CN112560956A - Target detection method and device, nonvolatile storage medium and electronic equipment - Google Patents

Target detection method and device, nonvolatile storage medium and electronic equipment Download PDF

Info

Publication number
CN112560956A
CN112560956A CN202011493902.5A CN202011493902A CN112560956A CN 112560956 A CN112560956 A CN 112560956A CN 202011493902 A CN202011493902 A CN 202011493902A CN 112560956 A CN112560956 A CN 112560956A
Authority
CN
China
Prior art keywords
layer
target detection
feature
feature layer
detection model
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN202011493902.5A
Other languages
Chinese (zh)
Inventor
蒋泽新
宋明岑
杨智慧
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Gree Electric Appliances Inc of Zhuhai
Zhuhai Gree Intelligent Equipment Co Ltd
Original Assignee
Gree Electric Appliances Inc of Zhuhai
Zhuhai Gree Intelligent Equipment Co Ltd
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 Gree Electric Appliances Inc of Zhuhai, Zhuhai Gree Intelligent Equipment Co Ltd filed Critical Gree Electric Appliances Inc of Zhuhai
Priority to CN202011493902.5A priority Critical patent/CN112560956A/en
Publication of CN112560956A publication Critical patent/CN112560956A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • 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
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • 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

Landscapes

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

Abstract

本发明公开了一种目标检测方法及装置、非易失性存储介质、电子设备。其中，该方法包括：获取目标检测模型的多个特征层，其中，上述多个特征层包括：第一特征层和第二特征层，上述第一特征层的语义信息少于上述第二特征层的语义信息；将上述多个特征层中的上述第一特征层与上述第二特征层进行融合处理，得到目标检测层，上述融合处理用于增强上述第一特征层的语义信息；基于上述目标检测层更新上述目标检测模型。本发明解决了现有技术中传统的目标检测模型的检测精度较低的技术问题。

Figure 202011493902

The invention discloses a target detection method and device, a non-volatile storage medium and electronic equipment. The method includes: acquiring multiple feature layers of a target detection model, wherein the multiple feature layers include: a first feature layer and a second feature layer, and the first feature layer has less semantic information than the second feature layer. The semantic information of the above-mentioned first characteristic layer and the above-mentioned second characteristic layer in the above-mentioned multiple characteristic layers are fused to obtain a target detection layer, and the above-mentioned fusion processing is used to enhance the semantic information of the above-mentioned first characteristic layer; based on the above-mentioned target The detection layer updates the above object detection model. The invention solves the technical problem of low detection accuracy of the traditional target detection model in the prior art.

Figure 202011493902

Description

Target detection method and device, nonvolatile storage medium and electronic equipment
Technical Field
The invention relates to the field of target detection, in particular to a target detection method and device, a nonvolatile storage medium and electronic equipment.
Background
In object detection, the classical target detection framework often does not perform well for small target detection tasks. Aiming at the problem of small target detection in an industrial environment, how to improve a classic target detection framework and improve the detection precision of small-scale targets is not provided at present.
Disclosure of Invention
The embodiment of the invention provides a target detection method and device, a nonvolatile storage medium and electronic equipment, which at least solve the technical problem of low detection precision of a traditional target detection model in the prior art.
According to an aspect of an embodiment of the present invention, there is provided a target detection method, including: obtaining a plurality of feature layers of a target detection model, wherein the plurality of feature layers comprise: the semantic information of the first characteristic layer is less than that of the second characteristic layer; performing fusion processing on the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer, wherein the fusion processing is used for enhancing semantic information of the first feature layer; and updating the target detection model based on the target detection layer.
Optionally, the fusing the first feature layer and the second feature layer in the plurality of feature layers to obtain the target detection layer includes: amplifying the first scale of the first characteristic layer into a second scale by adopting an interpolation mode, wherein the second scale is equal to the scale of a second characteristic layer adjacent to the first characteristic layer; adding the processed first characteristic layer and the second characteristic layer to obtain a new first characteristic layer; amplifying the third scale of the new first feature layer into a fourth scale by adopting an interpolation mode, wherein the fourth scale is equal to the scale of a new second feature layer adjacent to the new first feature layer; and adding the processed new first feature layer and the new second feature layer until the fusion processing of all the first feature layers and the second feature layers in the plurality of feature layers is finished, thereby obtaining the target detection layer.
Optionally, the method further includes: acquiring a real receptive field of the target detection model; adjusting an anchor frame of the target detection model based on the real receptive field, wherein the size of the anchor frame affects the anchor frame classification and the anchor frame regression of the target detection model; determining the number of samples of the target detection model according to the anchor frame, wherein the number of samples comprises: the number of positive samples and the number of negative samples.
Optionally, determining the number of samples of the target detection model according to the anchor frame includes: determining a loss function of the target detection model according to the anchor frame, wherein the loss function includes: a classification loss function and a regression loss function; the number of samples is determined based on the loss function.
Optionally, determining a loss function L of the target detection model according to the anchor frame by using the following calculation formula, including:
Figure BDA0002841505130000021
where i denotes the index of the anchor frame, PiRepresenting the prediction probability, P, of the above-mentioned anchor frame as the targeti *A true value indicating that the anchor frame is a target, and P is a true value when the anchor frame is the targeti *Is 1, otherwise Pi *Is 0; t is tiIndicating the predicted detection frame coordinate correction value,
Figure BDA0002841505130000022
representing the actual coordinate value of the detection frame; pi *LrRepresenting the calculation of the regression loss, N, against the anchor box of the sample onlycAnd NrRespectively representing the number of positive and negative anchor frames and the number of regressed positive anchor frames during classification, and lambda represents a balance parameter for balancing classification loss and regression loss.
Optionally, the classification loss function is a Softmax loss function, and the regression loss function is a Smooth-L1 loss function.
According to another aspect of the embodiments of the present invention, there is also provided an object detection apparatus, including: an obtaining module, configured to obtain a plurality of feature layers of a target detection model, where the plurality of feature layers include: the semantic information of the first characteristic layer is less than that of the second characteristic layer; a fusion processing module, configured to perform fusion processing on the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer, where the fusion processing is used to enhance semantic information of the first feature layer; and the updating module is used for updating the target detection model based on the target detection layer.
According to another aspect of the embodiments of the present invention, there is also provided a non-volatile storage medium storing a plurality of instructions, the instructions being adapted to be loaded by a processor and to perform any one of the above object detection methods.
According to another aspect of the embodiments of the present invention, there is also provided a processor, configured to execute a program, where the program is configured to execute any one of the above object detection methods when running.
According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including a memory and a processor, where the memory stores a computer program, and the processor is configured to execute the computer program to perform any one of the above object detection methods.
In the embodiment of the present invention, a plurality of feature layers of a target detection model are obtained, where the plurality of feature layers include: the semantic information of the first characteristic layer is less than that of the second characteristic layer; performing fusion processing on the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer, wherein the fusion processing is used for enhancing semantic information of the first feature layer; and updating the target detection model based on the target detection layer.
The first characteristic layer and the second characteristic layer of the target detection model are subjected to recursive fusion processing, the semantic features of the first characteristic layer are enhanced, and the purpose of improving the detection accuracy of the target detection model is achieved, so that the technical effect of enhancing the detection accuracy of the target detection model is achieved, and the technical problem that the detection accuracy of the traditional target detection model is low in the prior art is solved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:
FIG. 1 is a flow chart of a method of target detection according to an embodiment of the present invention;
FIG. 2 is a flow diagram of an alternative target detection method according to an embodiment of the invention;
fig. 3 is a schematic diagram of an implementation scenario of an optional fusion process performed on feature layers according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an object detection apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Example 1
In accordance with an embodiment of the present invention, there is provided an embodiment of an object detection method, it should be noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system such as a set of computer-executable instructions, and that while a logical order is illustrated in the flowchart, in some cases the steps illustrated or described may be performed in an order different than here.
Fig. 1 is a flowchart of an object detection method according to an embodiment of the present invention, as shown in fig. 1, the method includes the following steps:
step S102, obtaining a plurality of characteristic layers of a target detection model, wherein the plurality of characteristic layers comprise: the semantic information of the first characteristic layer is less than that of the second characteristic layer;
step S104, performing fusion processing on the first characteristic layer and the second characteristic layer in the plurality of characteristic layers to obtain a target detection layer, wherein the fusion processing is used for enhancing semantic information of the first characteristic layer;
step S106, updating the target detection model based on the target detection layer.
In the embodiment of the present invention, a plurality of feature layers of a target detection model are obtained, where the plurality of feature layers include: the semantic information of the first characteristic layer is less than that of the second characteristic layer; performing fusion processing on the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer, wherein the fusion processing is used for enhancing semantic information of the first feature layer; and updating the target detection model based on the target detection layer.
The shallow characteristic layer of the target detection model is used as a detection layer, the shallow characteristic layer and the deep characteristic layer are subjected to recursive fusion processing, the semantic characteristics of the shallow characteristic layer are enhanced, and the purpose of improving the detection accuracy of the target detection model is achieved, so that the technical effect of enhancing the detection accuracy of the target detection model is achieved, and the technical problem that the detection accuracy of the traditional target detection model is low in the prior art is solved.
Optionally, the target detection model may be any type of target detection model, for example, a Faster R-CNN target detection model, an SDD target detection model, a YOLO target detection model, and the like.
Taking the above target detection model as an SSD target detection model as an example, the SSD target detection model is a single-stage, efficient and general object detection framework. The effect of detecting the target in real time can be achieved while the precision is ensured, but when the SSD target detection model is used for detecting the small target, the detection effect is not ideal. According to analysis, the SSD target detection model is too deep in a feature detection layer, and the downsampling multiple is too large, so that feature information of a small target is lost. Secondly, the SSD anchor frame is laid at intervals which are too large, so that small targets are easily not in the anchor frame, and the small targets are also missed to be inspected.
In the embodiment of the application, aiming at the problem of low recall rate of small target detection, a shallow feature layer is used as a detection layer, and the shallow feature layer and a deep feature layer are fused to enhance semantic feature information; aiming at the problem that the positive and negative samples are extremely unbalanced in small target detection, a negative sample screening strategy is added, and the number of the positive and negative samples is balanced; the shallow characteristic layer is used as a detection layer, the deep characteristic layer and the shallow characteristic layer are fused, positive and negative samples are balanced, and the interference of the negative samples on detection results in target detection is reduced.
Optionally, the plurality of feature layers include: the semantic information of the first characteristic layer is less than that of the second characteristic layer.
It should be noted that the shallow feature and the deep feature are relative, for example, when two adjacent features are compared, the lower feature belongs to the shallow feature, and the upper feature belongs to the deep feature, but another upper feature adjacent to the upper feature is a deep feature of the upper feature, and the upper feature belongs to the shallow feature relative to the other upper feature.
In an optional embodiment, fig. 2 is a flowchart of an optional target detection method according to an embodiment of the present invention, and as shown in fig. 2, the fusing the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer includes:
step S202, enlarging the first scale of the first characteristic layer into a second scale by adopting an interpolation mode, wherein the second scale is equal to the scale of a second characteristic layer adjacent to the first characteristic layer;
step S204, adding the processed first characteristic layer and the second characteristic layer to obtain a new first characteristic layer;
step S206, enlarging the third dimension of the new first feature layer to a fourth dimension by adopting an interpolation mode, wherein the fourth dimension is equal to the dimension of a new second feature layer adjacent to the new first feature layer;
step S208 is to add the processed new first feature layer and the new second feature layer until the fusion process of all the first feature layers and the second feature layers in the plurality of feature layers is completed, thereby obtaining the target detection layer.
In order to solve the problem of the prior art, in the embodiment of the present application, a detection layer may be added to a shallow feature layer, however, semantic information of the shallow feature layer is not rich enough, and it is difficult to identify an object, and in order to solve the problem of less semantic information of the shallow feature layer, semantic information of the shallow feature layer may be enhanced by feature fusion with a deep feature layer through a feature pyramid, and a specific fusion processing procedure is as shown in fig. 3, based on an input image, a plurality of feature layers Conv3_3, Conv4_3, Conv5_3, Conv6_2, and Conv7_2 in an SSD object detection model are obtained as feature pyramid layers, starting from Conv7_2, the feature layer size is enlarged to be the same as the previous layer (Conv6_2) by means of interpolation, and the two are added to obtain a new feature layer P1, the P1 is subjected to the above interpolation operation and added to the Conv5_3 to obtain P2, the above operations were repeated to obtain P3 and P4, and P4 was used as the target detection layer.
In an optional embodiment, the method further includes:
step S302, acquiring a real receptive field of the target detection model;
step S304, adjusting an anchor frame of the target detection model based on the real receptive field, wherein the size of the anchor frame influences the anchor frame classification and the anchor frame regression of the target detection model;
step S306, determining the number of samples of the target detection model according to the anchor frame, where the number of samples includes: the number of positive samples and the number of negative samples.
In the embodiment of the present application, the size of the anchor frame of the target detection model may be, but is not limited to, set to 16 according to the size of the actual receptive field being twenty to forty percent of the theoretical receptive field, so that the anchor frame can be classified and regressed more accurately.
It should be noted that, in the embodiment of the present application, although the anchor frame laid by the target detection model can cover a small target more comprehensively, a large number of negative samples are generated, so that the number of the negative samples is far higher than that of the positive samples, the positive samples and the negative samples are extremely unbalanced, and the network training result at this time is necessarily biased toward the negative samples. For example, if only one of the 100 anchor boxes is a positive sample (target) and the other 99 anchor boxes are negative samples (background), the accuracy of the embodiment of the present application can be as high as 99% if the detection result is directly set as the background.
In an alternative embodiment, determining the number of samples of the target detection model according to the anchor frame includes:
step S402, determining a loss function of the target detection model according to the anchor frame, wherein the loss function includes: a classification loss function and a regression loss function;
in step S404, the number of samples is determined based on the loss function.
Aiming at the problem of imbalance of the positive and negative samples, the loss function of the target detection model can be determined through an anchor frame and sequenced, and the number of the positive samples and the number of the negative samples are respectively selected according to the proportion of 3:1 of the positive samples and the negative samples.
In an alternative embodiment, the loss function includes two parts, one is a classification loss function of whether the anchor frame is a human face, and the other is a regression loss function of the anchor frame as the coordinate correction value of the detection frame of the human face.
In an alternative embodiment, the classification loss function is a Softmax loss function, and the regression loss function is a Smooth-L1 loss function.
In an alternative embodiment, determining the loss function L of the target detection model according to the anchor frame by using the following calculation formula includes:
Figure BDA0002841505130000071
where i denotes the index of the anchor frame, PiRepresenting the prediction probability, P, of the above-mentioned anchor frame as the targeti *A true value indicating that the anchor frame is a target, and P is a true value when the anchor frame is the targeti *Is 1, otherwise Pi *Is 0; t is tiIndicating the predicted detection frame coordinate correction value,
Figure BDA0002841505130000072
representing the actual coordinate value of the detection frame; pi *LrRepresenting the calculation of the regression loss, N, against the anchor box of the sample onlycAnd NrRespectively representing the number of positive and negative anchor frames and the number of regressed positive anchor frames during classification, and lambda represents a balance parameter for balancing classification loss and regression loss.
Aiming at the condition that the detection effect of the SSD target on the small target is not good, the shallow feature layer is used as a detection layer and is fused with the deep feature layer; adjusting the size of the anchor frame according to the real receptive field; and the number of the positive samples and the negative samples is balanced, so that the detection precision of the small target is improved.
Example 2
According to an embodiment of the present invention, there is further provided an apparatus embodiment for implementing the object detection method, and fig. 4 is a schematic structural diagram of an object detection apparatus according to an embodiment of the present invention, as shown in fig. 4, the object detection apparatus includes: an obtaining module 400, a fusion processing module 402, and an updating module 404, wherein:
an obtaining module 400, configured to obtain a plurality of feature layers of a target detection model, where the plurality of feature layers include: the semantic information of the first characteristic layer is less than that of the second characteristic layer; a fusion processing module 402, configured to perform fusion processing on the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer, where the fusion processing is used to enhance semantic information of the first feature layer; an updating module 404, configured to update the object detection model based on the object detection layer.
It should be noted that the above modules may be implemented by software or hardware, for example, for the latter, the following may be implemented: the modules can be located in the same processor; alternatively, the modules may be located in different processors in any combination.
It should be noted here that the above-mentioned obtaining module 400, the fusion processing module 402 and the updating module 404 correspond to steps S102 to S106 in embodiment 1, and the above-mentioned modules are the same as the examples and application scenarios realized by the corresponding steps, but are not limited to what is disclosed in embodiment 1. It should be noted that the modules described above may be implemented in a computer terminal as part of an apparatus.
It should be noted that, reference may be made to the relevant description in embodiment 1 for alternative or preferred embodiments of this embodiment, and details are not described here again.
The above object detection device may further include a processor and a memory, where the above obtaining module 400, the fusion processing module 402, the updating module 404, and the like are all stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions.
The processor comprises a kernel, and the kernel calls a corresponding program unit from the memory, wherein one or more than one kernel can be arranged. The memory may include volatile memory in a computer readable medium, Random Access Memory (RAM) and/or nonvolatile memory such as Read Only Memory (ROM) or flash memory (flash RAM), and the memory includes at least one memory chip.
According to an embodiment of the present application, there is also provided an embodiment of a non-volatile storage medium. Optionally, in this embodiment, the nonvolatile storage medium includes a stored program, and the apparatus in which the nonvolatile storage medium is located is controlled to execute the any one of the object detection methods when the program runs.
Optionally, in this embodiment, the nonvolatile storage medium may be located in any one of a group of computer terminals in a computer network, or in any one of a group of mobile terminals, and the nonvolatile storage medium includes a stored program.
Optionally, the apparatus in which the non-volatile storage medium is controlled to perform the following functions when the program is executed: obtaining a plurality of feature layers of a target detection model, wherein the plurality of feature layers comprise: the semantic information of the first characteristic layer is less than that of the second characteristic layer; performing fusion processing on the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer, wherein the fusion processing is used for enhancing semantic information of the first feature layer; and updating the target detection model based on the target detection layer.
Optionally, the apparatus in which the non-volatile storage medium is controlled to perform the following functions when the program is executed: amplifying the first scale of the first characteristic layer into a second scale by adopting an interpolation mode, wherein the second scale is equal to the scale of a second characteristic layer adjacent to the first characteristic layer; adding the processed first characteristic layer and the second characteristic layer to obtain a new first characteristic layer; amplifying the third scale of the new first feature layer into a fourth scale by adopting an interpolation mode, wherein the fourth scale is equal to the scale of a new second feature layer adjacent to the new first feature layer; and adding the processed new first feature layer and the new second feature layer until the fusion processing of all the first feature layers and the second feature layers in the plurality of feature layers is finished, thereby obtaining the target detection layer.
Optionally, the apparatus in which the non-volatile storage medium is controlled to perform the following functions when the program is executed: acquiring a real receptive field of the target detection model; adjusting an anchor frame of the target detection model based on the real receptive field, wherein the size of the anchor frame affects the anchor frame classification and the anchor frame regression of the target detection model; determining the number of samples of the target detection model according to the anchor frame, wherein the number of samples comprises: the number of positive samples and the number of negative samples.
Optionally, the apparatus in which the non-volatile storage medium is controlled to perform the following functions when the program is executed: determining a loss function of the target detection model according to the anchor frame, wherein the loss function includes: a classification loss function and a regression loss function; the number of samples is determined based on the loss function.
Optionally, the apparatus in which the non-volatile storage medium is controlled to perform the following functions when the program is executed: determining a loss function L of the target detection model according to the anchor frame by adopting the following calculation formula, wherein the calculation formula comprises the following steps:
Figure BDA0002841505130000091
where i denotes the index of the anchor frame, PiRepresenting the prediction probability, P, of the above-mentioned anchor frame as the targeti *A true value indicating that the anchor frame is a target, and P is a true value when the anchor frame is the targeti *Is 1, otherwise Pi *Is 0; t is tiIndicating the predicted detection frame coordinate correction value,
Figure BDA0002841505130000092
representing the actual coordinate value of the detection frame; pi *LrRepresenting the calculation of the regression loss, N, against the anchor box of the sample onlycAnd NrRespectively representing the number of positive and negative anchor frames and the number of regressed positive anchor frames during classification, and lambda represents a balance parameter for balancing classification loss and regression loss.
According to an embodiment of the present application, there is also provided an embodiment of a processor. Optionally, in this embodiment, the processor is configured to execute a program, where the program executes the any one of the object detection methods.
According to an embodiment of the present application, there is also provided an embodiment of an electronic device, including a memory and a processor, where the memory stores a computer program, and the processor is configured to execute the computer program to perform any one of the above object detection methods.
There is further provided, in accordance with an embodiment of the present application, an embodiment of a computer program product, which, when executed on a data processing device, is adapted to execute a program initialized with the steps of the object detection method of any one of the above.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the above-described division of the units may be a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or may be integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable non-volatile storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a non-volatile storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned nonvolatile storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims (10)

1.一种目标检测方法，其特征在于，包括：1. a target detection method, is characterized in that, comprises: 获取目标检测模型的多个特征层，其中，所述多个特征层包括：第一特征层和第二特征层，所述第一特征层的语义信息少于所述第二特征层的语义信息；Acquire multiple feature layers of the target detection model, wherein the multiple feature layers include: a first feature layer and a second feature layer, and the semantic information of the first feature layer is less than the semantic information of the second feature layer ; 将所述多个特征层中的所述第一特征层与所述第二特征层进行融合处理，得到目标检测层，所述融合处理用于增强所述第一特征层的语义信息；Performing fusion processing on the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer, where the fusion processing is used to enhance the semantic information of the first feature layer; 基于所述目标检测层更新所述目标检测模型。The object detection model is updated based on the object detection layer. 2.根据权利要求1所述的方法，其特征在于，将所述多个特征层中的所述第一特征层与所述第二特征层进行融合处理，得到目标检测层，包括：2. The method according to claim 1, characterized in that, performing fusion processing on the first feature layer and the second feature layer in the plurality of feature layers to obtain a target detection layer, comprising: 采用插值的方式将所述第一特征层的第一尺度放大为第二尺度，其中，所述第二尺度与所述第一特征层相邻的第二特征层的尺度相等；Enlarging the first scale of the first feature layer to a second scale by means of interpolation, wherein the second scale is equal to the scale of the second feature layer adjacent to the first feature layer; 将处理后的第一特征层与所述第二特征层相加得到新的第一特征层；adding the processed first feature layer and the second feature layer to obtain a new first feature layer; 采用插值的方式将所述新的第一特征层的第三尺度放大为第四尺度，其中，所述第四尺度与所述新的第一特征层相邻的新的第二特征层的尺度相等；The third scale of the new first feature layer is enlarged to a fourth scale by means of interpolation, wherein the fourth scale is the scale of the new second feature layer adjacent to the new first feature layer equal; 将处理后的新的第一特征层与所述新的第二特征层相加，直至所述多个特征层中所有的所述第一特征层与所述第二特征层融合处理结束，得到所述目标检测层。Adding the processed new first feature layer and the new second feature layer until the fusion processing of all the first feature layers and the second feature layers in the plurality of feature layers ends, obtaining the target detection layer. 3.根据权利要求1所述的方法，其特征在于，所述方法还包括：3. The method according to claim 1, wherein the method further comprises: 获取所述目标检测模型的真实感受野；obtaining the real receptive field of the target detection model; 基于所述真实感受野调整所述目标检测模型的锚框，其中，所述锚框的大小影响所述目标检测模型的锚框分类和锚框回归；Adjust the anchor frame of the target detection model based on the true receptive field, wherein the size of the anchor frame affects the anchor frame classification and anchor frame regression of the target detection model; 根据所述锚框确定所述目标检测模型的样本数量，其中，所述样本数量包括：正样本数量和负样本数量。The number of samples of the target detection model is determined according to the anchor frame, wherein the number of samples includes: the number of positive samples and the number of negative samples. 4.根据权利要求3所述的方法，其特征在于，根据所述锚框确定所述目标检测模型的样本数量，包括：4. The method according to claim 3, wherein determining the number of samples of the target detection model according to the anchor frame, comprising: 根据所述锚框确定所述目标检测模型的损失函数，其中，所述损失函数包括：分类损失函数和回归损失函数；The loss function of the target detection model is determined according to the anchor frame, wherein the loss function includes: a classification loss function and a regression loss function; 基于所述损失函数确定所述样本数量。The number of samples is determined based on the loss function. 5.根据权利要求4所述的方法，其特征在于，采用如下计算公式，根据所述锚框确定所述目标检测模型的损失函数L，包括：5. The method according to claim 4, wherein the following calculation formula is used to determine the loss function L of the target detection model according to the anchor frame, comprising:
Figure FDA0002841505120000021
Figure FDA0002841505120000021
 其中，i表示锚框的索引，Pi表示所述锚框为目标的预测概率，Pi *表示所述锚框是目标的真实值，所述锚框为所述目标时Pi *为1，否则Pi *为0；ti表示预测的检测框坐标修正值，ti *表示检测框实际坐标值；Pi *Lr表示仅对正样本的锚框计算回归损失，Nc和Nr分别表示分类时正负锚框的数量和回归的正锚框数量，λ表示平衡参数，用于平衡分类损失和回归损失。Among them, i represents the index of the anchor frame, P i represents the predicted probability that the anchor frame is the target, P i * represents the true value of the anchor frame as the target, and P i * is 1 when the anchor frame is the target , otherwise P i * is 0; t i represents the predicted coordinate correction value of the detection frame, t i * represents the actual coordinate value of the detection frame; P i * L r means that the regression loss is only calculated for the anchor frame of the positive sample, N c and N r represents the number of positive and negative anchor boxes during classification and the number of positive anchor boxes for regression, respectively, and λ represents the balance parameter, which is used to balance the classification loss and regression loss. 6.根据权利要求4所述的方法，其特征在于，6. The method of claim 4, wherein 所述分类损失函数采用Softmax损失函数，所述回归损失函数采用Smooth-L1损失函数。The classification loss function adopts the Softmax loss function, and the regression loss function adopts the Smooth-L1 loss function. 7.一种目标检测装置，其特征在于，包括：7. A target detection device, characterized in that, comprising: 获取模块，用于获取目标检测模型的多个特征层，其中，所述多个特征层包括：第一特征层和第二特征层，所述第一特征层的语义信息少于所述第二特征层的语义信息；an acquisition module, configured to acquire multiple feature layers of the target detection model, wherein the multiple feature layers include: a first feature layer and a second feature layer, and the semantic information of the first feature layer is less than that of the second feature layer Semantic information of the feature layer; 融合处理模块，用于将所述多个特征层中的所述第一特征层与所述第二特征层进行融合处理，得到目标检测层，所述融合处理用于增强所述第一特征层的语义信息；A fusion processing module, configured to perform fusion processing on the first characteristic layer and the second characteristic layer in the plurality of characteristic layers to obtain a target detection layer, and the fusion processing is used to enhance the first characteristic layer semantic information; 更新模块，用于基于所述目标检测层更新所述目标检测模型。An update module, configured to update the target detection model based on the target detection layer. 8.一种非易失性存储介质，其特征在于，所述非易失性存储介质存储有多条指令，所述指令适于由处理器加载并执行权利要求1至6中任意一项所述的目标检测方法。8. A non-volatile storage medium, characterized in that the non-volatile storage medium stores a plurality of instructions, the instructions are adapted to be loaded by a processor and execute any one of claims 1 to 6. The described target detection method. 9.一种处理器，其特征在于，所述处理器用于运行程序，其中，所述程序被设置为运行时执行权利要求1至6中任意一项所述的目标检测方法。9 . A processor, wherein the processor is configured to run a program, wherein the program is configured to execute the target detection method according to any one of claims 1 to 6 when the program is run. 10 . 10.一种电子设备，包括存储器和处理器，其特征在于，所述存储器中存储有计算机程序，所述处理器被设置为运行所述计算机程序以执行权利要求1至6中任意一项所述的目标检测方法。10. An electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute any one of claims 1 to 6. The target detection method described above.
CN202011493902.5A 2020-12-16 2020-12-16 Target detection method and device, nonvolatile storage medium and electronic equipment Pending CN112560956A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011493902.5A CN112560956A (en) 2020-12-16 2020-12-16 Target detection method and device, nonvolatile storage medium and electronic equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011493902.5A CN112560956A (en) 2020-12-16 2020-12-16 Target detection method and device, nonvolatile storage medium and electronic equipment

Publications (1)

Publication Number Publication Date
CN112560956A true CN112560956A (en) 2021-03-26

Family

ID=75064445

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011493902.5A Pending CN112560956A (en) 2020-12-16 2020-12-16 Target detection method and device, nonvolatile storage medium and electronic equipment

Country Status (1)

Country Link
CN (1) CN112560956A (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108520229A (en) * 2018-04-04 2018-09-11 北京旷视科技有限公司 Image detecting method, device, electronic equipment and computer-readable medium
CN109741318A (en) * 2018-12-30 2019-05-10 北京工业大学 A single-stage multi-scale real-time detection method for specific targets based on effective receptive field
US20190287215A1 (en) * 2018-03-13 2019-09-19 Disney Enterprises, Inc. Image Processing Using A Convolutional Neural Network
CN110287927A (en) * 2019-07-01 2019-09-27 西安电子科技大学 Object detection method for remote sensing images based on deep multi-scale and context learning
CN111027547A (en) * 2019-12-06 2020-04-17 南京大学 Automatic detection method for multi-scale polymorphic target in two-dimensional image
CN111126472A (en) * 2019-12-18 2020-05-08 南京信息工程大学 Improved target detection method based on SSD
CN111524106A (en) * 2020-04-13 2020-08-11 北京推想科技有限公司 Skull fracture detection and model training method, device, equipment and storage medium
CN111652288A (en) * 2020-05-11 2020-09-11 北京航天自动控制研究所 Improved SSD small target detection method based on dense feature pyramid
CN111832655A (en) * 2020-07-16 2020-10-27 四川大学 A Multi-scale 3D Object Detection Method Based on Feature Pyramid Network
CN111914727A (en) * 2020-07-28 2020-11-10 联芯智能(南京)科技有限公司 Small target human detection method based on balanced sampling and nonlinear feature fusion
CN112052861A (en) * 2019-06-05 2020-12-08 高新兴科技集团股份有限公司 Method for calculating effective receptive field of deep convolutional neural network and storage medium

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190287215A1 (en) * 2018-03-13 2019-09-19 Disney Enterprises, Inc. Image Processing Using A Convolutional Neural Network
CN108520229A (en) * 2018-04-04 2018-09-11 北京旷视科技有限公司 Image detecting method, device, electronic equipment and computer-readable medium
CN109741318A (en) * 2018-12-30 2019-05-10 北京工业大学 A single-stage multi-scale real-time detection method for specific targets based on effective receptive field
CN112052861A (en) * 2019-06-05 2020-12-08 高新兴科技集团股份有限公司 Method for calculating effective receptive field of deep convolutional neural network and storage medium
CN110287927A (en) * 2019-07-01 2019-09-27 西安电子科技大学 Object detection method for remote sensing images based on deep multi-scale and context learning
CN111027547A (en) * 2019-12-06 2020-04-17 南京大学 Automatic detection method for multi-scale polymorphic target in two-dimensional image
CN111126472A (en) * 2019-12-18 2020-05-08 南京信息工程大学 Improved target detection method based on SSD
CN111524106A (en) * 2020-04-13 2020-08-11 北京推想科技有限公司 Skull fracture detection and model training method, device, equipment and storage medium
CN111652288A (en) * 2020-05-11 2020-09-11 北京航天自动控制研究所 Improved SSD small target detection method based on dense feature pyramid
CN111832655A (en) * 2020-07-16 2020-10-27 四川大学 A Multi-scale 3D Object Detection Method Based on Feature Pyramid Network
CN111914727A (en) * 2020-07-28 2020-11-10 联芯智能(南京)科技有限公司 Small target human detection method based on balanced sampling and nonlinear feature fusion

Similar Documents

Publication Publication Date Title
WO2018036293A1 (en) Image segmentation method, apparatus, and fully convolutional network system
CN110288199A (en) The method of product quality forecast
TW201941058A (en) Anomaly detection method and device
CN111445320B (en) Target community identification method and device, computer equipment and storage medium
CN113344910A (en) Defect labeling image generation method and device, computer equipment and storage medium
CN107818301A (en) Update the method, apparatus and electronic equipment of biometric templates
CN109213831A (en) Event detecting method and device calculate equipment and storage medium
CN111061624A (en) Policy execution effect determination method and device, electronic equipment and storage medium
CN105590307A (en) Transparency-based matting method and apparatus
CN109858282A (en) A kind of social network relationships data-privacy guard method and system
CN109685805B (en) Image segmentation method and device
CN113553950B (en) Abnormal event detection method, device, electronic device and storage medium
CN110876072A (en) Batch registered user identification method, storage medium, electronic device and system
CN113255670A (en) Unbalanced small sample target detection method and device and computer equipment
CN105335368A (en) Product clustering method and apparatus
CN113343918A (en) Power equipment identification method, system, medium and electronic equipment
CN115937205A (en) Surface defect tile image generation method, device, equipment and storage medium
CN104050291B (en) A kind of method for parallel processing and system of account balance data
CN112084865A (en) Target detection method, target detection device, electronic equipment and storage medium
CN108734366A (en) User identification method and its system
CN113850523A (en) ESG index determining method based on data completion and related product
CN107608844B (en) Method, system and device for testing hardware and computer readable storage medium
CN112560956A (en) Target detection method and device, nonvolatile storage medium and electronic equipment
CN118606218A (en) Test case selection method and device
CN108055638A (en) Obtain method, apparatus, computer-readable medium and the equipment of target location

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20210326

RJ01 Rejection of invention patent application after publication