WO2021151277A1 - 目标物损伤程度判定方法、装置、电子设备及存储介质 - Google Patents

目标物损伤程度判定方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2021151277A1
WO2021151277A1 PCT/CN2020/098985 CN2020098985W WO2021151277A1 WO 2021151277 A1 WO2021151277 A1 WO 2021151277A1 CN 2020098985 W CN2020098985 W CN 2020098985W WO 2021151277 A1 WO2021151277 A1 WO 2021151277A1
Authority
WO
WIPO (PCT)
Prior art keywords
frame
network
target object
sub
preset
Prior art date
Application number
PCT/CN2020/098985
Other languages
English (en)
French (fr)
Inventor
林春伟
刘伟锋
刘莉红
刘玉宇
肖京
Original Assignee
平安科技（深圳）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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021151277A1 publication Critical patent/WO2021151277A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • 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
    • G06V2201/08—Detecting or categorising vehicles

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a method, device, electronic device, and storage medium for determining the degree of damage to a target.
  • the computer vision technology is generally a two-stage detection algorithm.
  • the two-stage detection algorithm uses network layers such as RoIPooling (Region of Interest pooling) to extract feature maps of objects, and sends the feature maps to different branches to complete classification and position regression, and output a set of true frames .
  • the two-stage detection algorithm needs to define a series of dense and evenly distributed anchor frames. These anchor frames will be linked according to their size and different feature maps. The process of matching the anchor frame and the real frame implicitly determines where the real frame is The layer feature map is responsible for the prediction, and the judgment result of the damage degree is obtained after the matching is completed.
  • the inventor realizes that the two methods for judging the degree of damage based on the above have the following shortcomings: 1. Relying on manual damage determination requires a lot of labor cost, time-consuming and low accuracy; 2. Although the two-stage detection algorithm The accuracy can be improved, but there is no theoretical basis. The process of matching the anchor frame with the real frame is unknown, so the recognition accuracy is not enough.
  • This application provides a method, device, electronic device, and computer-readable storage medium for determining the degree of damage to a target.
  • the main purpose of the method is to set the process of matching the anchor frame with the real frame and solve the problem of insufficient recognition accuracy.
  • a method for determining the damage degree of a target object based on a picture includes:
  • Target damage degree classification model includes a fully convolutional neural network
  • the fully convolutional neural network includes a classification sub-network and a border regression sub-network
  • the standard picture training set is input to the full convolutional neural network, and the classification sub-network is used to generate the anchor frame and the real frame of each picture in the standard picture training set, and the frame regression sub-network is used to pair Performing frame regression between the anchor frame and the real frame to obtain a predicted loss value set, and obtain a standard target object damage degree classification model according to the predicted loss value set;
  • the standard target object damage degree classification model is used to analyze the damaged image set of the target object to obtain the target object damage degree identification result.
  • the present application also provides a device for determining the degree of damage to a target, the device comprising:
  • the classification module is used to obtain the original picture training set and the label set corresponding to the original picture training set, and classify the pictures in the original picture training set according to the corresponding labels to obtain the standard picture training set;
  • the frame regression module is used to input the standard picture training set to the fully convolutional neural network, and use the classification sub-network to generate the anchor frame and real frame of each picture in the standard picture training set, and pass all
  • the frame regression sub-network performs frame regression on the anchor frame and the real frame to obtain a set of predicted loss values, and obtain a standard target object damage degree classification model according to the set of predicted loss values;
  • the identification module is used to analyze the damaged image set of the target object using the standard target object damage degree classification model to obtain the target object damage degree identification result.
  • an electronic device which includes:
  • Memory storing at least one instruction
  • the processor implements the following steps when executing instructions stored in the memory:
  • Target damage degree classification model includes a fully convolutional neural network
  • the fully convolutional neural network includes a classification sub-network and a border regression sub-network
  • the standard picture training set is input to the full convolutional neural network, and the classification sub-network is used to generate the anchor frame and the real frame of each picture in the standard picture training set, and the frame regression sub-network is used to pair Performing frame regression between the anchor frame and the real frame to obtain a predicted loss value set, and obtain a standard target object damage degree classification model according to the predicted loss value set;
  • the standard target object damage degree classification model is used to analyze the damaged image set of the target object to obtain the target object damage degree identification result.
  • the present application also provides a computer-readable storage medium having at least one instruction stored in the computer-readable storage medium, and when the at least one instruction is executed by a processor in an electronic device, the following steps are implemented:
  • Target damage degree classification model includes a fully convolutional neural network
  • the fully convolutional neural network includes a classification sub-network and a border regression sub-network
  • the standard picture training set is input to the full convolutional neural network, and the classification sub-network is used to generate the anchor frame and the real frame of each picture in the standard picture training set, and the frame regression sub-network is used to pair Performing frame regression between the anchor frame and the real frame to obtain a predicted loss value set, and obtain a standard target object damage degree classification model according to the predicted loss value set;
  • the standard target object damage degree classification model is used to analyze the damaged image set of the target object to obtain the target object damage degree identification result.
  • This application classifies the pictures in the original picture training set according to the labels to obtain a standard picture training set, and trains a pre-built target damage degree classification model through the standard picture training set, wherein the pre-built target damage degree classification
  • the model includes a classification sub-network and a frame regression sub-network. After the pictures of the standard picture training set generate a real frame and a series of anchor frames through the classification sub-network, the frame regression can be performed through the frame regression sub-network, so the anchor
  • the process of matching the frame and the real frame is configurable. Therefore, the method, device and computer-readable storage medium for determining the damage degree of the target proposed in this application can improve the recognition accuracy.
  • FIG. 1 is a schematic flowchart of a method for determining the degree of damage to a target provided by an embodiment of the application;
  • FIG. 2 is a schematic flowchart of a frame regression method provided by an embodiment of the application
  • FIG. 3 is a functional block diagram of a device for determining the degree of damage to a target provided by an embodiment of the application
  • FIG. 4 is a schematic structural diagram of an electronic device for realizing the method for determining the damage degree of a target provided by an embodiment of the application.
  • This application provides a method for determining the damage degree of a target object.
  • FIG. 1 it is a schematic flowchart of a method for determining the damage degree of a target object based on a picture according to an embodiment of the present application.
  • the method can be executed by a device, and the device can be implemented by software and/or hardware.
  • the method for judging the damage degree of the target object based on the picture includes:
  • the original picture training set described in the embodiment of the present application includes massive object damage pictures.
  • the label set refers to extracting features of a preset type of all pictures in the original picture training set, and generating them based on the same characteristics.
  • the features of the preset type include the image viewing angle, the degree of damage, and the like.
  • the image viewing angle feature includes five different labels: front view, top view, bottom view, left side view, and right side view; the degree of damage includes five levels of level A, level B, level C, level D, and level E. Labels etc.
  • the pictures in the original picture training set are classified according to corresponding tags to form a standard picture training set. For example, if a picture in the original picture training set belongs to the front-view perspective, and the loss level is B, it is classified into the B-level picture set of the front-view set, and so on.
  • the fully convolutional neural network includes an input layer, a multi-x multi-convolutional layer, a pooling layer, and a 1x1 convolutional layer, which can classify the picture at the pixel level, thereby achieving the purpose of identifying a specific part of the picture.
  • the input layer is used to receive input pictures and perform normalization processing on the received input pictures, such as normalizing the pixel values of pixels in the input pictures from the range of [0,255] to [0,1 ] Interval.
  • the multi-x multi-convolutional layer includes a convolution kernel, a convolution step size, and padding, and is used to perform feature extraction on the input picture to obtain a feature map.
  • the convolution step length defines the distance between the positions when the convolution kernel scans the picture twice, for example, the convolution step length is At 1, the convolution kernel will scan through the pixels of the input picture one by one, and when the step size is n, it will skip n-1 pixels in the next scan.
  • the padding is used to increase the size of the input picture before passing the convolution kernel to offset the influence of the size shrinkage in the calculation.
  • a 16 ⁇ 16 picture passes through a unit step of a 5 ⁇ 5 convolution kernel. Before the same filling is performed, two layers will be filled in the horizontal and vertical directions, that is, two pixels are added on both sides to become a 20 ⁇ 20 image. After passing the convolution kernel, the output feature map size is 16 ⁇ 16, Keep the original size.
  • the pooling layer is used to perform further feature selection and filtering on the features extracted by the multiple x multiple convolutional layers.
  • the 1 ⁇ 1 convolutional layer is used to adjust the picture to any size and output.
  • the fully convolutional neural network in the original target damage degree classification model in the preferred embodiment of the present application further includes a classification sub-network and a bounding box regression sub-network.
  • the classified self-network is used to detect the target in the picture to obtain a real frame and a series of anchor frames
  • the frame regression sub-network is used to frame a series of anchor frames and the real frame in the picture. return.
  • the classification sub-network and the frame regression sub-network are both composed of one anchored frame branch and one non-anchor framed branch, and each anchored frame branch is composed of 4 layers of convolutional layers.
  • the number of feature maps output by the anchor box branches in the classification sub-network is the product of the number of categories and the number of anchor boxes, and the number of feature maps output by the anchor box branches in the border regression sub-network is 4 times the number of anchor boxes .
  • the anchor frame refers to the predicted bounding box generated by collecting a large number of regions in the input picture with the pixel as the center;
  • the bounding box regression refers to marking the predicted bounding box generated during the target detection process
  • the good real bounding box is the process of approaching the target, and then it is judged whether the predicted bounding box contains the target, and the edge of the predicted bounding box is adjusted to predict the target more accurately.
  • the classification sub-network and the bounding box regression sub-network described in the embodiment of the present application include a loss function.
  • the loss function refers to the difference between the forward calculation result of each iteration in the full convolutional neural network and the true value, so as to guide the next training to proceed in the correct direction.
  • the loss function includes a focal loss function (Focal Loss) and an intersection ratio loss function (IoU Loss), and the formulas are:
  • the Focal Loss and IoU Loss are the loss functions of the classification sub-network and the frame regression sub-network, respectively.
  • the S3 includes:
  • S31 Input the standard picture training set to the classification sub-network to obtain a picture feature set, and the pictures in the picture feature set include a real frame and a series of anchor frames;
  • S32 Input the picture feature set to the frame regression sub-network, and perform frame regression on the anchor frame and the real frame through the frame regression sub-network to obtain the intersection ratio of the anchor frame and the real frame;
  • the S31 includes:
  • the S32 includes:
  • the target object damage degree prediction loss value set is obtained.
  • intersection ratio is the ratio of the intersection area and the intersection area of the anchor frame and the real frame.
  • the method when the predicted loss value set does not meet the preset loss requirement, the method further includes:
  • the standard target object damage degree classification model is obtained according to the coefficient of the current loss function in the frame regression sub-network.
  • the damaged image set of the target object is received, and the damaged image set is preprocessed and transmitted to the standard target damage degree classification model to obtain the target damage degree identification result.
  • the pre-processing in the embodiment of the present application includes: extracting features of a preset type of all pictures in the damaged image set, for example, extracting the image perspective of the preset type, including a front view, a top view, and a bottom view , Left side view and right side view.
  • the damaged image set is transmitted to the standard target object damage degree classification model to obtain multiple target object damage degree identification results, and according to the multiple target object damage degree identification results, Get the final identification result of target damage degree.
  • the front view, top view, bottom view, left view, and right view pictures of the same damaged part of a target are input into the standard target damage degree classification model, and the standard target is used.
  • the object damage degree classification model analysis obtains the damage degree identification result corresponding to each picture angle of the same damaged part, and composes the target object damage degree identification result set.
  • the target damage degree evaluation result set it is judged whether there is a ratio of the evaluation result that exceeds the preset ratio value. If the ratio of the evaluation result exceeds the preset ratio value, the current damage degree identification is successful, and the final target object damage degree
  • the identification result is the identification result exceeding the preset ratio value. If the ratio of none of the evaluation results in the target object damage degree evaluation result set exceeds the preset ratio value, then the current damage degree evaluation fails.
  • the preset ratio value is 60%
  • the user inputs the front view, top view, bottom view, left side view, and right side view of the same damaged part of the damaged vehicle into the standard target damage classification model, and outputs
  • the target damage evaluation result set is as follows: front view, damage level B, top view, damage level B, bottom view, damage level B, left side view, damage level C, right view, damage level B, the same evaluation result If the ratio is 80%, which is greater than the preset ratio value, the output identification result is the damage level B.
  • This application classifies the pictures in the original picture training set according to the labels to obtain a standard picture training set, and trains a pre-built target damage degree classification model through the standard picture training set, wherein the pre-built target damage degree classification
  • the model includes a classification sub-network and a frame regression sub-network. After the pictures of the standard picture training set generate a real frame and a series of anchor frames through the classification sub-network, the frame regression can be performed through the frame regression sub-network, so the anchor
  • the process of matching the frame and the real frame is configurable. Therefore, the method, device and computer-readable storage medium for determining the damage degree of the target proposed in this application can improve the recognition accuracy.
  • FIG. 3 it is a functional block diagram of the device for determining the degree of damage to the target object of the present application.
  • the device 100 for determining the degree of damage to the target described in the present application may be installed in an electronic device.
  • the device for determining the degree of damage of the target object may include a classification module 101, a frame regression module 102, and an identification module 103.
  • the module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.
  • each module/unit is as follows:
  • the classification module 101 is configured to obtain an original picture training set and a label set corresponding to the original picture training set, and classify the pictures in the original picture training set according to corresponding labels to obtain a standard picture training set;
  • the frame regression module 102 is configured to input the standard picture training set to the fully convolutional neural network, and use the classification sub-network to generate the anchor frame and the real frame of each picture in the standard picture training set , Performing frame regression on the anchor frame and the real frame through the frame regression sub-network to obtain a set of predicted loss values, and obtain a standard target object damage degree classification model according to the set of predicted loss values;
  • the identification module 103 is configured to analyze the damaged image set of the target object by using the standard target object damage degree classification model to obtain the target object damage degree identification result.
  • each module of the target damage degree determination device is as follows:
  • the classification module 101 obtains an original picture training set and a label set corresponding to the original picture training set, and classifies the pictures in the original picture training set according to corresponding labels to obtain a standard picture training set.
  • the original picture training set described in this embodiment of the application includes massive object damage pictures.
  • the label set refers to extracting features of a preset type of all pictures in the original picture training set, and generating them based on the same characteristics.
  • the features of the preset type include the image viewing angle, the degree of damage, and the like.
  • the image viewing angle feature includes five different labels: front view, top view, bottom view, left side view, and right side view; the degree of damage includes five levels of level A, level B, level C, level D, and level E. Labels etc.
  • the classification module 101 described in the embodiment of the present application classifies the pictures in the original picture training set according to corresponding tags to form a standard picture training set. For example, if a picture in the original picture training set belongs to the front-view perspective, and the loss level is B, it is classified into the B-level picture set of the front-view set, and so on.
  • the frame regression module 102 inputs the standard picture training set to the fully convolutional neural network, and uses the classification sub-network to generate the anchor frame and the real frame of each picture in the standard picture training set, and passes all The frame regression sub-network performs frame regression on the anchor frame and the real frame to obtain a prediction loss value set, and obtains a standard target object damage classification model according to the prediction loss value set.
  • the frame regression module 102 specifically performs the following operations:
  • the standard picture training set to the classification sub-network to obtain a picture feature set
  • the pictures in the picture feature set include a real frame and a series of anchor frames
  • the set of predicted loss values may be stored in a node in a blockchain
  • the standard target object damage degree classification model is obtained and output.
  • the fully convolutional neural network includes an input layer, a multi-x multi-convolutional layer, a pooling layer, and a 1x1 convolutional layer, which can classify the picture at the pixel level, thereby achieving the purpose of identifying a specific part of the picture.
  • the input layer is used to receive input pictures and perform normalization processing on the received input pictures, such as normalizing the pixel values of pixels in the input pictures from the range of [0,255] to [0,1 ] Interval.
  • the multi-x multi-convolutional layer includes a convolution kernel, a convolution step size, and padding, and is used to perform feature extraction on the input picture to obtain a feature map.
  • the convolution step length defines the distance between the positions when the convolution kernel scans the picture twice, for example, the convolution step length is At 1, the convolution kernel will scan through the pixels of the input picture one by one, and when the step size is n, it will skip n-1 pixels in the next scan.
  • the padding is used to increase the size of the input picture before passing the convolution kernel to offset the influence of the size shrinkage in the calculation.
  • a 16 ⁇ 16 picture passes through a unit step of a 5 ⁇ 5 convolution kernel. Before the same filling is performed, two layers will be filled in the horizontal and vertical directions, that is, two pixels are added on both sides to become a 20 ⁇ 20 image. After passing the convolution kernel, the output feature map size is 16 ⁇ 16, Maintain the original size.
  • the pooling layer is used to perform further feature selection and filtering on the features extracted by the multiple x multiple convolutional layers.
  • the 1 ⁇ 1 convolutional layer is used to adjust the picture to any size and output.
  • the fully convolutional neural network in the original target damage degree classification model in the preferred embodiment of the present application further includes a classification sub-network and a bounding box regression sub-network.
  • the classified self-network is used to detect the target in the picture to obtain a real frame and a series of anchor frames
  • the frame regression sub-network is used to frame a series of anchor frames and the real frame in the picture. return.
  • the classification sub-network and the frame regression sub-network are both composed of one anchored frame branch and one non-anchor framed branch, and each anchored frame branch is composed of 4 layers of convolutional layers.
  • the number of feature maps output by the anchor box branches in the classification sub-network is the product of the number of categories and the number of anchor boxes, and the number of feature maps output by the anchor box branches in the border regression sub-network is 4 times the number of anchor boxes .
  • the anchor box refers to a predicted bounding box generated by collecting a large number of regions in the input picture with a pixel as the center;
  • the bounding box regression refers to marking the predicted bounding box generated during the target detection process
  • the good real bounding box is the process of approaching the target, and then it is judged whether the predicted bounding box contains the target, and the edge of the predicted bounding box is adjusted to predict the target more accurately.
  • the classification sub-network and the bounding box regression sub-network described in the embodiment of the present application include a loss function.
  • the loss function refers to the difference between the forward calculation result of each iteration in the full convolutional neural network and the true value, so as to guide the next training to proceed in the correct direction.
  • the loss function includes a focal loss function (Focal Loss) and an intersection ratio loss function (IoU Loss), and the formulas are:
  • the Focal Loss and IoU Loss are the loss functions of the classification sub-network and the frame regression sub-network, respectively.
  • the frame regression module 102 of the present application obtains the standard target damage degree classification model through the following operations: compare each loss value in the predicted loss value set with a preset loss value threshold; calculate that it is greater than the preset loss value The number of loss values of the loss value threshold and the ratio value of the number of loss values less than or equal to the preset loss value threshold; when the ratio value is greater than the preset ratio threshold, the adjustment frame regressor is executed The coefficient of the loss function preset in the network; when the ratio value is less than or equal to the preset ratio threshold value, the coefficient of the current loss function in the frame regression sub-network is directly used to obtain the standard target damage degree classification Model.
  • the identification module 103 uses the standard target object damage degree classification model to analyze the damaged image set of the target object to obtain the target object damage degree identification result.
  • the damaged image set of the target object is received, and the damaged image set is preprocessed and transmitted to the standard target damage degree classification model to obtain the target damage degree identification result.
  • the pre-processing in the embodiment of the present application includes: extracting features of a preset type of all pictures in the damaged image set, for example, extracting the preset type of picture angle of view, including a front view, a top view, and a bottom view. , Left side view and right side view.
  • the damaged image set is transmitted to the standard target object damage degree classification model to obtain multiple target object damage degree identification results, and according to the multiple target object damage degree identification results, Get the final identification result of target damage degree.
  • a target object such as front view, top view, bottom view, left side view, and right side view pictures of the same damaged part of a vehicle
  • the damage degree classification model analysis obtains the damage degree identification result corresponding to each picture angle of the same damaged part, and composes the target object damage degree identification result set.
  • the target damage degree evaluation result set it is judged whether there is a ratio of the evaluation result that exceeds the preset ratio value. If the ratio of the evaluation result exceeds the preset ratio value, the current damage degree identification is successful, and the final target object damage degree The identification result is the identification result exceeding the preset ratio value. If the ratio of none of the evaluation results in the target object damage degree evaluation result set exceeds the preset ratio value, then the current damage degree evaluation fails.
  • the preset ratio value is 60%
  • the user inputs the front view, top view, bottom view, left side view, and right side view of the same damaged part of the damaged vehicle into the standard target damage classification model, and outputs
  • the target damage evaluation result set is as follows: front view, damage level B, top view, damage level B, bottom view, damage level B, left side view, damage level C, right side view, damage level B, the same evaluation result If the ratio is 80%, which is greater than the preset ratio value, the output identification result is the damage level B.
  • FIG. 4 it is a schematic diagram of the structure of an electronic device that implements the method for determining the damage degree of a target object according to the present application.
  • the electronic device 1 may include a processor 10, a memory 11, and a bus, and may also include a computer program stored in the memory 11 and running on the processor 10, such as a picture-based target recognition program 12.
  • the memory 11 includes at least one type of readable storage medium.
  • the readable storage medium may be non-volatile or volatile.
  • the readable storage medium includes flash memory, mobile hard disk, and multimedia card.
  • Card-type memory for example: SD or DX memory, etc.
  • magnetic memory magnetic disk, optical disk, etc.
  • the memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, for example, a mobile hard disk of the electronic device 1.
  • the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart media card (SMC), and a secure digital (Secure Digital) equipped on the electronic device 1.
  • SD Secure Digital
  • flash card Flash Card
  • the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device.
  • the memory 11 can not only be used to store application software and various data installed in the electronic device 1, such as the code of the image-based target recognition program 12, etc., but also can be used to temporarily store data that has been output or will be output.
  • the processor 10 may be composed of integrated circuits in some embodiments, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same function or different functions, including one or more Combinations of central processing unit (CPU), microprocessor, digital processing chip, graphics processor, and various control chips, etc.
  • the processor 10 is the control unit of the electronic device, which uses various interfaces and lines to connect the various components of the entire electronic device, and runs or executes programs or modules stored in the memory 11 (such as executing Image-based target recognition programs, etc.), and call data stored in the memory 11 to execute various functions of the electronic device 1 and process data.
  • the bus may be a peripheral component interconnect standard (PCI) bus or an extended industry standard architecture (EISA) bus, etc.
  • PCI peripheral component interconnect standard
  • EISA extended industry standard architecture
  • the bus can be divided into address bus, data bus, control bus and so on.
  • the bus is configured to implement connection and communication between the memory 11 and at least one processor 10 and the like.
  • FIG. 3 only shows an electronic device with components. Those skilled in the art can understand that the structure shown in FIG. 3 does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure. Components, or combinations of certain components, or different component arrangements.
  • the electronic device 1 may also include a power source (such as a battery) for supplying power to various components.
  • the power source may be logically connected to the at least one processor 10 through a power management device, thereby controlling power
  • the device implements functions such as charge management, discharge management, and power consumption management.
  • the power supply may also include any components such as one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, and power status indicators.
  • the electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.
  • the electronic device 1 may also include a network interface.
  • the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used in the electronic device 1 Establish a communication connection with other electronic devices.
  • the electronic device 1 may also include a user interface.
  • the user interface may be a display (Display) and an input unit (such as a keyboard (Keyboard)).
  • the user interface may also be a standard wired interface or a wireless interface.
  • the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc.
  • the display can also be appropriately called a display screen or a display unit, which is used to display the information processed in the electronic device 1 and to display a visualized user interface.
  • the picture-based target recognition program 12 stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can realize:
  • Target damage degree classification model includes a fully convolutional neural network
  • the fully convolutional neural network includes a classification sub-network and a border regression sub-network
  • the standard picture training set is input to the full convolutional neural network, and the classification sub-network is used to generate the anchor frame and the real frame of each picture in the standard picture training set, and the frame regression sub-network is used to pair Performing frame regression between the anchor frame and the real frame to obtain a predicted loss value set, and obtain a standard target object damage degree classification model according to the predicted loss value set;
  • the standard target object damage degree classification model is used to analyze the damaged image set of the target object to obtain the target object damage degree identification result.
  • the integrated module/unit of the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile or volatile computer readable storage medium .
  • the computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory) .
  • modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • the functional modules in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software functional modules.
  • the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

Landscapes

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

Abstract

一种目标物损伤程度判定方法，涉及人工智能技术，包括：通过标签对获取的原始图片训练集进行分类，得到标准图片训练集（S1），构建包括分类子网络和边框回归子网络的目标物损伤程度分类模型（S2），利用分类子网络生成标准图片训练集中每个图片的锚框及真实框，通过边框回归子网络对锚框与真实框进行边框回归，得到预测损失值集，根据预测损失值集得到标准目标物损伤程度分类模型（S3），利用标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果（S4）。还涉及区块链技术，模型训练所需的数据可存储在区块链中的节点。可以提高识别目标物损伤程度的精度。

Description

目标物损伤程度判定方法、装置、电子设备及存储介质
本申请要求于2020年5月26日提交中国专利局、申请号为CN202010458345.7、发明名称为“目标物损伤程度判定方法、装置、电子设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域，尤其涉及一种目标物损伤程度判定方法、装置、电子设备及存储介质。
背景技术
车辆等物体的损伤程度的判断通常有两种方式：一是由人工判断，但每个人的判断标准以及经验丰富程度存在差异，导致损伤程度的判断结果的主观性较大且准确性不足；二是运用计算机视觉技术进行损伤程度的判定。所述计算机视觉技术一般为二阶段检测算法。所述二阶段检测算法是利用RoIPooling(Region of Interest pooling，感兴趣区域池)等网络层提取物体的特征图，并将所述特征图送入不同分支完成分类和位置回归，输出一组真实框。所述二阶段检测算法需要定义一系列稠密均匀分布的锚框，这些锚框会根据其尺寸大小和不同的特征图联系起来，锚框与真实框匹配的过程隐式地决定了真实框由哪层特征图负责预测，匹配完成得到损伤程度的判定结果。
发明人意识到基于以上的物体损伤程度的两种判断方式存在下述缺陷：1、依靠人工进行定损需要耗费大量人力成本、耗时较长且准确率较低；2、二阶段检测算法虽然可以提高准确率，但缺乏理论依据，所述锚框与真实框匹配的过程是未知的，因此识别精度不够。
发明内容
本申请提供一种目标物损伤程度判定方法、装置、电子设备及计算机可读存储介质，其主要目的在于设定锚框与真实框匹配的过程，解决识别精度不够的
为实现上述目的，本申请提供的一种基于图片的目标物损伤程度判定方法，包括：
获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
构建目标物损伤程度分类模型，其中，所述目标物损伤程度分类模型包括全卷积神经网络，所述全卷积神经网络包括分类子网络和边框回归子网络；
将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
为了解决上述问题，本申请还提供一种目标物损伤程度判定装置，所述装置包括：
分类模块，用于获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
边框回归模块，用于将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
鉴定模块，用于利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行 分析，得到目标物损伤程度鉴定结果。
为了解决上述问题，本申请还提供一种电子设备，所述电子设备包括：
存储器，存储至少一个指令；及
处理器，执行所述存储器中存储的指令时实现如下步骤：
获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
构建目标物损伤程度分类模型，其中，所述目标物损伤程度分类模型包括全卷积神经网络，所述全卷积神经网络包括分类子网络和边框回归子网络；
将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
为了解决上述问题，本申请还提供一种计算机可读存储介质，所述计算机可读存储介质中存储有至少一个指令，所述至少一个指令被电子设备中的处理器执行时实现如下步骤：
获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
构建目标物损伤程度分类模型，其中，所述目标物损伤程度分类模型包括全卷积神经网络，所述全卷积神经网络包括分类子网络和边框回归子网络；
将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
本申请将原始图片训练集中的图片按照标签进行分类，得到标准图片训练集，通过所述标准图片训练集训练预构建的目标物损伤程度分类模型，其中，所述预构建的目标物损伤程度分类模型中包括分类子网络和边框回归子网络，所述标准图片训练集的图片经过所述分类子网络生成真实框和一系列锚框后，可通过所述边框回归子网络进行边框回归，所以锚框与真实框匹配的过程是可设定的，因此本申请提出的目标物损伤程度判定方法、装置及计算机可读存储介质，可以提高识别精度。
附图说明
图1为本申请一实施例提供的目标物损伤程度判定方法的流程示意图；
图2为本申请一实施例提供的边框回归方法的流程示意图；
图3为本申请一实施例提供的目标物损伤程度判定装置的功能模块图；
图4为本申请一实施例提供的实现目标物损伤程度判定方法的电子设备的结构示意图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所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。
本申请提供一种目标物损伤程度判定方法。参照图1所示，为本申请一实施例提供的基于图片的目标物损伤程度判定方法的流程示意图。该方法可以由一个装置执行，该装置可以由软件和/或硬件实现。
在本实施例中，基于图片的目标物损伤程度判定方法包括：
S1、获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集。
较佳地，本申请实施例所述原始图片训练集包括海量物体损伤图片。所述标签集是指提取所述原始图片训练集中所有图片的预设种类的特征，并根据所述相同特征生成的。本申请实施例中，所述预设种类的特征包括图片视角、损伤程度等。例如，图片视角特征包括正视图、俯视图、仰视图、左侧视图和右侧视图5种不同的标签；所述损伤程度包括级别A、级别B、级别C、级别D、级别E五个级别的标签等。
本申请实施例将所述原始图片训练集中的图片按照对应标签进行分类，形成标准图片训练集。例如，所述原始图片训练集中的一张图片属于正视视角，损失级别为B，则将其分类到正视集合的B等级图片集内，以此类推。
S2、构建目标物损伤程度分类模型，其中，所述目标物损伤程度分类模型包括全卷积神经网络，所述全卷积神经网络包括分类子网络和边框回归子网络。
详细地，所述全卷积神经网络包括输入层、多x多卷积层、池化层及1x1卷积层，可以对图片进行像素级的分类，进而达到识别图片中特定部分的目的。
其中，所述输入层用于接收输入的图片并对接收的输入的图片进行标准化处理，如将输入的图片中的像素的像素值由[0，255]的范围归一化至[0,1]区间。
所述多x多卷积层包括卷积核、卷积步长和填充，用于对输入的图片进行特征提取得到特征图。其中，所述卷积核越大，可提取的图片的特征越复杂；所述卷积步长定义了卷积核相邻两次扫过所述图片时位置的距离，如卷积步长为1时，卷积核会逐个扫过输入的所述图片的像素，步长为n时会在下一次扫描跳过n-1个像素。所述填充用于在输入的所述图片通过卷积核之前，增大其尺寸以抵消计算中尺寸收缩产生的影响，如16×16的图片在经过单位步长的5×5的卷积核之前先进行相同填充，则会在水平和垂直方向填充两层，即两侧各增加2个像素变为20×20大小的图像，通过卷积核后，输出的特征图尺寸为16×16，保持了原本的尺寸。
所述池化层用于对所述多x多卷积层提取的特征进行进一步地特征选择和过滤。
所述1×1卷积层用于将图片调整为任意大小尺寸并输出。
进一步地，本申请优选实施例中所述原始目标物损伤程度分类模型中的全卷积神经网络还包含分类子网络和边框回归子网络。其中，所述分类自网络用于检测所述图片中的目标物得到真实框和一系列锚框，及所述边框回归子网络用于对所述图片中的一系列锚框与真实框进行边框回归。所述分类子网络和边框回归子网络均由1个有锚框分支和1个无锚框分支构成，每个有锚框分支分别由4层卷积层组成。所述分类子网络中的有锚框分支输出的特征图数量为类别数与锚框数的乘积，所述边框回归子网络中的有锚框分支输出的特征图数量为锚框数的4倍。其中，所述锚框是指，以像素点为中心，在输入图片中采集大量的区域，产生的预测边界框；所述边框回归是指目标检测过程中对产生的所述预测边界框以标注好的所述真实边界框为目标进行逼近的过程，然后判断这些所述预测边界框中是否包含目标，并调整所述预测边界框边缘从而更准确预测目标。
较佳地，本申请实施例所述分类子网络和边框回归子网络包括损失函数。
所述损失函数是指全卷积神经网络中每次迭代的前向计算结果与真实值的差距，从而指导下一步的训练向正确的方向进行的函数。本申请实施例中，所述损失函数包括焦点损失函数(Focal Loss)和交并比损失函数(IoU Loss)，公式分别为：
Figure PCTCN2020098985-appb-000001
Figure PCTCN2020098985-appb-000002
其中，所述Focal Loss和IoU Loss分别为所述分类子网络的和边框回归子网络的损失函数。其中，
Figure PCTCN2020098985-appb-000003
表示分类子网络的损失函数，
Figure PCTCN2020098985-appb-000004
表示为边框回归子网络的损失函数，l表示为卷积层数，i表示类别数，j表示锚框数,
Figure PCTCN2020098985-appb-000005
表示为损失系数。
S3、将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型。
详细地，参照图2所示，所述S3包括：
S31：将所述标准图片训练集输入至所述分类子网络得到图片特征集，所述图片特征集中的图片包括真实框与一系列锚框；
S32：将所述图片特征集输入至边框回归子网络，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到所述锚框与所述真实框的交并比；
S33：将所述锚框与所述真实框的交并比，通过非极大值抑制得到预测损失值集，所述预测损失值集可存储在区块链中的节点；
S34：判断所述预测损失值集是否满足预设损失要求。
在所述预测损失值集不满足预设损失要求时，调整边框回归子网络中的损失函数的系数后，返回S31；
S35：在所述预测损失值集满足预设损失要求，得到并输出所述标准目标物损伤程度分类模型。
具体地，所述S31包括：
在所述分类子网络中获取所述标准图片训练集内的图片，检测所述图片中的目标物得到真实框，并对所述图片中的所有像素点进行随机采样，得到一系列以像素点为中心的锚框，得到所述图片特征集。
详细地，所述S32包括：
将所述图片特征集输入至边框回归子网络；
利用所述边框回归子网络中的损失函数对所述图片中的一系列锚框与真实框进行边框回归，使得锚框向真实框不断逼近，得到所述锚框及真实框的交并比，从而得到所述目标物损伤程度预测损失值集。
其中，所述交并比为所述锚框及真实框相交面积和相并面积之比。
进一步地，本申请其他较佳实施例中，在所述预测损失值集不满足预设损失要求，该方法还包括：
将所述预测损失值集中内每个损失值与预设的损失值阈值进行比较；
计算大于所述预设的损失值阈值的损失值的数量和小于或者等于所述预设的损失值阈值的损失值的数量的比例值；
在所述比例值大于预设的比例阈值时，调整边框回归子网络中预设的损失函数的系数；
在所述比例值小于或者等于所述预设的比例阈值时，根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
S4、利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
本申请实施例接收所述目标物的受损图像集，并对所述受损图像集进行预处理后传输至所述标准目标物损伤程度分类模型中，得到目标物损伤程度鉴定结果。
详细地，本申请实施例中所述预处理包括：提取所述受损图像集中所有图片的预设种类的特征，如，提取所述预设种类的图片视角，包括正视图、俯视图、仰视图、左侧视图和右侧视图。
进一步地，本申请实施例将所述受损图像集传输至所述标准目标物损伤程度分类模型 中，得到多个目标物损伤程度鉴定结果，并根据所述多个目标物损伤程度鉴定结果，得到最终的目标物损伤程度鉴定结果。
例如，本申请实施例将目标物，如车辆，的同一受损部位的正视、俯视、仰视、左侧视和右侧视图图片输入至所述标准目标物损伤程度分类模型，利用所述标准目标物损伤程度分类模型分析得到同一受损部位的每种图片视角对应的损伤程度鉴定结果，组成目标物损伤程度鉴定结果集。在所述目标物损伤程度鉴定结果集判断是否有一种鉴定结果的比例超过预设比例值，若鉴定结果的比例超过预设比例值，则此次的损伤程度鉴定成功，最终的目标物损伤程度鉴定结果为超过预设比例值鉴定结果。若在所述目标物损伤程度鉴定结果集中没有一种鉴定结果的比例超过预设比例值，则此次的损伤程度鉴定失败。
例如，预设比例值为60％，用户将受损车辆的同一受损部位的正视图、俯视图、仰视图、左侧视图和右侧视图输入至所述标准目标物损伤程度分类模型中，输出的目标物损伤程度鉴定结果集如下：正视图、损伤级别B，俯视、损伤级别B，仰视图、损伤级别B，左侧视图、损伤级别C，右侧视图，损伤级别B，同一鉴定结果的比例为80％，大于预设比例值，则输出鉴定结果为损伤级别B。
本申请将原始图片训练集中的图片按照标签进行分类，得到标准图片训练集，通过所述标准图片训练集训练预构建的目标物损伤程度分类模型，其中，所述预构建的目标物损伤程度分类模型中包括分类子网络和边框回归子网络，所述标准图片训练集的图片经过所述分类子网络生成真实框和一系列锚框后，可通过所述边框回归子网络进行边框回归，所以锚框与真实框匹配的过程是可设定的，因此本申请提出的目标物损伤程度判定方法、装置及计算机可读存储介质，可以提高识别精度。
如图3所示，是本申请的目标物损伤程度判定装置的功能模块图。
本申请所述目标物损伤程度判定装置100可以安装于电子设备中。根据实现的功能，所述目标物损伤程度判定装置可以包括分类模块101、边框回归模块102、鉴定模块103。本发所述模块也可以称之为单元，是指一种能够被电子设备处理器所执行，并且能够完成固定功能的一系列计算机程序段，其存储在电子设备的存储器中。
在本实施例中，关于各模块/单元的功能如下：
所述分类模块101，用于获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
所述边框回归模块102，用于将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
所述鉴定模块103，用于利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
详细地，所述目标物损伤程度判定装置各模块的具体实施步骤如下：
所述分类模块101获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集。
本申请实施例所述原始图片训练集包括海量物体损伤图片。所述标签集是指提取所述原始图片训练集中所有图片的预设种类的特征，并根据所述相同特征生成的。本申请实施例中，所述预设种类的特征包括图片视角、损伤程度等。例如，图片视角特征包括正视图、俯视图、仰视图、左侧视图和右侧视图5种不同的标签；所述损伤程度包括级别A、级别B、级别C、级别D、级别E五个级别的标签等。
本申请实施例所述分类模块101将所述原始图片训练集中的图片按照对应标签进行分类，形成标准图片训练集。例如，所述原始图片训练集中的一张图片属于正视视角，损失级别为B，则将其分类到正视集合的B等级图片集内，以此类推。
所述边框回归模块102将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型。
详细地，所述边框回归模块102具体执行下述操作：
将所述标准图片训练集输入至所述分类子网络得到图片特征集，所述图片特征集中的图片包括真实框与一系列锚框；
将所述图片特征集输入至边框回归子网络，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到所述锚框与所述真实框的交并比；
将所述锚框与所述真实框的交并比，通过非极大值抑制得到预测损失值集，所述预测损失值集可存储在区块链中的节点；
判断所述预测损失值集是否满足预设损失要求；
在所述预测损失值集不满足预设损失要求时，调整边框回归子网络中的损失函数的系数，继续对所述锚框及真实框进行边框回归，得到所述锚框与所述真实框的交并比；
在所述预测损失值集满足预设损失要求，得到并输出所述标准目标物损伤程度分类模型。
详细地，所述全卷积神经网络包括输入层、多x多卷积层、池化层及1x1卷积层，可以对图片进行像素级的分类，进而达到识别图片中特定部分的目的。
其中，所述输入层用于接收输入的图片并对接收的输入的图片进行标准化处理，如将输入的图片中的像素的像素值由[0，255]的范围归一化至[0,1]区间。
所述多x多卷积层包括卷积核、卷积步长和填充，用于对输入的图片进行特征提取得到特征图。其中，所述卷积核越大，可提取的图片的特征越复杂；所述卷积步长定义了卷积核相邻两次扫过所述图片时位置的距离，如卷积步长为1时，卷积核会逐个扫过输入的所述图片的像素，步长为n时会在下一次扫描跳过n-1个像素。所述填充用于在输入的所述图片通过卷积核之前，增大其尺寸以抵消计算中尺寸收缩产生的影响，如16×16的图片在经过单位步长的5×5的卷积核之前先进行相同填充，则会在水平和垂直方向填充两层，即两侧各增加2个像素变为20×20大小的图像，通过卷积核后，输出的特征图尺寸为16×16，保持了原本的尺寸。
所述池化层用于对所述多x多卷积层提取的特征进行进一步地特征选择和过滤。
所述1×1卷积层用于将图片调整为任意大小尺寸并输出。
进一步地，本申请优选实施例中所述原始目标物损伤程度分类模型中的全卷积神经网络还包含分类子网络和边框回归子网络。其中，所述分类自网络用于检测所述图片中的目标物得到真实框和一系列锚框，及所述边框回归子网络用于对所述图片中的一系列锚框与真实框进行边框回归。所述分类子网络和边框回归子网络均由1个有锚框分支和1个无锚框分支构成，每个有锚框分支分别由4层卷积层组成。所述分类子网络中的有锚框分支输出的特征图数量为类别数与锚框数的乘积，所述边框回归子网络中的有锚框分支输出的特征图数量为锚框数的4倍。其中，所述锚框是指，以像素点为中心，在输入图片中采集大量的区域，产生的预测边界框；所述边框回归是指目标检测过程中对产生的所述预测边界框以标注好的所述真实边界框为目标进行逼近的过程，然后判断这些所述预测边界框中是否包含目标，并调整所述预测边界框边缘从而更准确预测目标。
较佳地，本申请实施例所述分类子网络和边框回归子网络包括损失函数。
所述损失函数是指全卷积神经网络中每次迭代的前向计算结果与真实值的差距，从而指导下一步的训练向正确的方向进行的函数。本申请实施例中，所述损失函数包括焦点损失函数(Focal Loss)和交并比损失函数(IoU Loss)，公式分别为：
Figure PCTCN2020098985-appb-000006
Figure PCTCN2020098985-appb-000007
其中，所述Focal Loss和IoU Loss分别为所述分类子网络的和边框回归子网络的损失函数。其中，
Figure PCTCN2020098985-appb-000008
表示分类子网络的损失函数，
Figure PCTCN2020098985-appb-000009
表示为边框回归子网络的损失函数，l表示为卷积层数，i表示类别数，j表示锚框数,
Figure PCTCN2020098985-appb-000010
表示为损失系数。
优选地，本申请边框回归模块102通过下述操作得到标准目标物损伤程度分类模型：将所述预测损失值集中内每个损失值与预设的损失值阈值进行比较；计算大于所述预设的损失值阈值的损失值的数量及小于或者等于所述预设的损失值阈值的损失值的数量的比例值；在所述比例值大于预设的比例阈值时，执行所述调整边框回归子网络中预设的损失函数的系数；在所述比例值小于或者等于所述预设的比例阈值时，直接根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
所述鉴定模块103利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
本申请实施例接收所述目标物的受损图像集，并对所述受损图像集进行预处理后传输至所述标准目标物损伤程度分类模型中，得到目标物损伤程度鉴定结果。
详细地，本申请实施例中所述预处理包括：提取所述受损图像集中所有图片的预设种类的特征，如，提取所述预设种类的图片视角，包括正视图、俯视图、仰视图、左侧视图和右侧视图。
进一步地，本申请实施例将所述受损图像集传输至所述标准目标物损伤程度分类模型中，得到多个目标物损伤程度鉴定结果，并根据所述多个目标物损伤程度鉴定结果，得到最终的目标物损伤程度鉴定结果。
例如，本申请实施例将目标物，如车辆的同一受损部位的正视、俯视、仰视、左侧视和右侧视图图片输入至所述标准目标物损伤程度分类模型，利用所述标准目标物损伤程度分类模型分析得到同一受损部位的每种图片视角对应的损伤程度鉴定结果，组成目标物损伤程度鉴定结果集。在所述目标物损伤程度鉴定结果集判断是否有一种鉴定结果的比例超过预设比例值，若鉴定结果的比例超过预设比例值，则此次的损伤程度鉴定成功，最终的目标物损伤程度鉴定结果为超过预设比例值鉴定结果。若在所述目标物损伤程度鉴定结果集中没有一种鉴定结果的比例超过预设比例值，则此次的损伤程度鉴定失败。
例如，预设比例值为60％，用户将受损车辆的同一受损部位的正视图、俯视图、仰视图、左侧视图和右侧视图输入至所述标准目标物损伤程度分类模型中，输出的目标物损伤程度鉴定结果集如下：正视图、损伤级别B，俯视、损伤级别B，仰视图、损伤级别B，左侧视图、损伤级别C，右侧视图，损伤级别B，同一鉴定结果的比例为80％，大于预设比例值，则输出鉴定结果为损伤级别B。
如图4所示，是本申请实现目标物损伤程度判定方法的电子设备的结构示意图。
所述电子设备1可以包括处理器10、存储器11和总线，还可以包括存储在所述存储器11中并可在所述处理器10上运行的计算机程序，如基于图片的目标识别程序12。
其中，所述存储器11至少包括一种类型的可读存储介质，所述可读存储介质可以是非易失性，也可以是易失性，所述可读存储介质包括闪存、移动硬盘、多媒体卡、卡型存储器(例如：SD或DX存储器等)、磁性存储器、磁盘、光盘等。所述存储器11在一些实施例中可以是电子设备1的内部存储单元，例如该电子设备1的移动硬盘。所述存储器11在另一些实施例中也可以是电子设备1的外部存储设备，例如电子设备1上配备的插接式 移动硬盘、智能存储卡(Smart Media Card，SMC)、安全数字(Secure Digital，SD)卡、闪存卡(Flash Card)等。进一步地，所述存储器11还可以既包括电子设备1的内部存储单元也包括外部存储设备。所述存储器11不仅可以用于存储安装于电子设备1的应用软件及各类数据，例如基于图片的目标识别程序12的代码等，还可以用于暂时地存储已经输出或者将要输出的数据。
所述处理器10在一些实施例中可以由集成电路组成，例如可以由单个封装的集成电路所组成，也可以是由多个相同功能或不同功能封装的集成电路所组成，包括一个或者多个中央处理器(Central Processing unit，CPU)、微处理器、数字处理芯片、图形处理器及各种控制芯片的组合等。所述处理器10是所述电子设备的控制核心(Control Unit)，利用各种接口和线路连接整个电子设备的各个部件，通过运行或执行存储在所述存储器11内的程序或者模块(例如执行基于图片的目标识别程序等)，以及调用存储在所述存储器11内的数据，以执行电子设备1的各种功能和处理数据。
所述总线可以是外设部件互连标准(peripheral component interconnect，简称PCI)总线或扩展工业标准结构(extended industry standard architecture，简称EISA)总线等。该总线可以分为地址总线、数据总线、控制总线等。所述总线被设置为实现所述存储器11以及至少一个处理器10等之间的连接通信。
图3仅示出了具有部件的电子设备，本领域技术人员可以理解的是，图3示出的结构并不构成对所述电子设备1的限定，可以包括比图示更少或者更多的部件，或者组合某些部件，或者不同的部件布置。
例如，尽管未示出，所述电子设备1还可以包括给各个部件供电的电源(比如电池)，优选地，电源可以通过电源管理装置与所述至少一个处理器10逻辑相连，从而通过电源管理装置实现充电管理、放电管理、以及功耗管理等功能。电源还可以包括一个或一个以上的直流或交流电源、再充电装置、电源故障检测电路、电源转换器或者逆变器、电源状态指示器等任意组件。所述电子设备1还可以包括多种传感器、蓝牙模块、Wi-Fi模块等，在此不再赘述。
进一步地，所述电子设备1还可以包括网络接口，可选地，所述网络接口可以包括有线接口和/或无线接口(如WI-FI接口、蓝牙接口等)，通常用于在该电子设备1与其他电子设备之间建立通信连接。
可选地，该电子设备1还可以包括用户接口，用户接口可以是显示器(Display)、输入单元(比如键盘(Keyboard))，可选地，用户接口还可以是标准的有线接口、无线接口。可选地，在一些实施例中，显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode，有机发光二极管)触摸器等。其中，显示器也可以适当的称为显示屏或显示单元，用于显示在电子设备1中处理的信息以及用于显示可视化的用户界面。
应该了解，所述实施例仅为说明之用，在专利申请范围上并不受此结构的限制。
所述电子设备1中的所述存储器11存储的基于图片的目标识别程序12是多个指令的组合，在所述处理器10中运行时，可以实现：
获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
构建目标物损伤程度分类模型，其中，所述目标物损伤程度分类模型包括全卷积神经网络，所述全卷积神经网络包括分类子网络和边框回归子网络；
将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
具体地，所述处理器10对上述指令的具体实现方法可参考图1对应实施例中相关步骤的描述，在此不赘述。
进一步地，所述电子设备1集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个非易失性或易失性计算机可读取存储介质中。所述计算机可读介质可以包括：能够携带所述计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器(ROM，Read-Only Memory)。
在本申请所提供的几个实施例中，应该理解到，所揭露的设备，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述模块的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式。
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的，作为模块显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能模块可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用硬件加软件功能模块的形式实现。
对于本领域技术人员而言，显然本申请不限于上述示范性实施例的细节，而且在不背离本申请的精神或基本特征的情况下，能够以其他的具体形式实现本申请。
因此，无论从哪一点来看，均应将实施例看作是示范性的，而且是非限制性的，本申请的范围由所附权利要求而不是上述说明限定，因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本申请内。不应将权利要求中的任何附关联图标记视为限制所涉及的权利要求。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
此外，显然“包括”一词不排除其他单元或步骤，单数不排除复数。系统权利要求中陈述的多个单元或装置也可以由一个单元或装置通过软件或者硬件来实现。第二等词语用来表示名称，而并不表示任何特定的顺序。
最后应说明的是，以上实施例仅用以说明本申请的技术方案而非限制，尽管参照较佳实施例对本申请进行了详细说明，本领域的普通技术人员应当理解，可以对本申请的技术方案进行修改或等同替换，而不脱离本申请技术方案的精神和范围。

Claims (20)

  1. 一种目标物损伤程度鉴定方法，其中，所述方法包括：
    获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
    构建目标物损伤程度分类模型，其中，所述目标物损伤程度分类模型包括全卷积神经网络，所述全卷积神经网络包括分类子网络和边框回归子网络；
    将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
    利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
  2. 如权利要求1所述的目标物损伤程度判定方法，其中，所述根据所述预测损失值集得到标准目标物损伤程度分类模型，包括：
    在所述预测损失值集不满足预设损失要求时，调整边框回归子网络中预设的损失函数的系数后重新通过所述边框回归子网络对所述锚框与真实框进行边框回归，直到所述预测损失值集满足所述预设损失要求，根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
  3. 如权利要求2所述的目标物损伤程度判定方法，其中，所述在所述预测损失值集不满足预设损失要求时，该方法还包括：
    将所述预测损失值集中内每个损失值与预设的损失值阈值进行比较；
    计算大于所述预设的损失值阈值的损失值的数量与小于或者等于所述预设的损失值阈值的损失值的数量之间的比例值；
    在所述比例值大于预设的比例阈值时，执行所述调整边框回归子网络中预设的损失函数的系数；
    在所述比例值小于或者等于所述预设的比例阈值时，直接根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
  4. 如权利要求1所述的目标物损伤程度判定方法，其中，所述利用所述分类子网络生成所述标准图片训练集内每个图片的锚框，包括：
    在所述分类子网络中获取所述标准图片训练集内的图片，检测所述图片中的目标物得到真实框，并对所述图片中的所有像素点进行随机采样，得到一系列以像素点为中心的锚框。
  5. 如权利要求3所述的目标物损伤程度判定方法，其中，所述通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，包括：
    利用所述边框回归子网络中预设的损失函数对所述图片中的一系列锚框与真实框进行边框回归，使所述锚框向所述真实框不断逼近，得到所述锚框与所述真实框的交并比，根据所述交并比得到所述预测损失值集。
  6. 如权利要求5所述的目标物损伤程度判定方法，其中，所述边框回归子网络中预设的损失函数为：
    Figure PCTCN2020098985-appb-100001
    其中，
    Figure PCTCN2020098985-appb-100002
    表示边框回归子网络的损失值，l表示卷积层数，i表示类别数，j表示锚框数,
    Figure PCTCN2020098985-appb-100003
    表示损失系数。
  7. 一种目标物损伤程度判定装置，其中，所述装置包括：
    分类模块，用于获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
    边框回归模块，用于将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
    鉴定模块，用于利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
  8. 如权利要求7所述的目标物损伤程度判定装置，其中，所述边框回归模块通过下述操作得到标准目标物损伤程度分类模型：
    将所述预测损失值集中内每个损失值与预设的损失值阈值进行比较；
    计算大于所述预设的损失值阈值的损失值的数量与小于或者等于所述预设的损失值阈值的损失值的数量之间的比例值；
    在所述比例值大于预设的比例阈值时，执行所述调整边框回归子网络中预设的损失函数的系数；
    在所述比例值小于或者等于所述预设的比例阈值时，直接根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
  9. 如权利要求7所述的目标物损伤程度判定装置，其中，所述边框回归模块通过执行下述操作得到所述预测损失值集：
    利用所述边框回归子网络中预设的损失函数对所述图片中的一系列锚框及真实框进行边框回归，使所述锚框向所述真实框不断逼近，得到所述锚框与所述真实框的交并比，根据所述交并比得到所述预测损失值集。
  10. 一种电子设备，其中，所述电子设备包括：
    至少一个处理器；以及，
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行时实现如下步骤：
    获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
    构建目标物损伤程度分类模型，其中，所述目标物损伤程度分类模型包括全卷积神经网络，所述全卷积神经网络包括分类子网络和边框回归子网络；
    将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
    利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
  11. 如权利要求10所述的电子设备，其中，所述根据所述预测损失值集得到标准目标物损伤程度分类模型，包括：
    在所述预测损失值集不满足预设损失要求时，调整边框回归子网络中预设的损失函数的系数后重新通过所述边框回归子网络对所述锚框与真实框进行边框回归，直到所述预测损失值集满足所述预设损失要求，根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
  12. 如权利要求11所述的电子设备，其中，所述在所述预测损失值集不满足预设损失要求时，所述指令被所述至少一个处理器执行时还实现如下步骤：
    将所述预测损失值集中内每个损失值与预设的损失值阈值进行比较；
    计算大于所述预设的损失值阈值的损失值的数量与小于或者等于所述预设的损失值阈值的损失值的数量之间的比例值；
    在所述比例值大于预设的比例阈值时，执行所述调整边框回归子网络中预设的损失函数的系数；
    在所述比例值小于或者等于所述预设的比例阈值时，直接根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
  13. 如权利要求10所述的电子设备，其中，所述利用所述分类子网络生成所述标准图片训练集内每个图片的锚框，包括：
    在所述分类子网络中获取所述标准图片训练集内的图片，检测所述图片中的目标物得到真实框，并对所述图片中的所有像素点进行随机采样，得到一系列以像素点为中心的锚框。
  14. 如权利要求12所述的电子设备，其中，所述通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，包括：
    利用所述边框回归子网络中预设的损失函数对所述图片中的一系列锚框与真实框进行边框回归，使所述锚框向所述真实框不断逼近，得到所述锚框与所述真实框的交并比，根据所述交并比得到所述预测损失值集。
  15. 如权利要求14所述的电子设备，其中，所述边框回归子网络中预设的损失函数为：
    Figure PCTCN2020098985-appb-100004
    其中，
    Figure PCTCN2020098985-appb-100005
    表示边框回归子网络的损失值，l表示卷积层数，i表示类别数，j表示锚框数,
    Figure PCTCN2020098985-appb-100006
    表示损失系数。
  16. 一种计算机可读存储介质，存储有计算机程序，其中，所述计算机程序被处理器执行时实现如下步骤：
    获取原始图片训练集及所述原始图片训练集对应的标签集，将所述原始图片训练集中的图片按照对应标签进行分类，得到标准图片训练集；
    构建目标物损伤程度分类模型，其中，所述目标物损伤程度分类模型包括全卷积神经网络，所述全卷积神经网络包括分类子网络和边框回归子网络；
    将所述标准图片训练集输入至所述全卷积神经网络，并利用所述分类子网络生成所述标准图片训练集内每个图片的锚框及真实框，通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，根据所述预测损失值集得到标准目标物损伤程度分类模型；
    利用所述标准目标物损伤程度分类模型对目标物的受损图像集进行分析，得到目标物损伤程度鉴定结果。
  17. 如权利要求16所述的计算机可读存储介质，其中，所述根据所述预测损失值集得到标准目标物损伤程度分类模型，包括：
    在所述预测损失值集不满足预设损失要求时，调整边框回归子网络中预设的损失函数的系数后重新通过所述边框回归子网络对所述锚框与真实框进行边框回归，直到所述预测损失值集满足所述预设损失要求，根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
  18. 如权利要求17所述的计算机可读存储介质，其中，所述在所述预测损失值集不满足预设损失要求时，所述计算机程序被处理器执行时还实现如下步骤：
    将所述预测损失值集中内每个损失值与预设的损失值阈值进行比较；
    计算大于所述预设的损失值阈值的损失值的数量与小于或者等于所述预设的损失值 阈值的损失值的数量之间的比例值；
    在所述比例值大于预设的比例阈值时，执行所述调整边框回归子网络中预设的损失函数的系数；
    在所述比例值小于或者等于所述预设的比例阈值时，直接根据边框回归子网络中当前的损失函数的系数，得到所述标准目标物损伤程度分类模型。
  19. 如权利要求16所述的计算机可读存储介质，其中，所述利用所述分类子网络生成所述标准图片训练集内每个图片的锚框，包括：
    在所述分类子网络中获取所述标准图片训练集内的图片，检测所述图片中的目标物得到真实框，并对所述图片中的所有像素点进行随机采样，得到一系列以像素点为中心的锚框。
  20. 如权利要求18所述的计算机可读存储介质，其中，所述通过所述边框回归子网络对所述锚框与真实框进行边框回归，得到预测损失值集，包括：
    利用所述边框回归子网络中预设的损失函数对所述图片中的一系列锚框与真实框进行边框回归，使所述锚框向所述真实框不断逼近，得到所述锚框与所述真实框的交并比，根据所述交并比得到所述预测损失值集。
PCT/CN2020/098985 2020-05-26 2020-06-29 目标物损伤程度判定方法、装置、电子设备及存储介质 WO2021151277A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010458345.7 2020-05-26
CN202010458345.7A CN111695609B (zh) 2020-05-26 2020-05-26 目标物损伤程度判定方法、装置、电子设备及存储介质

Publications (1)

Publication Number Publication Date
WO2021151277A1 true WO2021151277A1 (zh) 2021-08-05

Family

ID=72478395

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/098985 WO2021151277A1 (zh) 2020-05-26 2020-06-29 目标物损伤程度判定方法、装置、电子设备及存储介质

Country Status (2)

Country Link
CN (1) CN111695609B (zh)
WO (1) WO2021151277A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113837316A (zh) * 2021-10-13 2021-12-24 山东德普检测技术有限公司 一种基于农产品的异常区域检测方法、装置、设备及介质
CN113887455A (zh) * 2021-10-11 2022-01-04 东北大学 一种基于改进fcos的人脸口罩检测系统及方法
CN113963167A (zh) * 2021-10-29 2022-01-21 北京百度网讯科技有限公司 应用于目标检测的方法、装置及计算机程序产品
CN115063631A (zh) * 2022-06-20 2022-09-16 平安科技（深圳）有限公司 基于人工智能的车损检测方法及相关设备

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112329596B (zh) * 2020-11-02 2021-08-24 中国平安财产保险股份有限公司 目标物定损方法、装置、电子设备及计算机可读存储介质
CN114581652A (zh) * 2020-12-01 2022-06-03 北京四维图新科技股份有限公司 目标对象的检测方法、装置、电子设备以及存储介质
CN114120122B (zh) * 2021-11-29 2024-04-30 中国平安财产保险股份有限公司 基于遥感影像的灾损识别方法、装置、设备及存储介质
CN116630716B (zh) * 2023-06-06 2024-05-24 云途信息科技(杭州)有限公司 道路绿化破损识别方法、装置、计算机设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109657716A (zh) * 2018-12-12 2019-04-19 天津卡达克数据有限公司 一种基于深度学习的车辆外观损伤识别方法
CN109684956A (zh) * 2018-12-14 2019-04-26 深源恒际科技有限公司 一种基于深度神经网络的车辆损伤检测方法及系统
US20190332898A1 (en) * 2018-04-26 2019-10-31 Canon Kabushiki Kaisha Information processing apparatus for training neural network for recognition task and method thereof
CN110399884A (zh) * 2019-07-10 2019-11-01 浙江理工大学 一种特征融合自适应锚框模型车辆检测方法
CN110570316A (zh) * 2018-08-31 2019-12-13 阿里巴巴集团控股有限公司 训练损伤识别模型的方法及装置
CN111062413A (zh) * 2019-11-08 2020-04-24 深兰科技（上海）有限公司 一种道路目标检测方法、装置、电子设备及存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108038474B (zh) * 2017-12-28 2020-04-14 深圳励飞科技有限公司 人脸检测方法、卷积神经网络参数的训练方法、装置及介质
CN109117831B (zh) * 2018-09-30 2021-10-12 北京字节跳动网络技术有限公司 物体检测网络的训练方法和装置
CN111160379B (zh) * 2018-11-07 2023-09-15 北京嘀嘀无限科技发展有限公司 图像检测模型的训练方法及装置、目标检测方法及装置
CN110942000B (zh) * 2019-11-13 2022-09-06 南京理工大学 一种基于深度学习的无人驾驶车辆目标检测方法
CN111027547B (zh) * 2019-12-06 2022-08-09 南京大学 一种针对二维图像中的多尺度多形态目标的自动检测方法
CN111091105B (zh) * 2019-12-23 2020-10-20 郑州轻工业大学 基于新的边框回归损失函数的遥感图像目标检测方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190332898A1 (en) * 2018-04-26 2019-10-31 Canon Kabushiki Kaisha Information processing apparatus for training neural network for recognition task and method thereof
CN110570316A (zh) * 2018-08-31 2019-12-13 阿里巴巴集团控股有限公司 训练损伤识别模型的方法及装置
CN109657716A (zh) * 2018-12-12 2019-04-19 天津卡达克数据有限公司 一种基于深度学习的车辆外观损伤识别方法
CN109684956A (zh) * 2018-12-14 2019-04-26 深源恒际科技有限公司 一种基于深度神经网络的车辆损伤检测方法及系统
CN110399884A (zh) * 2019-07-10 2019-11-01 浙江理工大学 一种特征融合自适应锚框模型车辆检测方法
CN111062413A (zh) * 2019-11-08 2020-04-24 深兰科技（上海）有限公司 一种道路目标检测方法、装置、电子设备及存储介质

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113887455A (zh) * 2021-10-11 2022-01-04 东北大学 一种基于改进fcos的人脸口罩检测系统及方法
CN113887455B (zh) * 2021-10-11 2024-05-28 东北大学 一种基于改进fcos的人脸口罩检测系统及方法
CN113837316A (zh) * 2021-10-13 2021-12-24 山东德普检测技术有限公司 一种基于农产品的异常区域检测方法、装置、设备及介质
CN113837316B (zh) * 2021-10-13 2023-12-26 山东德普检测技术有限公司 一种基于农产品的异常区域检测方法、装置、设备及介质
CN113963167A (zh) * 2021-10-29 2022-01-21 北京百度网讯科技有限公司 应用于目标检测的方法、装置及计算机程序产品
CN113963167B (zh) * 2021-10-29 2022-05-27 北京百度网讯科技有限公司 应用于目标检测的方法、装置及计算机程序产品
CN115063631A (zh) * 2022-06-20 2022-09-16 平安科技（深圳）有限公司 基于人工智能的车损检测方法及相关设备

Also Published As

Publication number Publication date
CN111695609A (zh) 2020-09-22
CN111695609B (zh) 2024-06-28

Similar Documents

Publication Publication Date Title
WO2021151277A1 (zh) 目标物损伤程度判定方法、装置、电子设备及存储介质
CN108229509B (zh) 用于识别物体类别的方法及装置、电子设备
CN111652845B (zh) 异常细胞自动标注方法、装置、电子设备及存储介质
CN109583483B (zh) 一种基于卷积神经网络的目标检测方法和系统
WO2022247005A1 (zh) 图像中目标物识别方法、装置、电子设备及存储介质
CN112507934B (zh) 活体检测方法、装置、电子设备及存储介质
CN112137591B (zh) 基于视频流的目标物位置检测方法、装置、设备及介质
WO2021151313A1 (zh) 证件鉴伪方法、装置、电子设备及存储介质
WO2020253508A1 (zh) 异常细胞检测方法、装置及计算机可读存储介质
WO2021189856A1 (zh) 证件校验方法、装置、电子设备及介质
CN113222063B (zh) 快递纸箱垃圾分类方法、装置、设备及介质
CN116168351B (zh) 电力设备巡检方法及装置
CN112749653A (zh) 行人检测方法、装置、电子设备及存储介质
CN113627394B (zh) 人脸提取方法、装置、电子设备及可读存储介质
CN111985449A (zh) 救援现场图像的识别方法、装置、设备及计算机介质
CN111583215B (zh) 损伤图像智能定损方法、装置、电子设备及存储介质
CN114463685B (zh) 行为识别方法、装置、电子设备及存储介质
CN113657316B (zh) 移动交通违章监测方法、系统、电子设备及存储介质
CN113792671B (zh) 人脸合成图像的检测方法、装置、电子设备及介质
CN113792801B (zh) 人脸炫光程度的检测方法、装置、设备及存储介质
CN112541899B (zh) 证件的残缺检测方法、装置、电子设备及计算机存储介质
CN114049676A (zh) 疲劳状态检测方法、装置、设备及存储介质
WO2021217858A1 (zh) 基于图片的目标识别方法、装置、电子设备及可读存储介质
CN114708230A (zh) 基于图像分析的车架质量检测方法、装置、设备及介质
CN112561889A (zh) 目标物检测方法、装置、电子设备及存储介质

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: 20916798

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: 20916798

Country of ref document: EP

Kind code of ref document: A1