CN113537145A - Method, device and storage medium for rapidly solving false detection and missed detection in target detection - Google Patents
Method, device and storage medium for rapidly solving false detection and missed detection in target detection Download PDFInfo
- Publication number
- CN113537145A CN113537145A CN202110892872.3A CN202110892872A CN113537145A CN 113537145 A CN113537145 A CN 113537145A CN 202110892872 A CN202110892872 A CN 202110892872A CN 113537145 A CN113537145 A CN 113537145A
- Authority
- CN
- China
- Prior art keywords
- target
- verified
- detection
- false
- missed
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 353
- 238000000034 method Methods 0.000 title claims abstract description 67
- 239000013598 vector Substances 0.000 claims abstract description 131
- 238000012795 verification Methods 0.000 claims abstract description 85
- 238000000605 extraction Methods 0.000 claims description 22
- 238000005070 sampling Methods 0.000 claims description 14
- 238000004590 computer program Methods 0.000 claims description 10
- 230000015654 memory Effects 0.000 claims description 10
- 230000004044 response Effects 0.000 claims description 4
- 238000012549 training Methods 0.000 description 9
- 238000013135 deep learning Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000012512 characterization method Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method, a device and a storage medium for rapidly solving false detection and missed detection in target detection, aiming at a target to be verified and reported incorrectly or a target to be verified and reported incorrectly, corresponding pixel areas are input into a verification network model to obtain a group of multidimensional characteristic vectors, simultaneously, the real false-report target or the missing-report target is input into the check network model to obtain another group of most characteristic vectors, and by calculating the difference between the classes between the two groups of corresponding multi-dimensional characteristic vectors, the corresponding inter-class difference may reflect the difference in pixel distribution between the pixel region with the false positive target to be verified and the pixel region with the false positive target, or the difference in pixel distribution between the pixel region having the false positive target to be verified and the pixel region having the false negative target, therefore, by adopting the technical means of inter-class difference comparison, the model does not need to be retrained, and the quick solution of the target to be verified and misreported or the target to be verified and misreported can be realized.
Description
The present application claims priority from prior application of prior patent application having application number "202110720856.6" filed on 28/6/2021 and entitled "target detection verification method, apparatus and storage medium".
Technical Field
The invention relates to the technical field of computers, in particular to a method and a device for quickly solving false detection and missed detection in target detection and a storage medium.
Background
Currently, deep learning algorithms are applied for target detection. When the deep learning algorithm is used for target detection, the target detection result may include the situations of false detection and missed detection, and in the general situation, the pictures of the false detection and the missed detection are collected firstly, then the pictures are labeled manually, and the deep learning model is retrained by using the manually labeled pictures so as to reduce the probability of the later false detection or missed detection. Therefore, the prior art can not quickly solve the problems of error detection and missing detection in target detection.
Disclosure of Invention
The invention provides a method and a device for quickly solving the problems of error and missing detection in target detection, electronic equipment and a storage medium, which are used for solving the problem that the error and missing detection in the target detection can not be quickly solved in the prior art.
The invention provides a method for quickly solving false detection and missed detection in target detection, which comprises the following steps:
under the condition that a false alarm target to be verified is obtained, inputting a pixel region with the false alarm target to be verified into a verification network model, outputting a first multi-dimensional feature vector generated by downsampling, inputting a pixel image with the false alarm target into the verification network model, outputting a second multi-dimensional feature vector generated by downsampling, calculating a first inter-class difference between the first multi-dimensional feature vector and the second multi-dimensional feature vector by applying a Hilbert space norm, and if the first inter-class difference is smaller than a first preset threshold value, determining that the false alarm target to be verified is false and the target detection model has target false detection; and/or
Under the condition that a target to be verified and missed in report is obtained, inputting a pixel area with the target to be verified and missed in report into the verification network model, outputting a third multi-dimensional feature vector generated by down sampling, inputting a pixel image with the target to be verified and missed in report into the verification network model, outputting a fourth multi-dimensional feature vector generated by down sampling, calculating a second inter-class difference between the third multi-dimensional feature vector and the fourth multi-dimensional feature vector by applying a Hilbert space norm, and if the second inter-class difference is smaller than a second preset threshold value, determining that the target to be verified and missed in report is missed and the target detection model has target missing detection;
and the target to be checked and the target to be missed are input into the target detection model and output.
According to the method for quickly solving the false detection and the missed detection in the target detection, provided by the invention, the check network model is obtained by training a first residual error network, the target detection model is obtained by training a second residual error network, and the depth of the first residual error network is smaller than that of the second residual error network.
According to the method for rapidly solving the false detection and the missed detection in the target detection, provided by the invention, the first residual error network is resnet18, and the second residual error network is resnet 50.
According to the method for rapidly solving the false detection and the missed detection in the target detection, before the pixel area with the target to be verified and falsely reported is input into the verification network model or the pixel area with the target to be verified and missed reported is input into the verification network model, the method further comprises the following steps:
inputting a video frame set into the target detection model, and outputting a target set and a corresponding target area;
traversing the target area to obtain the area of the false report target to be verified or the area of the missing report target to be verified;
and extracting a pixel area with the false alarm target to be verified or the pixel area with the missing target to be verified from the video frame set.
According to the method for rapidly solving the false detection and the missed detection in the target detection, before the pixel image with the false-alarm target is input into the check network model, the method further comprises the following steps:
in response to an input of a user indicating the false-positive target in the target set, extracting a target region corresponding to the false-positive target from the target set, and extracting a pixel image with the false-positive target, which comprises the target region, from the video frame set; or
And responding to the input of a user indicating the false negative target in the target set, extracting a target area corresponding to the false negative target from the target set, and extracting a pixel image with the false negative target, which comprises the target area, from the video frame set.
According to the method for rapidly solving the false detection and the missed detection in the target detection, provided by the invention, the method further comprises the following steps:
outputting the target detection result of the target to be verified and missed to be reported to a result target set;
outputting the result target set to a user interface.
According to the method for rapidly solving the false detection and the missed detection in the target detection, provided by the invention, the method further comprises the following steps:
if the first inter-class difference is not smaller than the first preset threshold, determining that the false alarm target to be verified is not a false alarm target, and outputting a target detection result of the target to be verified to a result target set;
outputting the result target set to a user interface.
The invention also provides a device for rapidly solving the problems of error detection and missed detection in target detection, which comprises:
the device comprises a first extraction module, a second extraction module, a comparison module and a check module;
the first extraction module is used for inputting a pixel region with the false alarm target to be verified into a verification network model under the condition of acquiring the false alarm target to be verified, and outputting a first multi-dimensional feature vector generated by downsampling;
the second extraction module is used for inputting the pixel image with the false alarm target into the check network model and outputting a second multi-dimensional feature vector generated by downsampling;
the comparison module is used for calculating a first inter-class difference between the first multi-dimensional feature vector and the second multi-dimensional feature vector by applying a Hilbert space norm;
the checking module is used for determining that the target to be checked is false alarm and the target detection model has target false detection if the first inter-class difference is smaller than a first preset threshold; and/or
The first extraction module is specifically further configured to, under the condition that a target to be verified and missed-report is obtained, input a pixel region with the target to be verified and missed-report into the verification network model, and output a third multidimensional feature vector generated by downsampling;
the second extraction module is specifically further configured to input the pixel image with the under-sampling target into the check network model, and output a fourth multidimensional feature vector generated by down-sampling;
the comparison module is further specifically configured to apply a hilbert spatial norm to calculate a second inter-class difference between the third multidimensional feature vector and the fourth multidimensional feature vector;
the checking module is further specifically configured to determine that the target to be checked is false negative and the target detection model has target false negative if the difference between the second classes is smaller than a second preset threshold.
The invention also provides an electronic device, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, and is characterized in that the processor implements the steps of the method for quickly solving the false and missed detection in any one of the target detection when executing the program.
The present invention also provides a non-transitory computer readable storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the method for quickly solving the false or missing detection in the target detection.
The invention provides a method, a device, electronic equipment and a storage medium for rapidly solving false detection and missed detection in target detection, aiming at a target to be verified and reported incorrectly or a target to be verified and reported incorrectly, the corresponding pixel regions may be input to a check network model to obtain a set of multi-dimensional feature vectors, simultaneously, the real false-report target or the missing-report target is input into the check network model to obtain another group of most characteristic vectors, and by calculating the difference between the classes between the two groups of corresponding multi-dimensional characteristic vectors, the corresponding inter-class difference may reflect the difference in pixel distribution between the pixel region with the false positive target to be verified and the pixel region with the false positive target, or the difference in pixel distribution between the pixel region having the false positive target to be verified and the pixel region having the false negative target, therefore, the technical means of comparing the inter-class difference can realize the quick solution of the target to be verified and misreported or the target to be verified and misreported.
Therefore, compared with the prior art, the technical scheme provided by the embodiment of the invention does not need to retrain the model, and can realize quick solution of error and missing detection in target detection.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
Fig. 1 is a schematic flow chart of a method for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 2 is a second schematic flowchart of a method for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 3 is a third schematic flowchart of a method for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 4 is a fourth flowchart illustrating a method for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 5 is a fifth flowchart illustrating a method for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 6 is a sixth flowchart illustrating a method for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 7 is a seventh schematic flowchart illustrating a method for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 8 is an eighth schematic flowchart illustrating a method for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 9 is a schematic structural diagram of an apparatus for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 10 is a second schematic structural diagram of an apparatus for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 11 is a third schematic structural diagram of an apparatus for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 12 is a fourth schematic view of a device for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
FIG. 13 is a fifth schematic view of a device for rapidly resolving false and missed detections in target detection according to an embodiment of the present invention;
fig. 14 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, 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.
The following describes a method for rapidly solving false detection and missed detection in target detection according to an embodiment of the present invention with reference to fig. 1 to 3.
In a scenario of rapidly solving the false detection, referring to fig. 1, the method specifically includes the following steps:
step 110: the video frame set FIS (frame image set) is input into the object detection model to generate an object set OS (object set) and a corresponding object region ORS (object region set).
Step 120: and for a false alarm target o-error (epsilon OS) pointed by a user, taking out the r-error (epsilon ORS) corresponding to the false alarm target o-error, sending a pixel region in fi-error (epsilon FIS) corresponding to the r-error into a check network model, and performing downsampling to generate a 128-dimensional characterization vector v-error (epsilon FVS-error), wherein 128 dimensions are only examples and can be set as required.
Step 130: v-error serialization was added to the FVS-error.
Step 140: and traversing the ORS, sending the fii (epsilon FIS) pixel area corresponding to the target ri to be verified into the verification network model, and performing downsampling to generate a 128-dimensional characterization vector vi (epsilon FIS).
Step 150: and deserializing the FVS-error.
Step 160: traversing the FVS-error, applying a Hilbert spatial norm | |2Calculating inter-class difference cse | | | vi-v-error | | |2V-error ∈ FVS-error. The inter-class difference corresponds to the vector difference embodiment above.
Step 170: if cse < preset threshold 0.2, it is determined that the region where ri is located is false, the target detection model has target false detection, and it is determined that fii corresponding to the target false detection model does not contain a target, oi cannot be added into a result target set ros (result of object set).
Step 180: if cse > the preset threshold value of 0.2, it is determined that the region where ri is located is not false alarm, the target detection model has no target false detection, the target detection result oi to be verified is credible, it is determined that fii corresponding to the target detection result oi to be verified contains the target, and the corresponding target detection result oi to be verified is output to the ROS.
In the above-described embodiments of the present invention, 0.2 is only one example of the preset threshold, and does not constitute a limitation to the scope of protection.
In the embodiment of the present invention, the target verification method is explained by taking the video frame set as an example, and if the user reports that the multiple video frames have false detections, the above step 120-180 may be performed on each video frame.
In a scenario of rapidly solving missed detection in target detection, referring to fig. 2, taking missed detection verification as an example, the method specifically includes the following steps:
step 210: the video frame set FIS (frame image set) is input into the object detection model to generate an object set OS (object set) and a corresponding object region ORS (object region set).
Step 220: and for a false report target o-mis (epsilon OS) pointed out by the user, taking out the r-mis (epsilon ORS) corresponding to the false report target o-mis, sending a pixel area in fi-mis (epsilon FIS) corresponding to the r-mis into a check network model, and performing downsampling to generate a 128-dimensional characterization vector v-mis (epsilon FVS-mis).
Step 230: the v-miss serialization was appended to the FVS-miss.
Step 240: and traversing the ORS, sending the pixel region in fii (e.g. FIS) corresponding to the target ri to be verified into the verification network model, and generating a 128-dimensional characterization vector vi (e.g. FIS) by downsampling with reference to the step 140, which is not described herein again.
Step 250: dat documents are deserialized to generate the FVS-miss.
Step 260: traversing the FVS-miss, applying a Hilbert space norm | |2Calculating the difference between classes cse | | | vi-v-miss | | |2，v-miss∈FVS-miss。
Step 270: if cse < the preset threshold value of 0.2, the region where ri is located is determined to be a false report, the target detection model has target false detection, the corresponding fii contains the target is determined, the detection result oi of the target to be verified is output to the ROS, and therefore the confidence coefficient of the ROS is improved.
Step 280: the ROS is output to the user interface.
Step 290: if cse is greater than 0.2, it is determined that no false alarm exists in the ri-located region, no target false detection exists in the target detection model, and it is determined that fii corresponding to the target false alarm does not contain a target.
By adopting the embodiments shown in fig. 1 and fig. 2, the target detection can be checked for false detection and missed detection, and the checking result has sufficient confidence.
As can be seen from the above, the embodiment of the present invention provides a method for rapidly solving false detection in target detection, and as shown in fig. 4, the method may include the following steps:
step 310: under the condition of acquiring a false alarm target to be verified, inputting a pixel region with the false alarm target to be verified into a verification network model, and outputting a first multi-dimensional feature vector generated by downsampling;
step 330: calculating a first inter-class difference between the first and second multi-dimensional feature vectors using a Hilbert spatial norm;
step 340: and if the first inter-class difference is smaller than a first preset threshold value, determining that the false alarm target to be verified is false alarm and the target detection model has target false detection.
Optionally, if the first inter-class difference is not smaller than the first preset threshold, determining that the false-positive target to be verified is not a false-positive target, and outputting a target detection result of the target to be verified to a result target set;
outputting the result target set to a user interface.
As can be seen from the above, an embodiment of the present invention further provides a method for quickly resolving missing detection in target detection, and as shown in fig. 4, the method may include the following steps:
step 410: under the condition of obtaining a target to be verified and missed reported, inputting a pixel area with the target to be verified and missed reported into the verification network model, and outputting a third multi-dimensional feature vector generated by downsampling;
step 420: inputting the pixel image with the missing report target into the check network model, and outputting a fourth multi-dimensional feature vector generated by downsampling;
step 430: calculating a second inter-class difference between the third and fourth multi-dimensional feature vectors using a hilbert spatial norm;
step 440: and if the difference between the second types is smaller than a second preset threshold value, determining that the target to be verified is not reported and the target detection model has target missing detection.
Optionally, outputting a target detection result of the target to be verified and missed to be reported to a result target set;
outputting the result target set to a user interface.
In addition, if the difference between the second classes is not smaller than a second preset threshold, it is determined that the target to be verified is not the target to be missed and the target detection model does not have target missing detection.
And the target to be checked and the target to be missed are input into the target detection model and output. Therefore, when the target detection model is used for carrying out target detection on the video frame set, at least one situation of a to-be-verified false alarm target and a to-be-verified missing alarm target may occur, and at the moment, the error and missing detection of the situation or the situations can be rapidly solved.
Optionally, the check network model is obtained by training a first residual error network, the target detection model is obtained by training a second residual error network, and a depth of the first residual error network is smaller than a depth of the second residual error network.
Optionally, the first residual network is resnet18, and the second residual network is resnet 50.
Optionally, referring to fig. 5, before inputting the pixel region with the target to be checked for false alarm into the check network model, or inputting the pixel region with the target to be checked for false alarm into the check network model, the method further includes the following steps:
step 510: inputting a video frame set into the target detection model, and outputting a target set and a corresponding target area;
step 520: traversing the target area to obtain the area of the false report target to be verified or the area of the missing report target to be verified;
step 530: and extracting a pixel area with the false alarm target to be verified or the pixel area with the missing target to be verified from the video frame set.
Optionally, before inputting the pixel image with the false positive target into the check network model, the method further comprises:
in response to an input of a user indicating the false-positive target in the target set, extracting a target region corresponding to the false-positive target from the target set, and extracting a pixel image with the false-positive target, which comprises the target region, from the video frame set; or
And responding to the input of a user indicating the false negative target in the target set, extracting a target area corresponding to the false negative target from the target set, and extracting a pixel image with the false negative target, which comprises the target area, from the video frame set.
Optionally, referring to fig. 6, the method for quickly solving the false detection and the missed detection in the target detection provided in the embodiment of the present invention may further include the following steps:
step 610: outputting the target detection result of the target to be verified and missed to be reported to a result target set;
step 620: outputting the result target set to a user interface.
Optionally, referring to fig. 7, the method for quickly resolving an error and a missed detection in target detection provided in the embodiment of the present invention further includes:
step 710: if the first inter-class difference is not smaller than the first preset threshold, determining that the false alarm target to be verified is not a false alarm target, and outputting a target detection result of the target to be verified to a result target set;
step 720: outputting the result target set to a user interface.
The embodiment of the invention also provides a target detection and verification method, which particularly realizes the quick solution of false detection and missed detection in target detection. Referring to fig. 8, the method specifically includes the following steps:
step 810: extracting a first feature vector from an image to be verified to which a target detection result to be verified belongs, wherein the target detection result to be verified is obtained by performing target detection on the image to be verified by using a target detection model based on a deep learning network;
step 820: extracting a second feature vector from a reference image to which a credible target detection result belongs;
step 830: comparing a vector difference between the first feature vector and the second feature vector;
step 840: and checking the target detection result to be checked according to the vector difference.
In the embodiment of the invention, the target detection model is obtained by training the deep learning network by adopting a sample, and is used for detecting a target from an image and outputting a target detection result, wherein the target detection result reflects the probability that the target is contained in the image to be verified.
In the embodiment of the invention, the detection result of the target to be verified is configured as the detection result of the target to be verified, which may have target false detection or target missed detection, so that the detection result of the target to be verified needs to be further verified.
By using the scheme of the embodiment of the invention, the first characteristic vector reflects the pixel distribution condition in the image to be checked, and the second characteristic vector reflects the pixel distribution condition in the reference image. The trusted target detection result for the reference image is configured such that the detection result of the target is trusted, so that the second feature vector can be used as a trusted reference. The second feature vector is used as a reference, the first feature vector can be compared, and the verification result can indicate whether the target detection result to be verified of the image to be verified is consistent with the credible target detection result.
In the embodiment of the present invention, the trusted target detection result may also be obtained by performing target detection on the reference image by using the target detection model. In this way, the target detection result to be verified and the credible target detection result are obtained by adopting the same target detection model, and the logicality of the target detection result is consistent.
In the embodiment of the invention, a verification instruction of a target detection result to be verified can be received before a first feature vector is extracted from an image to be verified to which the target detection result to be verified belongs;
and determining the image to be verified corresponding to the target detection result to be verified according to the verification instruction.
The check instruction plays a triggering role. And the checking instruction defines a target detection result to be checked and a corresponding image to be checked. In this way, the image to be verified can be obtained according to the verification instruction, and step 810 is started to be executed.
The verification instruction can show that the target detection result to be verified may have false detection or missing detection on the target, and the false detection means that a non-target in the image to be verified is mistakenly identified as a real target. And the missed detection means that the target in the image to be verified is regarded as a non-target.
Specifically, if the verification instruction shows that the detection result of the target to be verified is false detection, the reference image is a false detection image of the target in error. And if the verification instruction shows that the detection result of the target to be verified is missed, the reference image is a missed detection image of the missed detection target.
The receiving of the verification instruction for the target detection result to be verified may include receiving the verification instruction for the target detection result to be verified from the user side, and thus, when the verification result is obtained, the verification result is sent to the user side.
The embodiment of the invention provides the target detection service and the target detection verification service, and a user can submit the target detection service through the user terminal and further submit the target detection verification service in a verification instruction mode when a target detection result to be verified is not trusted.
The target detection and verification system or the target detection system and the user side of the embodiment of the invention can form a server-client side framework.
In an optional embodiment of the present invention, the user side may further be an outwardly extending interface of the target detection and verification system, and is configured to receive a verification instruction input by a user.
In the embodiment of the present invention, the target detection result may be directly verified by using the method shown in fig. 8 when the target detection result is obtained by performing the target detection on the image by using the target detection model without passing the verification instruction of the user terminal. If the target detection model performs target detection on a video frame or a plurality of images to obtain a plurality of target detection results, part or all of the target detection results can be selected to be verified by using the method shown in fig. 1.
In the embodiment of the present invention, a first feature vector is extracted from an image to be verified to which a target detection result to be verified belongs, and specifically, the first feature vector is extracted from the image to be verified by using a verification network model, which is obtained by training a first residual error network, so that in order to ensure the accuracy and reliability of the verification result, a second feature vector can be extracted from the reference image by using the verification network model.
The first feature vector and the second feature vector are extracted by adopting the same check network model, so that the first feature vector and the second feature vector with the same dimensionality can be ensured to be extracted, and the vector difference comparison method is more comparable when comparing the vector differences.
The residual error network (ResNet) is composed of a series of residual error blocks, one residual error block is divided into a direct mapping part and a residual error part, the residual error part comprises a convolutional layer, and the convolutional layer is used for carrying out vector conversion on an image to obtain a feature vector.
The depth of the residual network is not limited, and may be ResNet18 or ResNet34 or other depth setting, and may be selected as needed.
In an alternative embodiment of the present invention, the check network model may also select other deep learning models, such as Convolutional Neural Network (CNN), Deep Convolutional Neural Network (DCNN), deconvolution neural network (DN), Generative Adaptive Network (GAN), Recurrent Neural Network (RNN), long short term memory network (LSTM), and neural Network (NTM), which are not limited herein.
In the embodiment of the present invention, the target detection model may also be obtained by training the second residual error network, and the depth of the first residual error network is smaller than that of the second residual error network. The second residual error network sets a deeper depth, more features can be extracted from the image, and the detection result of the target to be verified is more accurate. In contrast, the first residual error network is set to be relatively shallow in depth, so that the checking efficiency can be improved.
In one embodiment, the second residual network may be ResNet50 and the first residual network may be ResNet 18. This is merely an example, and other depths may be selected as needed.
Thus, mask-RCNN was used as the target detection model, and resnet50 was used as the backbone network for target detection. And (3) using the resnet18 as a verification network model for the image to be verified, verifying the detection result of the target to be verified of the image to be verified, removing the false alarm target and improving the confidence coefficient of the target which is not reported.
In the embodiment of the present invention, the inter-class difference between the first feature vector and the second feature vector may be calculated by using a hilbert space, and the inter-class difference represents a vector difference.
In the embodiment of the present invention, if the trusted target detection result is a target false detection result, if the inter-class difference is smaller than a first preset threshold, it is determined that the target detection result to be verified is a target false detection result, and if the inter-class difference is larger than the first preset threshold, it is determined that the target is detected by the target detection result to be verified.
The target false detection result refers to that the target detection model performs target false detection on the reference image, namely the reference image does not contain the target but reports that the target is detected. Therefore, if the inter-class difference is smaller than the first preset threshold, it indicates that the pixel distribution of the image to be verified is close to the pixel distribution of the reference image, and therefore the probability that the target false detection exists in the target detection result to be verified is higher in the verification result, and the target detection result to be verified is the target false detection result in the verification result.
Otherwise, determining that the target detection result to be verified does not belong to target false detection, and determining that the target detection result to be verified has higher reliability in detecting the target.
In the embodiment of the present invention, if the trusted target detection result is a target missed detection result, if the inter-class difference is smaller than a second preset threshold, it is determined that the target detection result to be verified is the target missed detection result, and if the inter-class difference is larger than the second preset threshold, it is determined that the target is not detected by the target detection result to be verified.
The target missing detection result refers to the target missing detection of the reference image by the target detection model, namely the reference image contains the target but reports that the target is not detected. Therefore, if the inter-class difference is smaller than the second preset threshold, it indicates that the pixel distribution of the image to be verified is close to the pixel distribution of the reference image, and therefore the verification result is that the target detection result to be verified has a high probability of target omission, and therefore the verification result is that the target detection result to be verified has a high reliability in detecting the target.
Otherwise, determining that the target detection result to be verified does not belong to target missing detection, and the target detection result to be verified does not detect the target with higher reliability.
In the embodiment of the present invention, the first preset threshold and the second preset threshold may be the same or different, and may be selected as needed.
In the embodiment of the invention, the false detection check and the missing detection check can be simultaneously carried out on the same target detection result to be checked, and the false detection check or the missing detection check can also be carried out only.
The verification result may be output to a user interface.
And receiving a verification instruction for the target detection result to be verified from the user side correspondingly to the above, wherein the verification result of the target detection result to be verified sent to the user side is specifically a result of quickly solving false detection or a result of quickly solving missed detection.
In the embodiment of the invention, the detection result of the target to be verified can be subjected to false detection verification or missed detection verification independently, and can also be subjected to both false detection and missed detection, and the false detection and the missed detection can be operated sequentially or simultaneously.
The target detection and verification device provided by the present invention is described below, and the target detection and verification device described below and the target detection and verification method described above may be referred to in correspondence with each other.
Referring to fig. 9, the apparatus for rapidly resolving an error and a missing detection in target detection provided by the embodiment of the present invention includes:
a first extraction module 910, a second extraction module 920, a comparison module 930, and a verification module 930;
the first extraction module 910 is configured to, when a false positive target to be verified is obtained, input a pixel region having the false positive target to be verified into a verification network model, and output a first multi-dimensional feature vector generated by downsampling;
the second extraction module 920 is configured to input a pixel image with a false alarm target into the check network model, and output a second multi-dimensional feature vector generated by downsampling;
the comparing module 930 is configured to apply a hilbert spatial norm to calculate a first inter-class difference between the first multi-dimensional feature vector and the second multi-dimensional feature vector;
the checking module 940 is configured to determine that the target to be checked is a false positive and the target detection model has target false detection if the first inter-class difference is smaller than a first preset threshold;
alternatively, the first and second electrodes may be,
the first extraction module 910 is further specifically configured to, when a missing report target to be verified is obtained, input a pixel region having the missing report target to be verified into the verification network model, and output a third multidimensional feature vector generated by downsampling;
the second extracting module 920 is further specifically configured to input the pixel image with the target of missing report into the check network model, and output a fourth multidimensional feature vector generated by downsampling;
the comparing module 930 is further specifically configured to apply a hilbert space norm to calculate a second inter-class difference between the third multidimensional feature vector and the fourth multidimensional feature vector;
the checking module 940 is further specifically configured to determine that the target to be checked is not reported and the target detection model has target missing detection if the second inter-class difference is smaller than a second preset threshold.
Optionally, referring to fig. 10, the apparatus further includes:
the target detection module 1010 is configured to input a video frame set into the target detection model and output a target set and a corresponding target region before inputting a pixel region with a target to be verified and misreported into a verification network model or inputting a pixel region with a target to be verified and misreported into the verification network model;
an obtaining module 1020, configured to traverse the target region to obtain a region of the false alarm target to be verified or a region of the missing report target to be verified;
and a pixel region extracting module 1030, configured to extract a pixel region with the false alarm target to be verified or a pixel region with the missing report target to be verified from the video frame set.
Optionally, the pixel region extracting module 1030 is further specifically configured to, before inputting the pixel image with the false-positive target into the check network model, in response to an input that a user indicates the false-positive target in the target set, extract a target region corresponding to the false-positive target from the target set, and extract the pixel image with the false-positive target including the target region from the video frame set; or
And responding to the input of a user indicating the false negative target in the target set, extracting a target area corresponding to the false negative target from the target set, and extracting a pixel image with the false negative target, which comprises the target area, from the video frame set.
Optionally, referring to fig. 11, the apparatus may further include:
the first output module 1110 outputs the target detection result of the target to be verified and missed to be reported to a result target set;
a second output module 1120, which outputs the result target set to a user interface.
Optionally, referring to fig. 12, the apparatus may further include:
a third output module 1210, configured to determine that the false positive target to be verified is not a false positive target if the first inter-class difference is not smaller than the first preset threshold, and output a target detection result of the target to be verified to a result target set;
the fourth output module 1220, outputs the result target set to a user interface.
Referring to fig. 13, an embodiment of the present invention further provides a target detection and verification apparatus, which specifically includes the following modules:
a first extraction module 1310, configured to extract a first feature vector from an image to be verified to which a target detection result to be verified belongs, where the target detection result to be verified is obtained by performing target detection on the image to be verified by using a target detection model based on a deep learning network;
a second extraction module 1320, configured to extract a second feature vector from a reference image to which the trusted target detection result belongs;
a comparison module 1330 that compares the vector difference between the first feature vector and the second feature vector;
the checking module 1340 checks the target detection result to be checked according to the vector difference.
Optionally, the first extracting module 1310 specifically extracts a first feature vector from the image to be checked by using a check network model, where the check network model is obtained by training a first residual error network;
the second extraction module 1320 extracts a second feature vector from the reference image by using the check network model.
Optionally, the target detection model is obtained by training a second residual error network, and a depth of the first residual error network is smaller than a depth of the second residual error network.
Optionally, the comparing module 1330 is specifically configured to:
calculating an inter-class difference between the first feature vector and the second feature vector by using a Hilbert space, wherein the inter-class difference characterizes the vector difference;
the check module 1340 is specifically configured to:
if the target is detected by the trusted target detection result, determining that the target is contained in the image to be verified if the inter-class difference is smaller than a preset threshold value, and determining that the target is not contained in the image to be verified if the inter-class difference is larger than the preset threshold value;
if the target is not detected by the trusted target detection result, determining that the target is not included in the image to be verified if the inter-class difference is smaller than a preset threshold, and determining that the target is included in the image to be verified if the inter-class difference is larger than the preset threshold.
Optionally, the first extraction module 1310 is specifically configured to:
before extracting a first characteristic vector from an image to be verified to which a target detection result to be verified belongs, receiving a verification instruction of the target detection result to be verified;
and determining the image to be verified to which the target detection result to be verified belongs according to the verification instruction.
Optionally, the first extraction module 1310 is further specifically configured to:
receiving a verification instruction of the target detection result to be verified from a user side;
the check module 1340 is further configured to:
and sending a verification result of the target detection result to be verified to the user side.
Optionally, if the verification instruction shows that the detection result of the target to be verified is false detection, the reference image is a false detection image of the target to be false detected;
and if the verification instruction shows that the detection result of the target to be verified is missed detection, the reference image is a missed detection image of the target to be missed detected.
Fig. 14 illustrates a physical structure diagram of an electronic device, and as shown in fig. 14, the electronic device may include: a processor (processor)1410, a communication Interface (Communications Interface)1420, a memory (memory)1430 and a communication bus 1440, wherein the processor 1410, the communication Interface 1420 and the memory 1430 communicate with each other via the communication bus 1440. Processor 1410 may invoke logic instructions in memory 1430 to perform a target detection check method comprising:
under the condition that a false alarm target to be verified is obtained, inputting a pixel region with the false alarm target to be verified into a verification network model, outputting a first multi-dimensional feature vector generated by downsampling, inputting a pixel image with the false alarm target into the verification network model, outputting a second multi-dimensional feature vector generated by downsampling, calculating a first inter-class difference between the first multi-dimensional feature vector and the second multi-dimensional feature vector by applying a Hilbert space norm, and if the first inter-class difference is smaller than a first preset threshold value, determining that the false alarm target to be verified is false and the target detection model has target false detection; and/or
Under the condition that a target to be verified and missed in report is obtained, inputting a pixel area with the target to be verified and missed in report into the verification network model, outputting a third multi-dimensional feature vector generated by down sampling, inputting a pixel image with the target to be verified and missed in report into the verification network model, outputting a fourth multi-dimensional feature vector generated by down sampling, calculating a second inter-class difference between the third multi-dimensional feature vector and the fourth multi-dimensional feature vector by applying a Hilbert space norm, and if the second inter-class difference is smaller than a second preset threshold value, determining that the target to be verified and missed in report is missed and the target detection model has target missing detection;
and the target to be checked and the target to be missed are input into the target detection model and output.
In addition, the logic instructions in the memory 1430 may be implemented in software functional units and stored in a computer readable storage medium when sold or used as a stand-alone product. 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 storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, which when executed by a computer, enable the computer to perform the object detection verification method provided by the above methods, the method comprising:
extracting a first feature vector from an image to be verified to which a target detection result to be verified belongs, wherein the target detection result to be verified is obtained by performing target detection on the image to be verified by using a target detection model based on a deep learning network;
extracting a second feature vector from a reference image to which a credible target detection result belongs;
comparing a vector difference between the first feature vector and the second feature vector;
checking the target detection result to be checked according to the vector difference; or
Under the condition that a false alarm target to be verified is obtained, inputting a pixel region with the false alarm target to be verified into a verification network model, outputting a first multi-dimensional feature vector generated by downsampling, inputting a pixel image with the false alarm target into the verification network model, outputting a second multi-dimensional feature vector generated by downsampling, calculating a first inter-class difference between the first multi-dimensional feature vector and the second multi-dimensional feature vector by applying a Hilbert space norm, and if the first inter-class difference is smaller than a first preset threshold value, determining that the false alarm target to be verified is false and the target detection model has target false detection; and/or
Under the condition that a target to be verified and missed in report is obtained, inputting a pixel area with the target to be verified and missed in report into the verification network model, outputting a third multi-dimensional feature vector generated by down sampling, inputting a pixel image with the target to be verified and missed in report into the verification network model, outputting a fourth multi-dimensional feature vector generated by down sampling, calculating a second inter-class difference between the third multi-dimensional feature vector and the fourth multi-dimensional feature vector by applying a Hilbert space norm, and if the second inter-class difference is smaller than a second preset threshold value, determining that the target to be verified and missed in report is missed and the target detection model has target missing detection;
and the target to be checked and the target to be missed are input into the target detection model and output.
In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program, which when executed by a processor is implemented to perform the object detection and verification method provided above, the method including:
extracting a first feature vector from an image to be verified to which a target detection result to be verified belongs, wherein the target detection result to be verified is obtained by performing target detection on the image to be verified by using a target detection model based on a deep learning network;
extracting a second feature vector from a reference image to which a credible target detection result belongs;
comparing a vector difference between the first feature vector and the second feature vector;
checking the target detection result to be checked according to the vector difference; or
Under the condition that a false alarm target to be verified is obtained, inputting a pixel region with the false alarm target to be verified into a verification network model, outputting a first multi-dimensional feature vector generated by downsampling, inputting a pixel image with the false alarm target into the verification network model, outputting a second multi-dimensional feature vector generated by downsampling, calculating a first inter-class difference between the first multi-dimensional feature vector and the second multi-dimensional feature vector by applying a Hilbert space norm, and if the first inter-class difference is smaller than a first preset threshold value, determining that the false alarm target to be verified is false and the target detection model has target false detection; and/or
Under the condition that a target to be verified and missed in report is obtained, inputting a pixel area with the target to be verified and missed in report into the verification network model, outputting a third multi-dimensional feature vector generated by down sampling, inputting a pixel image with the target to be verified and missed in report into the verification network model, outputting a fourth multi-dimensional feature vector generated by down sampling, calculating a second inter-class difference between the third multi-dimensional feature vector and the fourth multi-dimensional feature vector by applying a Hilbert space norm, and if the second inter-class difference is smaller than a second preset threshold value, determining that the target to be verified and missed in report is missed and the target detection model has target missing detection;
and the target to be checked and the target to be missed are input into the target detection model and output.
The above-described embodiments of the apparatus are merely illustrative, and 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 network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for rapidly solving false detection and missed detection in target detection is characterized by comprising the following steps:
under the condition that a false alarm target to be verified is obtained, inputting a pixel region with the false alarm target to be verified into a verification network model, outputting a first multi-dimensional feature vector generated by downsampling, inputting a pixel image with the false alarm target into the verification network model, outputting a second multi-dimensional feature vector generated by downsampling, calculating a first inter-class difference between the first multi-dimensional feature vector and the second multi-dimensional feature vector by applying a Hilbert space norm, and if the first inter-class difference is smaller than a first preset threshold value, determining that the false alarm target to be verified is false and the target detection model has target false detection; and/or
Under the condition that a target to be verified and missed in report is obtained, inputting a pixel area with the target to be verified and missed in report into the verification network model, outputting a third multi-dimensional feature vector generated by down sampling, inputting a pixel image with the target to be verified and missed in report into the verification network model, outputting a fourth multi-dimensional feature vector generated by down sampling, calculating a second inter-class difference between the third multi-dimensional feature vector and the fourth multi-dimensional feature vector by applying a Hilbert space norm, and if the second inter-class difference is smaller than a second preset threshold value, determining that the target to be verified and missed in report is missed and the target detection model has target missing detection;
and the target to be checked and the target to be missed are input into the target detection model and output.
2. The method according to claim 1, wherein the check network model is trained on a first residual network, the target detection model is trained on a second residual network, and a depth of the first residual network is smaller than a depth of the second residual network.
3. The method of claim 2, wherein the first residual network is resnet18, and the second residual network is resnet 50.
4. The method for rapidly solving the false and missing detection in the target detection according to claim 1, wherein before inputting the pixel region with the target to be verified and false-positive into the verification network model or inputting the pixel region with the target to be verified and missing-positive into the verification network model, the method further comprises:
inputting a video frame set into the target detection model, and outputting a target set and a corresponding target area;
traversing the target area to obtain the area of the false report target to be verified or the area of the missing report target to be verified;
and extracting a pixel area with the false alarm target to be verified or the pixel area with the missing target to be verified from the video frame set.
5. The method of claim 4, wherein before inputting the pixel image with false positive target into the check network model, the method further comprises:
in response to an input of a user indicating the false-positive target in the target set, extracting a target region corresponding to the false-positive target from the target set, and extracting a pixel image with the false-positive target, which comprises the target region, from the video frame set; or
And responding to the input of a user indicating the false negative target in the target set, extracting a target area corresponding to the false negative target from the target set, and extracting a pixel image with the false negative target, which comprises the target area, from the video frame set.
6. The method for rapidly resolving the false and missed detection in the target detection according to claim 5, wherein the method further comprises:
outputting the target detection result of the target to be verified and missed to be reported to a result target set;
outputting the result target set to a user interface.
7. The method for rapidly resolving the false and missed detection in the target detection according to claim 1, wherein the method further comprises:
if the first inter-class difference is not smaller than the first preset threshold, determining that the false alarm target to be verified is not a false alarm target, and outputting a target detection result of the target to be verified to a result target set;
outputting the result target set to a user interface.
8. The utility model provides a device that mistake, hourglass were examined and are solved fast in target detection which characterized in that includes:
the device comprises a first extraction module, a second extraction module, a comparison module and a check module;
the first extraction module is used for inputting a pixel region with the false alarm target to be verified into a verification network model under the condition of acquiring the false alarm target to be verified, and outputting a first multi-dimensional feature vector generated by downsampling;
the second extraction module is used for inputting the pixel image with the false alarm target into the check network model and outputting a second multi-dimensional feature vector generated by downsampling;
the comparison module is used for calculating a first inter-class difference between the first multi-dimensional feature vector and the second multi-dimensional feature vector by applying a Hilbert space norm;
the checking module is used for determining that the target to be checked is false alarm and the target detection model has target false detection if the first inter-class difference is smaller than a first preset threshold; and/or
The first extraction module is specifically further configured to, under the condition that a target to be verified and missed-report is obtained, input a pixel region with the target to be verified and missed-report into the verification network model, and output a third multidimensional feature vector generated by downsampling;
the second extraction module is specifically further configured to input the pixel image with the under-sampling target into the check network model, and output a fourth multidimensional feature vector generated by down-sampling;
the comparison module is further specifically configured to apply a hilbert spatial norm to calculate a second inter-class difference between the third multidimensional feature vector and the fourth multidimensional feature vector;
the checking module is further specifically configured to determine that the target to be checked is false negative and the target detection model has target false negative if the second inter-class difference is smaller than a second preset threshold.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for rapidly resolving false or missed detection in object detection according to any one of claims 1 to 7.
10. A non-transitory computer readable storage medium, having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for quickly resolving false or missed detections in object detection according to any one of claims 1 to 7.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2021107208566 | 2021-06-28 | ||
| CN202110720856 | 2021-06-28 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113537145A true CN113537145A (en) | 2021-10-22 |
| CN113537145B CN113537145B (en) | 2024-02-09 |
Family
ID=78090425
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110892872.3A Active CN113537145B (en) | 2021-06-28 | 2021-08-04 | Method, device and storage medium for rapidly solving false detection and missing detection in target detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113537145B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116858943A (en) * | 2023-02-03 | 2023-10-10 | 台州五标机械股份有限公司 | Hollow shaft intelligent preparation method and system for new energy automobile |
| CN117710944A (en) * | 2024-02-05 | 2024-03-15 | 虹软科技股份有限公司 | Model defect detection method, model training method, target detection method and target detection system |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108334830A (en) * | 2018-01-25 | 2018-07-27 | 南京邮电大学 | A kind of scene recognition method based on target semanteme and appearance of depth Fusion Features |
| WO2019144469A1 (en) * | 2018-01-24 | 2019-08-01 | 华讯方舟科技有限公司 | Image quality classification method, system and terminal device |
| CN111275038A (en) * | 2020-01-17 | 2020-06-12 | 平安医疗健康管理股份有限公司 | Image text recognition method and device, computer equipment and computer storage medium |
| CN111639653A (en) * | 2020-05-08 | 2020-09-08 | 浙江大华技术股份有限公司 | False detection image determining method, device, equipment and medium |
| CN111680579A (en) * | 2020-05-22 | 2020-09-18 | 常州工业职业技术学院 | Remote sensing image classification method for adaptive weight multi-view metric learning |
| CN111753658A (en) * | 2020-05-20 | 2020-10-09 | 高新兴科技集团股份有限公司 | Post sleep warning method and device and computer equipment |
| CN111782833A (en) * | 2020-06-09 | 2020-10-16 | 南京理工大学 | Fine-grained cross-media retrieval method based on multi-model network |
| CN112861673A (en) * | 2021-01-27 | 2021-05-28 | 长扬科技（北京）有限公司 | False alarm removal early warning method and system for multi-target detection of surveillance video |
-
2021
- 2021-08-04 CN CN202110892872.3A patent/CN113537145B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019144469A1 (en) * | 2018-01-24 | 2019-08-01 | 华讯方舟科技有限公司 | Image quality classification method, system and terminal device |
| CN108334830A (en) * | 2018-01-25 | 2018-07-27 | 南京邮电大学 | A kind of scene recognition method based on target semanteme and appearance of depth Fusion Features |
| CN111275038A (en) * | 2020-01-17 | 2020-06-12 | 平安医疗健康管理股份有限公司 | Image text recognition method and device, computer equipment and computer storage medium |
| CN111639653A (en) * | 2020-05-08 | 2020-09-08 | 浙江大华技术股份有限公司 | False detection image determining method, device, equipment and medium |
| CN111753658A (en) * | 2020-05-20 | 2020-10-09 | 高新兴科技集团股份有限公司 | Post sleep warning method and device and computer equipment |
| CN111680579A (en) * | 2020-05-22 | 2020-09-18 | 常州工业职业技术学院 | Remote sensing image classification method for adaptive weight multi-view metric learning |
| CN111782833A (en) * | 2020-06-09 | 2020-10-16 | 南京理工大学 | Fine-grained cross-media retrieval method based on multi-model network |
| CN112861673A (en) * | 2021-01-27 | 2021-05-28 | 长扬科技（北京）有限公司 | False alarm removal early warning method and system for multi-target detection of surveillance video |
Non-Patent Citations (1)
| Title |
|---|
| 唐洪涛;: "利用滑动窗口检测器的多目标跟踪误报检测", 控制工程, no. 11 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116858943A (en) * | 2023-02-03 | 2023-10-10 | 台州五标机械股份有限公司 | Hollow shaft intelligent preparation method and system for new energy automobile |
| CN117710944A (en) * | 2024-02-05 | 2024-03-15 | 虹软科技股份有限公司 | Model defect detection method, model training method, target detection method and target detection system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113537145B (en) | 2024-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110569721B (en) | Recognition model training method, image recognition method, device, equipment and medium | |
| JP7058373B2 (en) | Lesion detection and positioning methods, devices, devices, and storage media for medical images | |
| CN109241985B (en) | Image identification method and device | |
| CN108960211B (en) | Multi-target human body posture detection method and system | |
| EP2806374A1 (en) | Method and system for automatic selection of one or more image processing algorithm | |
| CN110276257B (en) | Face recognition method, device, system, server and readable storage medium | |
| EP2660753B1 (en) | Image processing method and apparatus | |
| US9947164B2 (en) | Automatic fault diagnosis method and device for sorting machine | |
| CN111461243B (en) | Classification method, classification device, electronic equipment and computer-readable storage medium | |
| US7643674B2 (en) | Classification methods, classifier determination methods, classifiers, classifier determination devices, and articles of manufacture | |
| CN113537145B (en) | Method, device and storage medium for rapidly solving false detection and missing detection in target detection | |
| CN110853033A (en) | Video detection method and device based on inter-frame similarity | |
| CN113869449A (en) | Model training method, image processing method, device, equipment and storage medium | |
| CN110532746B (en) | Face checking method, device, server and readable storage medium | |
| CN109271957B (en) | Face gender identification method and device | |
| US11354923B2 (en) | Human body recognition method and apparatus, and storage medium | |
| KR101545809B1 (en) | Method and apparatus for detection license plate | |
| CN113033305B (en) | Living body detection method, living body detection device, terminal equipment and storage medium | |
| CN114448664A (en) | Phishing webpage identification method and device, computer equipment and storage medium | |
| CN111723688B (en) | Human body action recognition result evaluation method and device and electronic equipment | |
| CN116091781B (en) | Data processing method and device for image recognition | |
| CN114119970B (en) | Target tracking method and device | |
| CN113361455B (en) | Training method of face counterfeit identification model, related device and computer program product | |
| CN111209567B (en) | Method and device for judging perceptibility of improving robustness of detection model | |
| CN115083006A (en) | Iris recognition model training method, iris recognition method and iris recognition 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |