CN111539390A - Small target image identification method, equipment and system based on Yolov3 - Google Patents
Small target image identification method, equipment and system based on Yolov3 Download PDFInfo
- Publication number
- CN111539390A CN111539390A CN202010573564.XA CN202010573564A CN111539390A CN 111539390 A CN111539390 A CN 111539390A CN 202010573564 A CN202010573564 A CN 202010573564A CN 111539390 A CN111539390 A CN 111539390A
- Authority
- CN
- China
- Prior art keywords
- loss
- coefficient
- setting
- category
- small target
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a small target image recognition method, equipment and system based on Yolov3, and belongs to the technical field of image recognition. According to the method, the area loss is added, so that the defect of the traditional coordinate independent loss summation is reduced, and the prediction precision of the loss function is improved, so that the small target identification precision is further improved; by weakening the network prediction layer with large characteristic size, enhancing the network prediction layer with small characteristic size, highlighting the identification depth of the small target layer and improving the identification precision of the small target; and after the preset time period, data backflow iteration is carried out on the recognition model, so that the small target recognition precision is further improved.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to a small target image recognition method, equipment and system based on Yolov 3.
Background
In recent years, with the development of deep learning technology, the image features and classification are gradually extracted and classified by using a novel convolutional neural network instead of the traditional manual method; under the scene that the novel convolutional neural network technology is continuously popularized, in order to meet the monitoring scene and the identification requirement, the accuracy of small target identification in the image is required.
The prior art provides a small target image identification method, which comprises the steps of firstly confirming the type of a target to be detected, obtaining the ratio of the width and height of all target labeled frames to the width and height of an original image, and clustering 9 anchor frames with different sizes by using K-means to realize the identification of the small target image.
However, when the method provided by the prior art is used, the small target layer identification depth is not highlighted, so that the accuracy and precision of small target image identification are low.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a method, a device and a system for identifying a small target image based on Yolov3, wherein the method comprises the following steps:
in one aspect, a small target image recognition method based on Yolov3 is provided, and includes:
setting the labeling category of the small target image to be recognized in the recognition model;
setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient;
setting a loss function as a weighting function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient;
training the recognition model according to the labeling type and the loss function until the training loss stops decreasing;
and identifying the small target image to be identified according to the identification model.
Optionally, the training of the recognition model according to the labeling category and the loss function is performed until training loss stops decreasing, and the method further includes:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
Optionally, after the small target image to be recognized is recognized according to the recognition model, the method further includes:
and after a preset time period, performing data reflow iteration on the recognition model.
Optionally, the weakening of the network prediction layer with a large feature size and the enhancing of the network prediction layer with a small feature size include:
setting a second block x 2 of the network layer prediction layer to block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
Optionally, the setting of the annotation category of the small target image to be identified in the identification model further includes:
and acquiring the labeling category of the small target image to be identified, which is customized by the user through the picture/video.
In another aspect, a Yolov 3-based small target image recognition device is provided, including:
the setting module is used for setting the labeling category of the small target image to be recognized in the recognition model;
the optimization module is used for setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient;
the optimization module is further used for setting a loss function as a weighted function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient;
the training module is used for training the recognition model according to the labeling type and the loss function until the training loss stops decreasing;
and the identification module is used for identifying the small target image to be identified according to the identification model.
Optionally, the optimization module is further configured to:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
Optionally, the optimization module is further configured to:
and after a preset time period, performing data reflow iteration on the recognition model.
Optionally, the optimization module is further specifically configured to:
setting a second block x 2 of the network layer prediction layer to block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
Optionally, the setting module is specifically configured to:
and acquiring the labeling category of the small target image to be identified, which is customized by the user through the picture/video.
In another aspect, a Yolov 3-based small object image recognition device is provided, the device comprising a processor and a memory connected to the processor, the memory storing a set of program codes, the processor executing the program codes stored in the memory to:
setting the labeling category of the small target image to be recognized in the recognition model;
setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient;
setting a loss function as a weighting function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient;
training the recognition model according to the labeling type and the loss function until the training loss stops decreasing;
and identifying the small target image to be identified according to the identification model.
Optionally, the processor executes the program code stored in the memory to further implement the following operations:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
Optionally, the processor executes the program code stored in the memory to further implement the following operations:
and after a preset time period, performing data reflow iteration on the recognition model.
Optionally, the processor executes the program code stored in the memory to further implement the following operations:
setting a second block x 2 of the network layer prediction layer to block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
Optionally, the processor executes the program code stored in the memory to further implement the following operations:
and acquiring the labeling category of the small target image to be identified, which is customized by the user through the picture/video.
In another aspect, a Yolov 3-based small target image recognition system is provided, including:
the setting equipment is used for setting the labeling category of the small target image to be recognized in the recognition model;
the optimization equipment is used for setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient;
the optimization equipment is further used for setting a loss function as a weighted function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient;
the training equipment is used for training the recognition model according to the labeling type and the loss function until the training loss stops decreasing;
and the identification equipment is used for identifying the small target image to be identified according to the identification model.
Optionally, the optimization device is further configured to:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
Optionally, the optimization device is further configured to:
and after a preset time period, performing data reflow iteration on the recognition model.
Optionally, the optimization device is further specifically configured to:
setting a second block x 2 of the network layer prediction layer to block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
Optionally, the setting device is specifically configured to:
and acquiring the labeling category of the small target image to be identified, which is customized by the user through the picture/video.
The invention provides a method, a device and a system for identifying a small target image based on Yolov3, wherein the method comprises the following steps: setting the labeling category of the small target image to be recognized in the recognition model; setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient; setting a loss function as a weighting function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient; training the recognition model according to the labeling type and the loss function until the training loss stops decreasing; and identifying the small target image to be identified according to the identification model.
The technical scheme provided by the invention has the following beneficial effects:
1. by adding area loss, the defect of the traditional coordinate independent loss summation is reduced, and the prediction precision of the loss function is improved, so that the small target identification precision is further improved;
2. by weakening the network prediction layer with large characteristic size, enhancing the network prediction layer with small characteristic size, highlighting the identification depth of the small target layer and improving the identification precision of the small target;
3. and after the preset time period, data backflow iteration is carried out on the recognition model, so that the small target recognition precision is further improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a flowchart of a method for identifying a small target image based on Yolov3 according to an embodiment of the present invention;
FIG. 2 is a flowchart of a method for identifying a small target image based on Yolov3 according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a small target image recognition device based on Yolov3 according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a small target image recognition device based on Yolov3 according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a Yolov 3-based small target image recognition system according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying 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.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, 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 some, but not all, embodiments of the present invention. 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.
In addition, the term "and/or" herein is only one kind of association relationship describing an associated object, and means that there may be three kinds of relationships, for example, a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship, unless otherwise specified.
The Yolov 3-based small target image identification method provided by the embodiment of the invention is mainly applied to monitoring scenes, social scenes based on videos/pictures, beautification scenes or processing scenes of the videos/pictures, and in addition, the method provided by the embodiment of the invention can also be applied to other scenes, and the embodiment of the invention does not limit specific scenes.
To facilitate the understanding of the method described in the embodiment of the present invention by those skilled in the art, the existing Yolov3 described in the embodiment of the present invention may be:
the loss function includes at least: coordinate loss, width and height loss, confidence loss and category loss;
the network prediction layer is as follows:
network layer predicting layer second block x 2;
the third block of the network layer prediction layer 8;
network layer predicting layer fourth block 8;
the network layer predicts the fifth block 4 of the layer.
Example one
The embodiment of the invention provides a small target image identification method based on Yolov3, and as shown in fig. 1, the method comprises the following steps:
101. and setting the labeling type of the small target image to be recognized in the recognition model.
Specifically, the labeling category of the small target image to be identified, which is customized by the user through the picture/video, is obtained.
102. And setting a loss function in the recognition model according to the area loss.
Specifically, an area loss coefficient, a confidence coefficient and a category loss coefficient are set;
and setting the loss function as a weighted function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss and the category loss coefficient.
103. Weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
Specifically, a second block x 2 of the network layer prediction layer is set as block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
104. And training the recognition model according to the labeling type and the loss function until the training loss stops decreasing.
105. And identifying the small target image to be identified according to the identification model.
106. And after a preset time period, performing data reflow iteration on the recognition model.
Example two
An embodiment of the present invention provides a small target image identification method based on Yolov3, and as shown in fig. 2, the method includes:
201. and setting the labeling type of the small target image to be recognized in the recognition model.
Specifically, selecting the labeling category of the small target image from the small target data set;
and setting the annotation category as the annotation category of the small target image to be identified in the identification model.
In practical applications, in order to improve the user experience, the process described in step 201 may further be:
in a social scene, acquiring a labeling category of a small target image to be identified, which is customized by a user through a picture/video, wherein the process specifically comprises the following steps:
identifying a small target image selected by a user through a gesture in a picture/video;
identifying the annotation category of the small target image, wherein the gesture of the user at least comprises inputting a track at least comprising the small target image on an interface displaying pictures/videos; the identification process can be realized through a pre-configured identification model, the identification model can be pre-trained by the method of the embodiment of the invention, or can be realized through the existing identification algorithm or the identification model;
displaying the small target image and the recognition result to a user;
and after the user inputs the confirmation information, setting the annotation category of the small target image as the annotation category of the small target image to be identified.
Optionally, in a social scene or a beautification scene or a processing scene of a video/picture, the annotation category of the small target image to be identified, which is customized by the user through the picture/video, is obtained, and the process specifically includes:
acquiring a picture/video input by a user, wherein the picture/video at least comprises a small target image required by the user;
and identifying the small target image and the labeling type of the small target image in the same way as the steps, and the detailed description is omitted here.
And displaying the small target image and the recognition result to the user, and setting the annotation category of the small target image as the annotation category of the small target image to be recognized after the user inputs the confirmation information.
In addition, in the monitoring scenario, step 201 may also be actively triggered after the monitored area is changed after the installation of the monitoring device is completed, so that when the target in the monitoring scenario changes, the changed small target identification may be implemented, thereby improving the reliability in the monitoring scenario.
After the installation of the monitoring device is completed, the labeling category of the small target image to be recognized in the recognition model is periodically set, and the process may be:
after a preset time period, all small target images in the monitored area and the labeling types of all the small target images are acquired, and if the labeling types are different from the previous labeling types, step 201 and the subsequent steps are triggered.
202. An area loss coefficient, a confidence loss coefficient, and a category loss coefficient are set.
Specifically, the setting process may be set through an empirical value, or may be set through a training result or a recognition result, and the specific setting manner is not limited in the embodiment of the present invention.
The process of setting the area loss coefficient, the confidence coefficient and the category loss coefficient according to the training result or the recognition result can also be setting the area loss coefficient, the confidence coefficient and the category loss coefficient according to the training result or the recognition result; the process of setting the area loss coefficient, the confidence coefficient and the category loss coefficient according to the training result or the recognition result may be:
respectively setting an area loss coefficient interval, a confidence coefficient interval and a category loss coefficient interval, respectively randomly selecting values in the intervals to form a plurality of area loss coefficient, confidence coefficient and category loss coefficient value groups, and executing the following steps for any group of values:
setting the value group as an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient, and executing the processes from the step 201 to the step 206 to obtain a trained recognition model;
inputting test data into the identification model to obtain a test result;
continuing to execute the steps on the next group of numerical values until all the numerical values in the interval complete the steps to obtain a plurality of test results;
and evaluating the test result, and setting the numerical value group with the test result which is most consistent with the expectation as an area loss coefficient, a confidence coefficient and a category loss coefficient.
203. And setting the loss function as a weighted function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss and the category loss coefficient.
Specifically, the loss function may be specifically:
wherein, therein
In the interest of a loss of area of the frame,
in order to be a loss of confidence,
is a category loss;
the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient are respectively;
in addition, in practical application, the area loss in the embodiment of the present invention may be obtained by coordinate loss and width-height loss, and the embodiment of the present invention does not limit a specific obtaining manner;
it should be noted that steps 202 to 203 are processes for implementing the setting of the loss function in the recognition model according to the area loss, and the processes may be implemented in other ways besides the ways described in the above steps, and the embodiment of the present invention is not limited to the specific ways.
The loss function in the existing Yolov3 algorithm is the summation of coordinate loss and width and height loss, and the width and height loss is calculated independently, so that the recognition precision is low, the reliability and the accuracy of small target recognition are affected, and through area loss, the problems are avoided, end-to-end loss evaluation is realized, and the reliability and the accuracy of small target recognition are further improved.
204. And enhancing the network prediction layer with small feature size.
Specifically, a second block x 2 of the network layer prediction layer is set as block x 6;
the third block 8 of the network layer prediction layer is set to block 10.
205. And weakening the network prediction layer with large feature size.
Specifically, a fourth block × 8 of the network layer prediction layer is set as a block × 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
It should be noted that the manner of enhancing the network prediction layer with small feature size and weakening the network prediction layer with large feature size described in step 204 to step 205 is merely exemplary, and in practical applications, the process is implemented by the training result or the recognition result.
By weakening the network prediction layer with large characteristic size, enhancing the network prediction layer with small characteristic size, and utilizing the advantages of the network prediction layer with small characteristic size in small target image recognition, the reliability of small target image recognition is further strengthened.
206. And training the recognition model according to the labeling type and the loss function until the training loss stops decreasing.
Specifically, a training parameter is set, and the recognition model is trained according to the training parameter, the label category and the loss function until the training loss stops decreasing.
207. And identifying the small target image to be identified according to the identification model.
Specifically, a picture containing a small target image to be recognized is input to the recognition model, and a recognition result is output.
208. And after a preset time period, performing data reflow iteration on the recognition model.
Specifically, the data reflow iteration mode includes training the recognition model again.
Besides performing periodic iteration according to a preset time period, the iterative processing method may further include the following steps:
the manner of actively triggering after the monitored area changes and detecting whether the monitored area changes is the same as the manner described in step 201, and is not described herein again.
The method can also be triggered actively by the user, and the active triggering mode of the user comprises clicking a virtual key or periodically carrying out testing through a testing image and then actively triggering.
EXAMPLE III
An embodiment of the present invention provides a Yolov 3-based small target image recognition device 3, and as shown in fig. 3, the device includes:
the setting module 31 is used for setting the labeling category of the small target image to be recognized in the recognition model;
an optimization module 32, configured to set an area loss coefficient, a confidence loss coefficient, and a category loss coefficient;
the optimization module 32 is further configured to set the loss function as a weighted function of the area loss, the confidence coefficient loss, and the category loss according to the area loss coefficient, the confidence coefficient loss, and the category loss coefficient;
the training module 33 is configured to train the recognition model according to the labeled category and the loss function until the training loss stops decreasing;
and the identification module 34 is used for identifying the small target image to be identified according to the identification model.
Optionally, the optimization module 32 is further configured to:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
Optionally, the optimization module 32 is further configured to:
and after a preset time period, performing data reflow iteration on the recognition model.
Optionally, the optimization module 32 is further specifically configured to:
setting a second block x 2 of the network layer prediction layer to block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
Optionally, the setting module 31 is specifically configured to:
and acquiring the annotation category of the small target image to be identified, which is customized by the user through the picture/video.
Example four
An embodiment of the present invention provides a Yolov 3-based small target image recognition device 4, which is shown in fig. 4, and includes a processor 41 and a memory 42 connected to the processor, where the memory 42 is used to store a set of program codes, and the processor 41 executes the program codes stored in the memory 42 to implement the following operations:
setting the labeling category of the small target image to be recognized in the recognition model;
setting a loss function in the recognition model according to the area loss;
training the recognition model according to the labeling type and the loss function until the training loss stops decreasing;
and identifying the small target image to be identified according to the identification model.
Optionally, the processor 41 executes the program code stored in the memory 42 for implementing the following operations:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
Optionally, the processor 41 executes the program code stored in the memory 42 for implementing the following operations:
and after a preset time period, performing data reflow iteration on the recognition model.
Optionally, the processor 41 executes the program code stored in the memory 42 for implementing the following operations:
setting a second block x 2 of the network layer prediction layer to block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
Optionally, the processor 41 executes the program code stored in the memory 42 for implementing the following operations:
and acquiring the annotation category of the small target image to be identified, which is customized by the user through the picture/video.
EXAMPLE five
An embodiment of the present invention provides a Yolov 3-based small target image recognition system, and as shown in fig. 5, the system includes:
the setting device 51 is used for setting the labeling category of the small target image to be recognized in the recognition model;
an optimization device 52 for setting an area loss coefficient, a confidence loss coefficient, and a category loss coefficient;
the optimization device 52 is further configured to set the loss function as a weighted function of the area loss, the confidence loss, and the category loss according to the area loss coefficient, the confidence loss coefficient, and the category loss coefficient;
the training device 53 is used for training the recognition model according to the labeled category and the loss function until the training loss stops decreasing;
and the recognition device 54 is used for recognizing the small target image to be recognized according to the recognition model.
Optionally, the optimizing device 52 is further configured to:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
Optionally, the optimizing device 52 is further configured to:
and after a preset time period, performing data reflow iteration on the recognition model.
Optionally, the optimization device 52 is specifically configured to:
setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient;
and setting the loss function as a weighted function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss and the category loss coefficient.
Optionally, the optimization device 52 is further specifically configured to:
setting a second block x 2 of the network layer prediction layer to block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
Optionally, the setting device 51 is specifically configured to:
and acquiring the annotation category of the small target image to be identified, which is customized by the user through the picture/video.
The embodiment of the invention provides a small target image identification method, equipment and system based on Yolov3, which reduce the defect of the traditional independent loss summation of coordinates by adding area loss, improve the prediction precision of a loss function and further improve the identification precision of a small target; by weakening the network prediction layer with large characteristic size, enhancing the network prediction layer with small characteristic size, highlighting the identification depth of the small target layer and improving the identification precision of the small target; and after the preset time period, data backflow iteration is carried out on the recognition model, so that the small target recognition precision is further improved.
All the above-mentioned optional technical solutions can be combined arbitrarily to form the optional embodiments of the present invention, and are not described herein again.
It should be noted that: when the small target image recognition device and the small target image recognition system provided by the above embodiments execute the small target image recognition method, only the division of the above functional modules is taken as an example, and in practical applications, the above function distribution may be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules, so as to complete all or part of the above described functions. In addition, the small target image recognition method, device and system embodiments provided by the above embodiments belong to the same concept, and specific implementation processes thereof are detailed in the method embodiments and are not described herein again.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, where the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (10)
1. A small target image identification method based on Yolov3 is characterized by comprising the following steps:
setting the labeling category of the small target image to be recognized in the recognition model;
setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient;
setting a loss function as a weighting function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient;
training the recognition model according to the labeling type and the loss function until the training loss stops decreasing;
and identifying the small target image to be identified according to the identification model.
2. The method of claim 1, wherein the training of the recognition model according to the label class and the loss function is performed until training loss stops decreasing, the method further comprising:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
3. The method according to claim 1 or 2, wherein after the small target image to be recognized is recognized according to the recognition model, the method further comprises:
and after a preset time period, performing data reflow iteration on the recognition model.
4. The method of claim 3, wherein weakening the network prediction layer with large feature size and enhancing the network prediction layer with small feature size comprises:
setting a second block x 2 of the network layer prediction layer to block x 6;
setting a third block 8 of the network layer prediction layer as block 10;
setting a fourth block 8 of the network layer prediction layer as block 4;
the fifth block 4 of the network layer prediction layer is set to block 2.
5. The method according to claim 4, wherein the setting of the label category of the small target image to be identified in the identification model further comprises:
and acquiring the labeling category of the small target image to be identified, which is customized by the user through the picture/video.
6. A Yolov 3-based small object image recognition device, the device comprising:
the setting module is used for setting the labeling category of the small target image to be recognized in the recognition model;
the optimization module is used for setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient;
the optimization module is further used for setting a loss function as a weighted function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient;
the training module is used for training the recognition model according to the labeling type and the loss function until the training loss stops decreasing;
and the identification module is used for identifying the small target image to be identified according to the identification model.
7. The apparatus of claim 6, wherein the optimization module is further configured to:
weakening the network prediction layer with large characteristic size and enhancing the network prediction layer with small characteristic size.
8. The apparatus of claim 6 or 7, wherein the optimization module is further configured to:
and after a preset time period, performing data reflow iteration on the recognition model.
9. The device of claim 8, wherein the setup module is specifically configured to:
and acquiring the labeling category of the small target image to be identified, which is customized by the user through the picture/video.
10. A Yolov 3-based small object image recognition system, the system comprising:
the setting equipment is used for setting the labeling category of the small target image to be recognized in the recognition model;
the optimization equipment is used for setting an area loss coefficient, a confidence coefficient loss coefficient and a category loss coefficient;
the optimization equipment is further used for setting a loss function as a weighted function of the area loss, the confidence coefficient loss and the category loss according to the area loss coefficient, the confidence coefficient loss coefficient and the category loss coefficient;
the training equipment is used for training the recognition model according to the labeling type and the loss function until the training loss stops decreasing;
and the identification equipment is used for identifying the small target image to be identified according to the identification model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010573564.XA CN111539390A (en) | 2020-06-22 | 2020-06-22 | Small target image identification method, equipment and system based on Yolov3 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010573564.XA CN111539390A (en) | 2020-06-22 | 2020-06-22 | Small target image identification method, equipment and system based on Yolov3 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111539390A true CN111539390A (en) | 2020-08-14 |
Family
ID=71980840
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010573564.XA Pending CN111539390A (en) | 2020-06-22 | 2020-06-22 | Small target image identification method, equipment and system based on Yolov3 |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111539390A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733747A (en) * | 2021-01-14 | 2021-04-30 | 哈尔滨市科佳通用机电股份有限公司 | Identification method, system and device for relieving falling fault of valve pull rod |
| CN113591662A (en) * | 2021-07-24 | 2021-11-02 | 深圳市铁越电气有限公司 | Method, system and storage medium for recognizing smoking calling behavior |
| CN114092829A (en) * | 2021-11-22 | 2022-02-25 | 南京华苏科技有限公司 | Small target detection method based on improved YOLOV4 in high-altitude operation scene |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711381A (en) * | 2019-01-04 | 2019-05-03 | 北京环境特性研究所 | Target identification method, device and the computer equipment of remote sensing images |
| CN109711437A (en) * | 2018-12-06 | 2019-05-03 | 武汉三江中电科技有限责任公司 | A kind of transformer part recognition methods based on YOLO network model |
| CN110766098A (en) * | 2019-11-07 | 2020-02-07 | 中国石油大学(华东) | Traffic scene small target detection method based on improved YOLOv3 |
-
2020
- 2020-06-22 CN CN202010573564.XA patent/CN111539390A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711437A (en) * | 2018-12-06 | 2019-05-03 | 武汉三江中电科技有限责任公司 | A kind of transformer part recognition methods based on YOLO network model |
| CN109711381A (en) * | 2019-01-04 | 2019-05-03 | 北京环境特性研究所 | Target identification method, device and the computer equipment of remote sensing images |
| CN110766098A (en) * | 2019-11-07 | 2020-02-07 | 中国石油大学(华东) | Traffic scene small target detection method based on improved YOLOv3 |
Non-Patent Citations (2)
| Title |
|---|
| 徐义鎏 等: "改进损失函数的 Yolov3车型检测算法", 《信息通信》 * |
| 袁铭阳 等: "一种基于改进 YOLOv3 的遥感影像飞机目标检测算法", 《测绘科学技术学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733747A (en) * | 2021-01-14 | 2021-04-30 | 哈尔滨市科佳通用机电股份有限公司 | Identification method, system and device for relieving falling fault of valve pull rod |
| CN113591662A (en) * | 2021-07-24 | 2021-11-02 | 深圳市铁越电气有限公司 | Method, system and storage medium for recognizing smoking calling behavior |
| CN113591662B (en) * | 2021-07-24 | 2024-07-09 | 深圳市铁越电气有限公司 | Recognition method, system and storage medium for smoking and calling behaviors |
| CN114092829A (en) * | 2021-11-22 | 2022-02-25 | 南京华苏科技有限公司 | Small target detection method based on improved YOLOV4 in high-altitude operation scene |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR102635987B1 (en) | Method, apparatus, device and storage medium for training an image semantic segmentation network | |
| CN112712051B (en) | Object tracking method, device, computer equipment and storage medium | |
| CN112381104B (en) | Image recognition method, device, computer equipment and storage medium | |
| CN110781960B (en) | Training method, classification method, device and equipment of video classification model | |
| CN111985385A (en) | Behavior detection method, device and equipment | |
| CN111061898A (en) | Image processing method, device, computer equipment and storage medium | |
| CN115713715B (en) | Human behavior recognition method and recognition system based on deep learning | |
| CN113065379B (en) | Image detection method and device integrating image quality and electronic equipment | |
| CN111652181B (en) | Target tracking method and device and electronic equipment | |
| CN111539390A (en) | Small target image identification method, equipment and system based on Yolov3 | |
| CN113762382B (en) | Model training and scene recognition method, device, equipment and medium | |
| CN111708913A (en) | Label generation method and device and computer readable storage medium | |
| CN115100739B (en) | Man-machine behavior detection method, system, terminal device and storage medium | |
| CN113705293A (en) | Image scene recognition method, device, equipment and readable storage medium | |
| CN110427998A (en) | Model training, object detection method and device, electronic equipment, storage medium | |
| CN111432206A (en) | Artificial intelligence-based video resolution processing method, device and electronic device | |
| CN111950507B (en) | Data processing and model training method, device, equipment and medium | |
| CN115830697A (en) | Student classroom behavior identification method, device, equipment and storage medium | |
| CN112784691A (en) | Target detection model training method, target detection method and device | |
| CN117173624A (en) | Object recognition model learning method in computing device | |
| CN109740527B (en) | A kind of image processing method in video frame | |
| CN115049675A (en) | Generation area determination and light spot generation method, apparatus, medium, and program product | |
| CN112733883B (en) | Point supervision target detection method | |
| CN112241736A (en) | Text detection method and device | |
| CN113192108A (en) | Human-in-loop training method for visual tracking model and related device |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200814 |