，包括：
将所述电力监控图像输入至预训练的行人检测网络，得到行人检测结果；所述行人检测结果包括存在行人和不存在行人中的至少一种；
若所述行人检测结果为存在行人且所述图像内容变化程度大于预设阈值，则判定所述电力监控图像中存在人体。
在其中一个实施例中，所述确定所述待识别电力目标在所述电力监控图像中的关键位置点，还包括：
将所述电力监控图像输入预训练的关键点检测网络，得到针对所述待识别电力目标的关键点热力图；
根据所述关键点热力图，确定所述待识别电力目标在所述电力监控图像中的关键位置点。
在其中一个实施例中，在所述确定所述待识别电力目标在所述电力监控图像中的关键位置点的步骤之后，所述方法还包括：
根据所述关键位置点，在所述电力监控图像中生成包围所述待识别电力目标的目标框；
若所述目标框的面积处于预设面积范围，则执行所述根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像的步骤；
若所述目标框的面积不处于预设面积范围，则返回所述获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
在其中一个实施例中，所述根据所述电力目标图像，识别出所述待识别电力目标的目标分类结果，包括：
将所述电力目标图像输入至预训练的目标分类网络，生成所述待识别电力目标的目标分类结果。
一种电力目标检测装置，所述装置包括：
获取模块，用于获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像；所述电力监控图像包括有待识别电力目标；
确定模块，用于确定所述待识别电力目标在所述电力监控图像中的关键位置点；
控制模块，用于根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像；在所述电力目标图像中，所述待识别电力目标处于中心主图区；所述第二变焦倍数大于所述第一变焦倍数；
识别模块，用于根据所述电力目标图像，确定所述待识别电力目标的目标分类结果。
一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，所述处理器执行所述计算机程序时实现上述的方法的步骤。
一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现上述的方法的步骤。
上述电力目标检测方法、装置、计算机设备和存储介质，通过获取图像采集设备处于第一变焦倍数时拍摄到的包括有待识别电力目标的电力监控图像，并确定待识别电力目标在电力监控图像中的关键位置点；然后，根据关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到电力目标 图像；在电力目标图像中，待识别电力目标处于中心主图区；第二变焦倍数大于第一变焦倍数；最后，根据电力目标图像，确定待识别电力目标的目标分类结果；如此，可以在电力监控图像识别出待识别电力目标在电力监控图像中的关键位置点后，并基于该关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到清晰度高的电力目标图像，使得电力目标图像可以包含尽可能多的待识别电力目标的特征，进而可以基于该电力目标图像，准确地对待识别电力目标进行分类。
附图说明
图1为一个实施例中一种电力目标检测方法的应用环境图；
图2为一个实施例中一种电力目标检测方法的流程示意图；
图3为一个实施例中一种关键点检测网络的训练流程图；
图4为一个实施例中一种关键点检测网络的网络结构示意图；
图5为一个实施例中一种二倍下采样模块的结构示意图；
图6a为一个实施例中一种零部件目标预测过程展示图；
图6b为一个实施例中一种动物目标预测过程展示图；
图7为另一个实施例中一种电力目标检测方法的流程示意图；
图8为一个实施例中一种电力目标检测方法的处理流程框图；
图9为一个实施例中一种电力目标检测装置的结构框图；
图10为一个实施例中计算机设备的内部结构图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供的电力目标检测方法，可以应用于如图1所示的应用环境中。其中，计算机设备110通过网络与图像采集设备120进行通信。其中，计算机设备110获取图像采集设备120处于第一变焦倍数时拍摄到的电力监控图像； 电力监控图像包括有待识别电力目标。然后，计算机设备110确定待识别电力目标在电力监控图像中的关键位置点。再然后，计算机设备110根据关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到电力目标图像；在电力目标图像中，待识别电力目标处于中心主图区；第二变焦倍数大于第一变焦倍数。最后，计算机设备110根据电力目标图像，确定待识别电力目标的目标分类结果。实际应用中，计算机设备110可以但不限于是各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备。
在一个实施例中，如图2所示，提供了一种电力目标检测方法，以该方法应用于图1中的计算机设备110为例进行说明，包括以下步骤：
步骤S210，获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像；电力监控图像包括有待识别电力目标。
其中，电力监控图像可以是指对电力生产区域进行监控时采集的图像。实际应用中，
其中，电力目标可以是指在电力生产中存在各种需要被检测的目标。例如，在电厂的生产区域，则需要检测小目标有设备的零部件(螺丝、螺母等)以及小动物(蛇、老鼠和猫等)。
其中，图像采集设备可以是指摄像头。实际应用中，摄像头具有变焦功能。例如，数码变焦功能、光学变焦功能。
具体实现中，计算机设备获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像。
步骤S220，确定待识别电力目标在电力监控图像中的关键位置点。
具体实现中，当计算机设备获取到电力监控图像时，计算机设备则确定待识别电力目标在电力监控图像中的关键位置点。具体来说，计算机设备可以将电力监控图像输入至预训练的关键点识别模型。通过该预训练的关键点识别模识别出待识别电力目标在电力监控图像中的关键位置点。
步骤S230，根据关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到电力目标图像。
其中，在电力目标图像中，待识别电力目标处于中心主图区。
其中，第二变焦倍数大于第一变焦倍数。
具体实现中，当计算机设备确定出待识别电力目标在电力监控图像中的关键位置点后，计算机设备则控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到电力目标图像。具体来说，计算机设备将该关键位置点发送至图像采集设备，进而控制该图像采集设备进行聚焦放大，拍摄该待识别电力目标使待识别电力目标处于中心主图区，以获取到电力目标图像。
步骤S240，根据电力目标图像，确定待识别电力目标的目标分类结果。
具体实现中，当计算机设备获取到图像采集装置拍摄到的电力目标图像后，计算机设备则根据电力目标图像，确定出待识别电力目标的目标分类结果。具体来说，计算机设备可以将电力目标图像输入至预训练的目标分类网络，通过该预训练的目标分类网络对电力目标图像进行识别，得到针对待识别电力目标的目标分类结果。
实际应用中，预训练的目标分类网络可以采用MobileNetV2结构，其中，该目标分类网络可以使用数据增强的方法强化对多尺度目标识别的准确性，即在训练时通过调整待识别目标在画面中的占比进行数据增强，来减少因物体在目标框中的大小对目标识别的准确性的影响。
上述电力目标检测方法中，通过获取图像采集设备处于第一变焦倍数时拍摄到的包括有待识别电力目标的电力监控图像，并确定待识别电力目标在电力监控图像中的关键位置点；然后，根据关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到电力目标图像；在电力目标图像中，待识别电力目标处于中心主图区；第二变焦倍数大于第一变焦倍数；最后，根据电力目标图像，确定待识别电力目标的目标分类结果；如此，可以在电力监控图像识别出待识别电力目标在电力监控图像中的关键位置点后，并基于该关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到清晰度高的电力目标图像，使得电力目标图像可以包含尽可能多的待识别电力目标的特征，进而可以基于该电力目标图像，准确地对待识别电力目标进行分类。
在另一个实施例中，在获取图像采集设备处于第一变焦倍数时拍摄到的电 力监控图像的步骤之后，方法还包括：检测电力监控图像中是否存在人体；若电力监控图像中不存在人体，则执行确定待识别电力目标在电力监控图像中的关键位置点的步骤；若电力监控图像中存在人体，则返回获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
具体实现中，在计算机设备获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤之后，计算机设备还需要检测电力监控图像中是否存在人体；若电力监控图像中不存在人体，则触发后续的目标检测网络即执行确定待识别电力目标在电力监控图像中的关键位置点的步骤；若电力监控图像中存在人体，则不触发后续的目标检测网络，返回获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤即输入下一图像帧序列。
本实施例的技术方案，在获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤之后，通过检测电力监控图像中是否存在人体，从而决定是否触发执行确定待识别电力目标在电力监控图像中的关键位置点的步骤，如此，可以检测算法并前置了一个触发机制，在达成某些条件时才触发检测算法进行检测，去除了人体目标的干扰，减少了计算量节省了功耗。
在另一个实施例中，在检测电力监控图像中是否存在人体的步骤之前，还包括：获取电力监控图像与预设的历史背景图像之间的差异；若差异大于预设差异阈值，则返回检测电力监控图像中是否存在人体的步骤。
在另一个实施例中，检测电力监控图像中是否存在人体，包括：将电力监控图像输入至预训练的行人检测网络，得到行人检测结果；行人检测结果包括存在行人和不存在行人中的至少一种；若行人检测结果为存在行人，则判定电力监控图像中存在人体。
具体实现中，计算机设备在检测电力监控图像中是否存在人体之前，具体包括：计算机设备可以获取电力监控图像与预设的历史背景图像之间的差异；若差异大于预设差异阈值，则再触发判定电力监控图像中是否存在人体。
具体来说，计算机设备可以对多帧电力监控图像使用混合高斯模型对运动目标背景建立实时模型，然后取连续的两帧图像进行差分，从而得到运动目标 轮廓，若运动目标轮廓超过预先设定的目标阈值即图像内容变化程度大于预设的阈值时，则判定电力监控图像中存在人体。
其中，混合高斯背景建模是基于像素样本统计信息的背景表示方法，利用像素在较长时间内大量样本值的概率密度等统计信息(如模式数量、每个模式的均值和标准差)表示背景，然后使用统计差分(如3σ原则)进行目标像素判断，可以对复杂动态背景进行建模，在混合高斯背景模型中，认为像素之间的颜色信息互不相关，对各像素点的处理都是相互独立的。对于视频图像中的每一个像素点，其值在序列图像中的变化可看作是不断产生像素值的随机过程，即用高斯分布来描述每个像素点的颜色呈现规律。
混合高斯背景建模的具体流程如下：
(1)每个新像素值X t同当前K个模型按公式(1)进行比较，直到找到匹配新像素值的分布模型，即同该模型的均值偏差在2.5σ内；
|X t-μ i,t-1|≤2.5σ i,t-1      (1)；
其中,X t是t时刻像素X的RGB分量，σ i,t-1为t-1时刻第i个高斯分布的标准差。
(2)如果所匹配的背景模式符合背景要求，则该像素属于背景，否则属于前景。
(3)各模式权值按公式(2)更新对于匹配的模式M k,t＝1,否则M k,t＝0；
w k,t＝(1-α)*w k,t-1+α*M k,t       (2)；
其中，α表示学习率，w k,t，w k,t-1为权值。
(4)未匹配模式的均值μ与标准差σ不变，匹配模式的参数如下公式(3)、公式(4)和公式(5)更新；
ρ＝α*η(X t|μ k,σ k)       (3)；
μ t＝(1-ρ)*μ t-1+ρ*X t       (4)；
Figure PCTCN2021079280-appb-000001
其中，α表示学习率，ρ是参数学习率。
(5)如果第一步中没有任何模式匹配，则权重最小的模式被替换，即该模式的均值为当前像素值，标准差为初始较大值，权重为较小值；
(6)各模式根据w/α 2按降序排列，权重大、标准差小的模式排列靠前；
(7)选前B个模式作为背景，B的计算如公式(6)。
Figure PCTCN2021079280-appb-000002
其中，T为权重阈值，取值范围为[0.5，1]，通过T的设定可以选出描述背景的最佳分布。
其中，图像差分法通过对当前获取的图像与背景历史图像相减，来检测当前图像是否存在异常，是一种简单有效的方法。通过图像的减法运算，可以提高图像处理的速度。
但是，对于实际应用场景，背景会受到各种自然因素的影响。如果背景固定不变，将会带来很大的误差。本发明采用高斯背景建模算法先对输入的图像进行背景建模，然后差分，从而更适用于实际应用场景。差分的计算公式如下：
△F i＝F i-F b      (7)；
其中，□F i表示图像差分后的结果，此处取两帧图片给对应的像素做差，取绝对值然后取所有做差后像素值的平均值，F i为当前帧图像，F b表示由混合高斯背景建模算法得到的背景图像。
最后，计算机设备根据图像差分后的结果，判定电力监控图像中存在人体。
条件触发判断
得到差分图像后对于差分后的结果进行判断，判断条件如公式(8)。
Figure PCTCN2021079280-appb-000003
其中，ΔF i表示图像差分后的结果，T为根据实验得到的一个阈值，当ΔF i小于阈值时，则说明当前图片与背景差异度较低，标志位flag1置False，当ΔF i大于等于阈值时，则说明当前图片与背景差异度符合需要检测的要求，标志位flag1置True。如此，可以基于该标志位flag1判断该电力监控图像中存在人体。
具体实现中，计算机设备将电力监控图像输入至预训练的行人检测网络，得到行人检测结果；行人检测结果包括存在行人和不存在行人中的至少一种。具体来说，预训练的行人检测网络可以是开源的RFBNet300网络。如果预训练 的行人检测网络输出的行人检测结果为存在行人，则将标志位flag2置False，如果预训练的行人检测网络输出的行人检测结果为不存在行人，则标志位flag2置True。然后，计算机设备将标志位flag1和标志位flag2进行求与运算，若求与结果为True(真)则执行所述确定所述待识别电力目标在所述电力监控图像中的关键位置点的步骤。
本实施例的技术方案，通过获取电力监控图像与预设的历史背景图像之间的差异；若差异大于预设差异阈值，则返回检测电力监控图像中是否存在人体的步骤；从而实现只有在检测出电力监控图像与预设的历史背景图像之间的差异大于预设差异阈值时，才进一步执行后续的目标检测处理，减少了多余的计算量，节省了功耗。
在另一个实施例中，确定待识别电力目标在电力监控图像中的关键位置点，包括：将电力监控图像输入预训练的关键点检测网络，得到针对待识别电力目标的关键点热力图；根据关键点热力图，确定待识别电力目标在电力监控图像中的关键位置点。
具体实现中，计算机设备在确定待识别电力目标在电力监控图像中的关键位置点的过程中，具体包括：计算机设备将将电力监控图像输入预训练的关键点检测网络，得到针对待识别电力目标的关键点热力图；根据关键点热力图，确定待识别电力目标在电力监控图像中的关键位置点。
需要说明的是，需要对关键点检测网络进行训练，得到预训练的关键点检测网络。图3实例性地提供了一种关键点检测网络的训练流程图；如图3所示，需要对输入样本进行数据预处理，然后，进行神经网络前向运算，计算损失函数，并反向传播更新该关键点检测网络的网络参数。
数据预处理过程：训练数据为标注了目标框的图片，为了增加样本多样性，首先对数据进行数据增强，包含随机翻转，随机颜色变化、随机对比度变化、随机亮度变化、随机尺度变化等等，然后进行标准化操作，减去均值并除以方差。最后生成目标的热力图，即在每个目标的中心根据目标框大小生成不同方差的二维高斯分布。
神经网络前向运算；将处理好的数据输入网络中，得到输出，包含预测的热力图，和每个预测中心相对于目标框的偏移。
计算损失函数；计算输出热力图与预测热力图的损失，使用Focal loss损失函数(焦点损失函数)，并计算预测目标框偏移的损失，使用smooth L1损失(平滑的L1损失)，损失函数如下：
Figure PCTCN2021079280-appb-000004
Figure PCTCN2021079280-appb-000005
训练网络：使用Adam优化器和反向传播算法更新网络的参数，不断迭代直到网络收敛。
图4提供了一种关键点检测网络的网络结构示意图；其中，concatenate操作为按通道连接特征图，二倍上采样常采用的是双线性插值算法。其中，该关键点检测网络包括二倍下采样模块。为了便于本领域技术人员的理解，图5提供了一种二倍下采样模块的结构示意图。
输入网络得到预测的热力图与预测的目标框偏差，进行后处理。后处理步骤为：
后处理，(1)对热力图进行阈值操作，设置最小阈值为A，则响应值低于A的位置；(2)进行最大池化操作操作提取响应极值点；(3)取极值点位置的预测偏移得到目标包围框。
为了便于本领域技术人员的理解，图6a提供了一种零部件目标预测过程展示图；图6b提供了一种动物目标预测过程展示图。
本实施例的技术方案，通过将电力监控图像输入预训练的关键点检测网络，得到针对待识别电力目标的关键点热力图，并根据关键点热力图，从而实现准确地确定待识别电力目标在电力监控图像中的关键位置点。
在另一个实施例中，在确定待识别电力目标在电力监控图像中的关键位置点的步骤之后，方法还包括：根据关键位置点，在电力监控图像中生成包围待 识别电力目标的目标框；若目标框的面积处于预设面积范围，则执行根据关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到电力目标图像的步骤；若目标框的面积不处于预设面积范围，则返回获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
具体实现中，计算机设备在确定待识别电力目标在电力监控图像中的关键位置点的步骤之后，方法还包括：计算机设备根据关键位置点，在电力监控图像中生成包围待识别电力目标的目标框；然后，计算机设备判断目标框的面积S是否处于预设面积范围；若计算机设备判定该目标框的面积S处于预设面积范围即S满足如下公式：
S 1＜S＜S 2；
其中，S 1和S 2是设置的区域大小限定参数，当目标框的面积大小处于设置范围内时则需要进行二次识别即执行根据关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到电力目标图像的步骤；若目标框的面积不处于预设面积范围，则不需要进行二次识别即返回获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
本实施例的技术方案，通过根据关键位置点，在电力监控图像中生成包围待识别电力目标的目标框；若目标框的面积处于预设面积范围，则执行根据关键位置点，控制图像采集设备处于第二变焦倍数拍摄待识别电力目标，得到电力目标图像的步骤；若目标框的面积不处于预设面积范围，则返回获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤，从而实现基于电力监控图像中的待识别电力目标的大小决定是否进行进一步地识别，即进行二次判断，进一步增加了检测准确性。
在另一个实施例中，如图7所示，提供了一种电力目标检测方法，以该方法应用于图1中的计算机设备110为例进行说明，包括以下步骤：步骤S710，获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像；所述电力监控 图像包括有待识别电力目标。步骤S720，获取所述电力监控图像与预设的历史背景图像之间的差异。步骤S730，若所述差异大于预设差异阈值，则检测所述电力监控图像中是否存在人体。步骤S740，若所述电力监控图像中不存在人体，则确定所述待识别电力目标在所述电力监控图像中的关键位置点。步骤S750，根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像；在所述电力目标图像中，所述待识别电力目标处于中心主图区；所述第二变焦倍数大于所述第一变焦倍数。步骤S760，根据所述电力目标图像，确定所述待识别电力目标的目标分类结果。需要说明的是，上述步骤的具体限定可以参见上文对一种电力目标检测方法的具体限定。
应该理解的是，虽然图2和图7的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图2和图7中的至少一部分步骤可以包括多个步骤或者多个阶段，这些步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤中的步骤或者阶段的至少一部分轮流或者交替地执行。
为了便于本领域技术人员的理解，图8提供了一种电力目标检测方法的处理流程框图；其中，首先输入图像帧序列；计算机设备判断是否满足检测要求；若满足检测要求则将图像帧序列输入至检测网络；并进行后处理；然后，计算机设备进一步判断是否需要进行二次识别；若需要进行二次识别，则执行二次识别；输出识别结果，并返回输入下一个图像帧序列。
在一个实施例中，如图9所示，提供了一种电力目标检测装置，包括：
获取模块910，用于获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像；所述电力监控图像包括有待识别电力目标；
确定模块920，用于确定所述待识别电力目标在所述电力监控图像中的关键 位置点；
控制模块930，用于根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像；在所述电力目标图像中，所述待识别电力目标处于中心主图区；所述第二变焦倍数大于所述第一变焦倍数；
识别模块940，用于根据所述电力目标图像，确定所述待识别电力目标的目标分类结果。
在其中一个实施例中，所述电力目标检测装置，还包括：人体检测模块，用于检测所述电力监控图像中是否存在人体；若所述电力监控图像中不存在人体，则执行所述确定所述待识别电力目标在所述电力监控图像中的关键位置点的步骤；若所述电力监控图像中存在人体，则返回所述获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
在其中一个实施例中，所述电力目标检测装置，还包括：差异获取模块，用于获取所述电力监控图像与预设的历史背景图像之间的差异；若所述差异大于预设差异阈值，则返回所述检测所述电力监控图像中是否存在人体的步骤。
在其中一个实施例中，所述人体检测模块，具体用于将所述电力监控图像输入至预训练的行人检测网络，得到行人检测结果；所述行人检测结果包括存在行人和不存在行人中的至少一种；若所述行人检测结果为存在行人且所述图像内容变化程度大于预设阈值，则判定所述电力监控图像中存在人体。
在其中一个实施例中，所述确定模块920，具体用于将所述电力监控图像输入预训练的关键点检测网络，得到针对所述待识别电力目标的关键点热力图；根据所述关键点热力图，确定所述待识别电力目标在所述电力监控图像中的关键位置点。
在其中一个实施例中，所述电力目标检测装置，还包括：判断模块，用于根据所述关键位置点，在所述电力监控图像中生成包围所述待识别电力目标的目标框；若所述目标框的面积处于预设面积范围，则执行所述根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像的步骤；若所述目标框的面积不处于预设面积范围，则返回所 述获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
在其中一个实施例中，所述识别模块940，具体用于将所述电力目标图像输入至预训练的目标分类网络，生成所述待识别电力目标的目标分类结果。
关于一种电力目标检测装置的具体限定可以参见上文中对于一种电力目标检测方法的限定，在此不再赘述。上述一种电力目标检测装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是终端，其内部结构图可以如图10所示。该计算机设备包括通过系统总线连接的处理器、存储器、通信接口、显示屏和输入装置。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的通信接口用于与外部的终端进行有线或无线方式的通信，无线方式可通过WIFI、运营商网络、NFC(近场通信)或其他技术实现。该计算机程序被处理器执行时以实现一种电力目标检测方法。该计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏，该计算机设备的输入装置可以是显示屏上覆盖的触摸层，也可以是计算机设备外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图10中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，提供了一种计算机设备，包括存储器和处理器，存储器存储有计算机程序，计算机程序被处理器执行时，使得处理器执行上述一种电 力目标检测方法的步骤。此处一种电力目标检测方法的步骤可以是上述各个实施例的一种电力目标检测方法中的步骤。
在一个实施例中，提供了一种计算机可读存储介质，存储有计算机程序，计算机程序被处理器执行时，使得处理器执行上述一种电力目标检测方法的步骤。此处一种电力目标检测方法的步骤可以是上述各个实施例的一种电力目标检测方法中的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和易失性存储器中的至少一种。非易失性存储器可包括只读存储器(Read-Only Memory，ROM)、磁带、软盘、闪存或光存储器等。易失性存储器可包括随机存取存储器(Random Access Memory，RAM)或外部高速缓冲存储器。作为说明而非局限，RAM可以是多种形式，比如静态随机存取存储器(Static Random Access Memory，SRAM)或动态随机存取存储器(Dynamic Random Access Memory，DRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种电力目标检测方法，其特征在于，包括：
    获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像；所述电力监控图像包括有待识别电力目标；
    确定所述待识别电力目标在所述电力监控图像中的关键位置点；
    根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像；在所述电力目标图像中，所述待识别电力目标处于中心主图区；所述第二变焦倍数大于所述第一变焦倍数；
    根据所述电力目标图像，确定所述待识别电力目标的目标分类结果。
  2. 根据权利要求1所述的方法，其特征在于，在所述获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤之后，所述方法还包括：
    检测所述电力监控图像中是否存在人体；
    若所述电力监控图像中不存在人体，则执行所述确定所述待识别电力目标在所述电力监控图像中的关键位置点的步骤；
    若所述电力监控图像中存在人体，则返回所述获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
  3. 根据权利要求2所述的方法，其特征在于，在所述检测所述电力监控图像中是否存在人体的步骤之前，还包括：
    获取所述电力监控图像与预设的历史背景图像之间的差异；
    若所述差异大于预设差异阈值，则返回所述检测所述电力监控图像中是否存在人体的步骤。
  4. 根据权利要求2所述的方法，其特征在于，所述检测所述电力监控图像中是否存在人体，包括：
    将所述电力监控图像输入至预训练的行人检测网络，得到行人检测结果；所述行人检测结果包括存在行人和不存在行人中的至少一种；
    若所述行人检测结果为存在行人，则判定所述电力监控图像中存在人体。
  5. 根据权利要求1所述的方法，其特征在于，所述确定所述待识别电力目标在所述电力监控图像中的关键位置点，包括：
    将所述电力监控图像输入预训练的关键点检测网络，得到针对所述待识别电力目标的关键点热力图；
    根据所述关键点热力图，确定所述待识别电力目标在所述电力监控图像中的关键位置点。
  6. 根据权利要求1所述的方法，其特征在于，在所述确定所述待识别电力目标在所述电力监控图像中的关键位置点的步骤之后，所述方法还包括：
    根据所述关键位置点，在所述电力监控图像中生成包围所述待识别电力目标的目标框；
    若所述目标框的面积处于预设面积范围，则执行所述根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像的步骤；
    若所述目标框的面积不处于预设面积范围，则返回所述获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像的步骤。
  7. 根据权利要求1至6任意一项所述的方法，其特征在于，所述根据所述电力目标图像，识别出所述待识别电力目标的目标分类结果，包括：
    将所述电力目标图像输入至预训练的目标分类网络，生成所述待识别电力目标的目标分类结果。
  8. 一种电力目标检测装置，其特征在于，所述装置包括：
    获取模块，用于获取图像采集设备处于第一变焦倍数时拍摄到的电力监控图像；所述电力监控图像包括有待识别电力目标；
    确定模块，用于确定所述待识别电力目标在所述电力监控图像中的关键位置点；
    控制模块，用于根据所述关键位置点，控制所述图像采集设备处于第二变焦倍数拍摄所述待识别电力目标，得到电力目标图像；在所述电力目标图像中，所述待识别电力目标处于中心主图区；所述第二变焦倍数大于所述第一变焦倍数；
    识别模块，用于根据所述电力目标图像，确定所述待识别电力目标的目标 分类结果。
  9. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至7中任一项所述的方法的步骤。
  10. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1至7中任一项所述的方法的步骤。
PCT/CN2021/079280 2020-12-12 2021-03-05 电力目标检测方法、装置、计算机设备和存储介质 WO2022121130A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011452606.0A CN112989910A (zh) 2020-12-12 2020-12-12 电力目标检测方法、装置、计算机设备和存储介质
CN202011452606.0 2020-12-12

Publications (1)

Publication Number Publication Date
WO2022121130A1 true WO2022121130A1 (zh) 2022-06-16

Family

ID=76344935

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/079280 WO2022121130A1 (zh) 2020-12-12 2021-03-05 电力目标检测方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN112989910A (zh)
WO (1) WO2022121130A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116451168A (zh) * 2023-06-15 2023-07-18 北京国电通网络技术有限公司 异常电力信息生成方法、装置、电子设备和可读介质
CN116953416A (zh) * 2023-09-19 2023-10-27 英迪格(天津)电气有限公司 一种铁路变配电装置运行状态的监控系统

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113989695B (zh) * 2021-09-18 2022-05-20 北京远度互联科技有限公司 目标跟踪方法、装置、电子设备及存储介质
CN115187603A (zh) * 2022-09-13 2022-10-14 国网浙江省电力有限公司 一种基于深度神经网络的电力设备检测方法及装置
CN117612095A (zh) * 2023-11-27 2024-02-27 中国南方电网有限责任公司 一种电力安全控制方法及装置

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150294176A1 (en) * 2012-12-26 2015-10-15 Canon Kabushiki Kaisha Automatic tracking image pickup system
CN105100724A (zh) * 2015-08-13 2015-11-25 电子科技大学 一种基于视觉分析的智能家居远程安全监控方法及装置
CN107336417A (zh) * 2017-06-13 2017-11-10 上海斐讯数据通信技术有限公司 一种基于机器视觉的模具保护方法及系统
CN107766801A (zh) * 2017-09-28 2018-03-06 中国南方电网有限责任公司超高压输电公司检修试验中心 一种基于红外热点追踪的绝缘子识别方法及装置
CN110139037A (zh) * 2019-05-20 2019-08-16 深圳铂石空间科技有限公司 对象监测方法及装置、存储介质和电子设备
CN110149482A (zh) * 2019-06-28 2019-08-20 Oppo广东移动通信有限公司 对焦方法、装置、电子设备和计算机可读存储介质
CN111339846A (zh) * 2020-02-12 2020-06-26 深圳市商汤科技有限公司 图像识别方法及装置、电子设备和存储介质

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150294176A1 (en) * 2012-12-26 2015-10-15 Canon Kabushiki Kaisha Automatic tracking image pickup system
CN105100724A (zh) * 2015-08-13 2015-11-25 电子科技大学 一种基于视觉分析的智能家居远程安全监控方法及装置
CN107336417A (zh) * 2017-06-13 2017-11-10 上海斐讯数据通信技术有限公司 一种基于机器视觉的模具保护方法及系统
CN107766801A (zh) * 2017-09-28 2018-03-06 中国南方电网有限责任公司超高压输电公司检修试验中心 一种基于红外热点追踪的绝缘子识别方法及装置
CN110139037A (zh) * 2019-05-20 2019-08-16 深圳铂石空间科技有限公司 对象监测方法及装置、存储介质和电子设备
CN110149482A (zh) * 2019-06-28 2019-08-20 Oppo广东移动通信有限公司 对焦方法、装置、电子设备和计算机可读存储介质
CN111339846A (zh) * 2020-02-12 2020-06-26 深圳市商汤科技有限公司 图像识别方法及装置、电子设备和存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116451168A (zh) * 2023-06-15 2023-07-18 北京国电通网络技术有限公司 异常电力信息生成方法、装置、电子设备和可读介质
CN116451168B (zh) * 2023-06-15 2023-09-12 北京国电通网络技术有限公司 异常电力信息生成方法、装置、电子设备和可读介质
CN116953416A (zh) * 2023-09-19 2023-10-27 英迪格(天津)电气有限公司 一种铁路变配电装置运行状态的监控系统
CN116953416B (zh) * 2023-09-19 2023-12-08 英迪格(天津)电气有限公司 一种铁路变配电装置运行状态的监控系统

Also Published As

Publication number Publication date
CN112989910A (zh) 2021-06-18

Similar Documents

Publication Publication Date Title
WO2022121130A1 (zh) 电力目标检测方法、装置、计算机设备和存储介质
US11222239B2 (en) Information processing apparatus, information processing method, and non-transitory computer-readable storage medium
CN109255352B (zh) 目标检测方法、装置及系统
CN109271870B (zh) 行人重识别方法、装置、计算机设备及存储介质
CN108960163B (zh) 手势识别方法、装置、设备和存储介质
US10534957B2 (en) Eyeball movement analysis method and device, and storage medium
US8805018B2 (en) Method of detecting facial attributes
JP6351240B2 (ja) 画像処理装置、画像処理方法及びプログラム
CN110097050B (zh) 行人检测方法、装置、计算机设备及存储介质
US11900676B2 (en) Method and apparatus for detecting target in video, computing device, and storage medium
CN111126208B (zh) 行人归档方法、装置、计算机设备及存储介质
JP6756406B2 (ja) 画像処理装置、画像処理方法および画像処理プログラム
CN108764143B (zh) 图像处理方法、装置、计算机设备和存储介质
CN112446379A (zh) 一种动态大场景自适应智能处理方法
CN114155365B (zh) 模型训练方法、图像处理方法及相关装置
CN111062263A (zh) 手部姿态估计的方法、设备、计算机设备和存储介质
JP2012059076A (ja) 情報処理装置、その認識方法及びプログラム
KR20140095333A (ko) 영상에서 객체 추적 방법 및 장치
Ribeiro et al. Face detection on infrared thermal image
CN113743426A (zh) 一种训练方法、装置、设备以及计算机可读存储介质
CN113780145A (zh) 精子形态检测方法、装置、计算机设备和存储介质
CN113706481A (zh) 精子质量检测方法、装置、计算机设备和存储介质
CN116740652A (zh) 一种基于神经网络模型的锈斑面积扩大的监测方法与系统
CN108875488B (zh) 对象跟踪方法、对象跟踪装置以及计算机可读存储介质
Mei et al. Fast template matching in multi-modal image under pixel distribution mapping

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21901847

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21901847

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 14.12.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 21901847

Country of ref document: EP

Kind code of ref document: A1