CN111177434B - Data reflow method for improving accuracy of cv algorithm - Google Patents
Data reflow method for improving accuracy of cv algorithm Download PDFInfo
- Publication number
- CN111177434B CN111177434B CN201911407242.1A CN201911407242A CN111177434B CN 111177434 B CN111177434 B CN 111177434B CN 201911407242 A CN201911407242 A CN 201911407242A CN 111177434 B CN111177434 B CN 111177434B
- Authority
- CN
- China
- Prior art keywords
- image
- algorithm
- processing
- detection result
- type
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Library & Information Science (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
- Studio Devices (AREA)
Abstract
The invention provides a data reflow method for improving the accuracy of a cv algorithm, which not only allows a user to perform label processing and storage processing on an image with an identification error so as to realize transferable editing on the image with the identification error, thereby facilitating the subsequent efficient classification processing and data reflow processing on the image with the identification error.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a data reflow method for improving the accuracy of a cv algorithm.
Background
At present, when a cv algorithm is used for training an algorithm model by using a corresponding training data set, when the model is trained and mature, the detection accuracy of scenes similar to the training data set can be improved, but in practical application, the training data sets cannot cover all scenes, so that the recognition effect on scenes inconsistent with the training data sets can be reduced, in addition, even in the same scene, if factors such as light interference, too far target distance, close background color to a detection target color value and the like occur, the recognition accuracy can be influenced, and images with wrong recognition of the cv algorithm have high utilization value on the directional optimization of the cv algorithm, and the higher fitting detection effect of the cv algorithm on more scenes can be realized. However, the conventional search interface generally displays the detected image result and the captured information thereof, and only allows the user to delete the image with the wrong identification, but does not effectively utilize the image with the wrong identification, but only performs screenshot processing or downloading processing on the image even if the image with the wrong identification is utilized, and then transmits the image to an algorithm maintainer for optimization.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides a data reflow method for improving the accuracy of a cv algorithm, which comprises the following steps: step S1, screening processing on preset conditions is carried out on image detection results of a cv algorithm, so that a corresponding image detection result list is obtained; s2, auditing one or more image attribute information in the image detection result list to identify the error existence condition in the image detection result; step S3, classifying the image detection result according to the identification result of the error existence condition to obtain a first type image set and a second type image set; s4, carrying out data reflow processing on the labeling of the first type of image set, and carrying out directional optimization processing on a cv algorithm according to the image subjected to the data reflow processing; therefore, the data reflow method for improving the accuracy of the cv algorithm not only allows a user to perform label processing and storage processing on the images with the identification errors so as to realize transferable editing on the images with the identification errors, thereby facilitating efficient classification processing and data reflow processing on the images with the identification errors, but also can perform directional optimization processing on the cv algorithm by utilizing the images with the identification errors after reflow, and further improves the detection accuracy of the cv algorithm and the applicability to different scenes.
The invention provides a data reflow method for improving the accuracy of a cv algorithm, which is characterized by comprising the following steps of:
step S1, screening processing on preset conditions is carried out on image detection results of a cv algorithm, so that a corresponding image detection result list is obtained;
s2, auditing one or more image attribute information in the image detection result list to identify the error existence condition in the image detection result;
step S3, classifying the image detection results according to the recognition results of the error existence condition to obtain a first type image set and a second type image set;
s4, carrying out data reflow processing on the labeling on the first type image set, and carrying out directional optimization processing on a cv algorithm according to the image subjected to the data reflow processing;
further, in the step S1, the image detection result of the cv algorithm is subjected to a screening process with respect to a predetermined condition, whereby a corresponding image detection result list is obtained specifically including,
step S101, integrating the image detection result of the cv algorithm into an image processing platform, and providing a corresponding condition screening area and an image detection result display area in an interface of the preset image processing platform;
step S102, selecting and setting corresponding screening conditions through the condition screening area;
step S103, carrying out screening processing on the image detection result according to the screening condition, and displaying the result of the screening processing in the image detection result display area;
further, in the step S101, integrating the image detection result of the cv algorithm into an image processing platform, and providing a corresponding condition screening area and an image detection result display area in an interface of the preset image processing platform specifically includes,
step S1011, obtaining image material scene attribute information corresponding to an image detection result of the cv algorithm, and performing precompression processing on the image detection result according to the image material scene attribute information;
step S1012, integrating the image detection result subjected to the precompression processing into the image processing platform according to a predetermined sequence mode;
step S1013, providing the condition screening area and the image detection result display area for two mutually non-overlapping areas of the interface of the image processing platform respectively;
or alternatively, the process may be performed,
in the step S102, the condition selection area is used to select and set the corresponding selection conditions including,
step S1021, providing a plurality of switchable selection screening conditions in the condition screening area, wherein the plurality of switchable selection screening conditions at least comprise an image static object screening condition, an image dynamic object screening condition and an image pixelation screening condition;
step S1022, selecting and setting the plurality of screening conditions which can be switched and selected to generate corresponding screening trigger instructions;
further, in the step S2, one or more image attribute information in the image detection result list is subjected to auditing processing to identify that an error exists in the image detection result specifically includes,
step S201, obtaining image scenes corresponding to all image results in the image detection result list, and calculating visual evaluation parameters of all image results according to the image scenes;
step S202, checking and processing the visual evaluation parameters and the one or more pieces of image attribute information in a comparison mode;
step S203, according to the result of the auditing process, identifying the error existence condition in the image detection result;
further, in the step S201, image scenes corresponding to all the image results in the image detection result list are obtained, and the visualized evaluation parameters of all the image results are calculated according to the image scenes,
acquiring image scenes corresponding to all image results, and performing pixel-level color, texture or contour calculation processing on the image scenes to calculate and obtain the visual evaluation parameters;
or alternatively, the process may be performed,
in the step S202, the process of checking the visual evaluation parameter against the one or more image attribute information specifically includes,
performing manual auditing processing of numerical value comparison on the visual evaluation parameters and the one or more pieces of image attribute information;
or alternatively, the process may be performed,
in the step S203, the step of recognizing that the error exists in the image detection result specifically includes,
if the result of the auditing process indicates that the visual evaluation parameter is not matched with the one or more image attribute information, determining that a cv algorithm detection error exists in the corresponding image detection result, otherwise, determining that the cv algorithm detection error does not exist in the corresponding image detection result;
further, in the step S3, the image detection result is classified according to the recognition result of the error existence condition, so as to obtain a first type image set and a second type image set, which specifically include,
step S301, if a certain image detection result is identified as having a cv algorithm detection error, classifying the corresponding certain image detection result as the first type image, and if a certain image detection result is identified as not having a cv algorithm detection error, classifying the corresponding certain image detection result as the second type image;
step S302, performing aggregation processing on the first type of images or the second type of images according to a time sequence and/or a place sequence corresponding to the image detection result to obtain a first type of image set or a second type of image set;
step S303, performing quality pre-judging processing on the images corresponding to the first type of image set or the second type of image set respectively so as to eliminate images which do not meet the preset quality condition;
further, in the step S302, the first type of images or the second type of images are subjected to a collection process according to a time sequence and/or a place sequence corresponding to the image detection result, so as to obtain the first type of image set or the second type of image set, which specifically includes,
step S3021, obtaining detection time information and/or scene location information corresponding to image detection processing of the specific target in the preset scene by using the image detection result;
step S3022, performing screening processing on the detection time information and/or the scene location information to obtain the time sequence and/or the location sequence;
step S3023, performing an arrangement type aggregation process on the first type of images or the second type of images according to the time sequence and/or the location sequence, so as to obtain the first type of image set or the second type of image set;
further, in the step S4, the data reflow processing concerning the labeling is performed on the first type image set, and the directional optimization processing for the cv algorithm is specifically performed according to the image subjected to the data reflow processing,
step S401, a cv algorithm detection error state of each image in the first type image set is obtained, and adaptive label picking processing is carried out according to the cv algorithm detection error state;
step S402, according to the result of the label choosing processing, classifying and storing all the images in the first type image set;
step S403, according to the result of the classified storage processing, exporting and refluxing the corresponding image to the corresponding cv algorithm processing center, and carrying out directional optimization processing on the cv algorithm through the exported image;
further, in the step S401, the adaptive label picking process according to the cv algorithm detecting the error state further specifically includes,
after adaptively selecting a label according to the error state detected by the cv algorithm, removing the image subjected to the label selecting process so as to enable the image to be in an unqueriable state;
or alternatively, the process may be performed,
in the step S402, according to the result of the label picking processing, the classification storage processing is performed on all the images in the first type image set, which specifically includes,
step S4021, determining all currently-checked tag information according to the result of the tag checking processing, and performing numerical conversion processing on all the tag information to obtain a plurality of identifiable numerical tags;
step S4022, storing all images in the first type image set to different storage spaces according to the plurality of identifiable digitized tags, so as to implement the classified storage processing;
or alternatively, the process may be performed,
in the step S403, according to the result of the classification storage processing, the corresponding image is exported and returned to the corresponding cv algorithm processing center, and the directional optimization processing for the cv algorithm by the exported image specifically includes,
step S4031, according to the result of the classified storage processing, periodically updating the stored image and exporting and refluxing the updated and stored image to a corresponding cv algorithm processing center;
step S4032, using the updated and stored image as output data of the cv algorithm to realize the directional optimization processing;
further, in the step S4, recording and auditing processing is further performed on all images in the second class image set, so that the images are in a queriable state;
or alternatively, the process may be performed,
before the step S1, a step S0 is further included, specifically, selecting, by the target server, a plurality of mobile terminals by using a preset selection method, wherein,
the target server sequentially transmits the image detection result of the cv algorithm through a plurality of movable terminals in a relay type so as to send the image detection result to a preset server, and the transmission of the image detection result is realized among different movable terminals through a Bluetooth network;
the preset selection method includes the steps of,
the total energy consumption of each group of alternative mobile terminals is calculated according to the following formula (1)
In the above formula (1), ni is the total energy consumption of the current group of alternative mobile terminals, T is the total number of mobile terminals in the current group of alternative mobile terminals, M is the data amount of the image detection result of the cv algorithm, and Q elec For the energy consumption, mu, of the Bluetooth module in each alternative mobile terminal fs D, for the energy consumption of the signal amplifying circuit corresponding to the Bluetooth module i,i+1 Alpha is the distance between the ith alternative movable terminal and the (i+1) th alternative movable terminal mov D, a mobile energy consumption factor for each alternative movable terminal i The total moving distance of the ith alternative movable terminal in the corresponding time period from the time when the image detection result is received to the time when the image detection result is transmitted to the (i+1) th alternative movable terminal;
finally, the corresponding group of alternative mobile terminals with the smallest total energy consumption is selected as the plurality of mobile terminals.
Compared with the prior art, the data reflow method for improving the accuracy of the cv algorithm not only allows a user to perform label processing and storage processing on the images with the identification errors so as to realize transferable editing on the images with the identification errors, thereby facilitating efficient classification processing and data reflow processing on the images with the identification errors, but also can perform directional optimization processing on the cv algorithm by utilizing the images with the identification errors after reflow, and further improves the detection accuracy of the cv algorithm and the applicability to different scenes.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a data reflow method for improving accuracy of cv algorithm.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, a flow chart of a data reflow method for improving accuracy of cv algorithm according to an embodiment of the invention is shown. The data reflow method for improving the accuracy of the cv algorithm comprises the following steps:
step S1, screening processing on preset conditions is carried out on image detection results of a cv algorithm, so that a corresponding image detection result list is obtained;
s2, auditing one or more image attribute information in the image detection result list to identify the error existence condition in the image detection result;
step S3, classifying the image detection result according to the identification result of the error existence condition to obtain a first type image set and a second type image set;
and S4, carrying out data reflow processing on the labeling of the first type of image set, and carrying out directional optimization processing on a cv algorithm according to the image subjected to the data reflow processing.
Preferably, in this step S1, the image detection result of the cv algorithm is subjected to a screening process with respect to a predetermined condition, whereby a corresponding image detection result list is obtained specifically including,
step S101, integrating the image detection result of the cv algorithm into an image processing platform, and providing a corresponding condition screening area and an image detection result display area in an interface of the preset image processing platform;
step S102, selecting and setting corresponding screening conditions through the condition screening area;
step S103, the screening processing is performed on the image detection result according to the screening condition, and the result of the screening processing is displayed in the image detection result display area.
Preferably, in the step S101, in integrating the image detection result of the cv algorithm into an image processing platform, and providing a corresponding condition screening area and an image detection result display area in an interface of the preset image processing platform specifically includes,
step S1011, obtaining image material scene attribute information corresponding to the image detection result of the cv algorithm, and performing precompression processing on the image detection result according to the image material scene attribute information;
step S1012, integrating the image detection result after the pre-compression processing into the image processing platform according to a predetermined sequence pattern;
in step S1013, the two mutually non-overlapping areas of the interface of the image processing platform are respectively provided with the condition filtering area and the image detection result display area.
Preferably, in the step S1011, the pre-compression processing of the image detection result includes,
according to the following formula (2), a collection node sequence of the image detection result is formulated
D＝((m1，f1),(m2，f2)......(m2n，f2n)) (2)
In the formula (2), (mj, fj) represents that the size of the corresponding acquired data amount at the acquisition node mj is fj, and 2n represents the total number of the acquisition nodes;
determining a fitted regression line for the sequence of acquisition nodes as a function of the position of the acquisition node mj in the image detection result as an independent variable and the size fj of the corresponding acquired data amount at the acquisition node mj as a dependent variable according to the following formula (3),
f＝γ 2 the value range of +alpha m +delta and gamma is (0, kappa) (3)
In the formula (3), alpha is a first parameter, delta is a second parameter, gamma is a third parameter, and kappa has a value smaller than 1 and larger than 0;
performing a linear fitting process on the first parameter α and the second parameter δ to obtain a regression equation represented by the following formula (4)
Obtaining a data compression algorithm of a unitary linear equation by using a regression equation expressed by the formula (4), and pre-compressing the image detection result by using the data compression algorithm, wherein the data compression algorithm is as follows:
for the data compression algorithm of a single acquisition node, according to the acquisition node sequence and a preset fitting regression line L (m), the absolute value of the difference between the fitting value of the local 2n+1 acquisition nodes and the size f2n+1 of the acquired data quantity of the 2n+1 acquisition nodes is smaller than gamma 2 When a new acquisition node sequence d= ((m 1, f 1), (m 2, f 2));
for a data compression algorithm of the aggregation node, noise data elimination is carried out on collected data of all nodes before the aggregation node, an average value of the collected data of all the nodes before the aggregation node is determined, the obtained collected data is applied to the process similar to that of a single collected node, a corresponding fitting regression line set is calculated, and a starting time point and a stopping time point of the fitting regression line set and a corresponding regression coefficient are concentrated to finally generate compressed data;
through the precompression processing, the data with noise eliminated can be obtained by constructing a proper acquisition node sequence mathematical model, so that the data volume is smaller than the original transmission data, and the data is transmitted after being compressed, thereby greatly reducing the energy consumption of network communication.
Preferably, in this step S102, the selection and setting of the corresponding screening conditions by the condition screening area specifically includes,
step S1021, providing a plurality of switchable selection screening conditions in the condition screening area, wherein the plurality of switchable selection screening conditions at least comprise an image static object screening condition, an image dynamic object screening condition and an image pixelation screening condition;
step S1022, selecting and setting the plurality of switchable selected screening conditions to generate corresponding screening trigger instructions.
Preferably, in the step S2, one or more image attribute information in the image detection result list is subjected to an audit process to identify that an error exists in the image detection result specifically includes,
step S201, obtaining image scenes corresponding to all image results in the image detection result list, and calculating visual evaluation parameters of all image results according to the image scenes;
step S202, checking the contrast type of the visual evaluation parameter and the one or more image attribute information;
step S203, according to the result of the auditing process, the error existence condition in the image detection result is identified.
Preferably, in the step S201, image scenes corresponding to all the image results in the image detection result list are acquired, and the visual evaluation parameters of all the image results are calculated according to the image scenes,
and acquiring image scenes corresponding to all image results, and performing pixel-level color, texture or contour calculation processing on the image scenes to calculate and obtain the visual evaluation parameters.
Preferably, in the step S202, the process of checking the visual evaluation parameter against the one or more image attribute information specifically includes,
and comparing the visual evaluation parameter with the one or more image attribute information in numerical value.
Preferably, in the step S203, the identifying of the error presence in the image detection result specifically includes,
if the result of the auditing process indicates that the visual evaluation parameter is not matched with the one or more image attribute information, the corresponding image detection result is determined to have the cv algorithm detection error, otherwise, the corresponding image detection result is determined to have no cv algorithm detection error.
Preferably, in the step S3, the image detection result is classified according to the recognition result of the error existence condition, so as to obtain a first type image set and a second type image set specifically including,
step S301, if a certain image detection result is identified as having a cv algorithm detection error, classifying the corresponding certain image detection result as the first type image, and if a certain image detection result is identified as not having a cv algorithm detection error, classifying the corresponding certain image detection result as the second type image;
step S302, performing aggregation processing on the first type of images or the second type of images according to a time sequence and/or a place sequence corresponding to the image detection result to obtain a first type of image set or a second type of image set;
step S303, performing quality pre-judging processing on the images corresponding to the first type of image set or the second type of image set respectively so as to eliminate images which do not meet the preset quality condition.
Preferably, in the step S302, the first type of image or the second type of image is subjected to a collection process according to a time sequence and/or a location sequence corresponding to the image detection result, so as to obtain the first type of image set or the second type of image set specifically including,
step S3021, obtaining detection time information and/or scene location information corresponding to image detection processing of a specific target in a preset scene by using the image detection result;
step S3022, performing screening processing on the detection time information and/or the scene location information to obtain the time sequence and/or the location sequence;
step S3023, performing a permutation type aggregation process on the first type of image or the second type of image according to the time sequence and/or the location sequence to obtain the first type of image set or the second type of image set.
Preferably, in the step S4, the first type image set is subjected to a data reflow process concerning labeling, and the performing a directional optimization process on the cv algorithm based on the image subjected to the data reflow process specifically includes,
step S401, a cv algorithm detection error state of each image in the first type image set is obtained, and adaptive label picking processing is carried out according to the cv algorithm detection error state;
step S402, according to the result of the label choosing processing, classifying and storing all the images in the first type image set;
step S403, according to the result of the classified storage processing, exporting and refluxing the corresponding image to the corresponding cv algorithm processing center, and carrying out directional optimization processing on the cv algorithm through the exported image.
Preferably, in the step S401, the adaptive label picking process according to the cv algorithm detecting the error state further specifically includes,
and after adaptively checking the label according to the cv algorithm detection error state, removing the image subjected to the checked label processing so as to enable the image to be in an unqueriable state.
Preferably, in the step S402, according to the result of the label picking process, the classification storage process is performed on all the images in the first type image set, which specifically includes,
step S4021, determining all currently-checked tag information according to the result of the checked tag processing, and performing numeric conversion processing on all the tag information to obtain a plurality of identifiable numeric tags;
step S4022, storing all the images in the first type image set into different storage spaces according to the plurality of identifiable digitized labels, so as to implement the classified storage processing.
Preferably, in the step S403, the deriving of the corresponding image is returned to the corresponding cv algorithm processing center according to the result of the classification storage processing, and the performing the directional optimization processing on the cv algorithm by the derived image specifically includes,
step S4031, according to the result of the classified storage processing, periodically updating the stored image and exporting and refluxing the updated and stored image to the corresponding cv algorithm processing center;
and step S4032, taking the updated and stored image as output data of the cv algorithm to realize the directional optimization processing.
Preferably, in the step S4, recording and auditing processing is further performed on all images in the second type of image set, so that the images are in a queriable state.
Preferably, before this step S1, a step S0 is further included, in which a plurality of mobile terminals are selected by the target server using a preset selection method, wherein,
the target server sequentially transmits the image detection result of the cv algorithm through a plurality of movable terminals in a relay type so as to send the image detection result to a preset server, and the transmission of the image detection result is realized among different movable terminals through a Bluetooth network;
the preset selection method includes the steps of,
the total energy consumption of each group of alternative mobile terminals is calculated according to the following formula (1)
In the above formula (1), ni is the total energy consumption of the current group of alternative mobile terminals, and T is the current group of alternative mobile terminalsThe total number of the movable terminals, M is the data size of the image detection result of the cv algorithm, Q elec For the energy consumption, mu, of the Bluetooth module in each alternative mobile terminal fs D, for the energy consumption of the signal amplifying circuit corresponding to the Bluetooth module i,i+1 Alpha is the distance between the ith alternative movable terminal and the (i+1) th alternative movable terminal mov D, a mobile energy consumption factor for each alternative movable terminal i The total moving distance of the ith alternative movable terminal in the corresponding time period from the time when the image detection result is received to the time when the image detection result is transmitted to the (i+1) th alternative movable terminal;
finally, selecting the corresponding group of alternative movable terminals with the smallest total energy consumption as the plurality of movable terminals;
therefore, the image detection result is transmitted between different servers by utilizing a plurality of alternative movable terminals, so that the directional optimization processing of the cv algorithm can be finished under the condition that the remote network connection between the different servers is lacked, and the energy consumption of the image detection result in the transmission process can be effectively reduced.
As can be seen from the foregoing embodiments, the data reflow method for improving the accuracy of the cv algorithm not only allows the user to perform tag processing and storage processing on the image with the identification error, so as to implement transferable editing on the image with the identification error, thereby facilitating efficient classification processing and data reflow processing on the image with the identification error, but also can perform directional optimization processing on the cv algorithm by using the image with the identification error after reflow, thereby improving the detection accuracy of the cv algorithm and the applicability to different scenes.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (8)
1. The data reflow method for improving the accuracy of the cv algorithm is characterized by comprising the following steps of:
step S1, screening processing on preset conditions is carried out on image detection results of a cv algorithm, so that a corresponding image detection result list is obtained;
s2, auditing one or more image attribute information in the image detection result list to identify the error existence condition in the image detection result;
step S201, obtaining image scenes corresponding to all image results in the image detection result list, and calculating visual evaluation parameters of all image results according to the image scenes;
step S202, checking and processing the visual evaluation parameters and the one or more pieces of image attribute information in a comparison mode;
step S203, according to the result of the auditing process, identifying the error existence condition in the image detection result;
step S3, classifying the image detection results according to the recognition results of the error existence condition to obtain a first type image set and a second type image set;
step S301, if a certain image detection result is identified as having a cv algorithm detection error, classifying the corresponding certain image detection result as the first type image, and if a certain image detection result is identified as not having a cv algorithm detection error, classifying the corresponding certain image detection result as the second type image;
step S302, performing aggregation processing on the first type of images or the second type of images according to a time sequence and/or a place sequence corresponding to the image detection result to obtain a first type of image set or a second type of image set;
step S303, performing quality pre-judging processing on the images corresponding to the first type of image set or the second type of image set respectively so as to eliminate images which do not meet the preset quality condition;
and S4, carrying out data reflow processing on the labeling on the first type image set, and carrying out directional optimization processing on a cv algorithm according to the image subjected to the data reflow processing.
2. The data reflow method of claim 1, wherein the accuracy of the cv algorithm is improved by:
in said step S1, the image detection result of the cv algorithm is subjected to a screening process with respect to a predetermined condition, whereby a corresponding image detection result list is obtained specifically including,
step S101, integrating the image detection result of the cv algorithm into an image processing platform, and providing a corresponding condition screening area and an image detection result display area in an interface of the image processing platform;
step S102, selecting and setting corresponding screening conditions through the condition screening area;
step S103, performing the screening process on the image detection result according to the screening condition, and displaying the result of the screening process in the image detection result display area.
3. The data reflow method of claim 2, wherein the accuracy of the cv algorithm is improved by:
in the step S101, integrating the image detection result of the cv algorithm into an image processing platform, and providing a corresponding condition screening area and an image detection result display area in an interface of the image processing platform specifically includes,
step S1011, obtaining image material scene attribute information corresponding to an image detection result of the cv algorithm, and performing precompression processing on the image detection result according to the image material scene attribute information;
step S1012, integrating the image detection result subjected to the precompression processing into the image processing platform according to a predetermined sequence mode;
step S1013, providing the condition screening area and the image detection result display area for two mutually non-overlapping areas of the interface of the image processing platform respectively;
or alternatively, the process may be performed,
in the step S102, the condition selection area is used to select and set the corresponding selection conditions including,
step S1021, providing a plurality of switchable selection screening conditions in the condition screening area, wherein the plurality of switchable selection screening conditions at least comprise an image static object screening condition, an image dynamic object screening condition and an image pixelation screening condition;
step S1022, selecting and setting the plurality of screening conditions capable of being selected in a switching manner to generate a corresponding screening trigger instruction.
4. The data reflow method of claim 1, wherein the accuracy of the cv algorithm is improved by:
in the step S201, image scenes corresponding to all the image results in the image detection result list are obtained, and the visual evaluation parameters of all the image results are calculated according to the image scenes,
acquiring image scenes corresponding to all image results, and performing pixel-level color, texture or contour calculation processing on the image scenes to calculate and obtain the visual evaluation parameters;
or alternatively, the process may be performed,
in the step S202, the process of comparing the visual evaluation parameter with the one or more image attribute information specifically includes a manual checking process of comparing the visual evaluation parameter with the one or more image attribute information in terms of value;
or alternatively, the process may be performed,
in the step S203, the step of recognizing that the error exists in the image detection result specifically includes,
and if the result of the auditing processing indicates that the visual evaluation parameter is not matched with the one or more image attribute information, determining that the corresponding image detection result has the cv algorithm detection error, otherwise, determining that the corresponding image detection result has no cv algorithm detection error.
5. The data reflow method of claim 1, wherein the accuracy of the cv algorithm is improved by:
in the step S302, the first type of images or the second type of images are subjected to a collection process according to a time sequence and/or a place sequence corresponding to the image detection result, so as to obtain the first type of image set or the second type of image set, which specifically includes,
step S3021, obtaining detection time information and/or scene location information corresponding to image detection processing of the specific target in the preset scene by using the image detection result;
step S3022, performing screening processing on the detection time information and/or the scene location information to obtain the time sequence and/or the location sequence;
step S3023, performing an arrangement type aggregation process on the first type of images or the second type of images according to the time sequence and/or the location sequence, so as to obtain the first type of image set or the second type of image set.
6. The data reflow method of claim 1, wherein the accuracy of the cv algorithm is improved by:
in the step S4, the data reflow processing concerning labeling is performed on the first type image set, and the directional optimization processing for the cv algorithm is specifically performed according to the image subjected to the data reflow processing,
step S401, a cv algorithm detection error state of each image in the first type image set is obtained, and adaptive label picking processing is carried out according to the cv algorithm detection error state;
step S402, according to the result of the label choosing processing, classifying and storing all the images in the first type image set;
step S403, exporting and reflowing the corresponding image to the corresponding cv algorithm processing center according to the result of the classification storage processing, and performing directional optimization processing on the cv algorithm through the exported image.
7. The method for improving accuracy of cv algorithm according to claim 6, wherein:
in the step S401, the adaptive label picking process according to the cv algorithm detecting the error state further specifically includes,
after adaptively selecting a label according to the error state detected by the cv algorithm, removing the image subjected to the label selecting process so as to enable the image to be in an unqueriable state;
or alternatively, the process may be performed,
in the step S402, according to the result of the label picking processing, the classification storage processing is performed on all the images in the first type image set, which specifically includes,
step S4021, determining all currently-checked tag information according to the result of the tag checking processing, and performing numerical conversion processing on all the tag information to obtain a plurality of identifiable numerical tags;
step S4022, storing all images in the first type image set to different storage spaces according to the plurality of identifiable digitized tags, so as to implement the classified storage processing;
or alternatively, the process may be performed,
in the step S403, according to the result of the classification storage processing, the corresponding image is exported and returned to the corresponding cv algorithm processing center, and the directional optimization processing for the cv algorithm by the exported image specifically includes,
step S4031, according to the result of the classified storage processing, periodically updating the stored image and exporting and refluxing the updated and stored image to a corresponding cv algorithm processing center;
and step S4032, taking the updated and stored image as output data of the cv algorithm to realize the directional optimization processing.
8. The data reflow method of claim 1, wherein the accuracy of the cv algorithm is improved by:
in the step S4, recording and auditing are further performed on all images in the second class image set, so that the images are in a queriable state;
or alternatively, the process may be performed,
before the step S1, a step S0 is further included, in which a target server selects a plurality of mobile terminals by using a preset selection method, where the target server sequentially transmits the image detection result of the cv algorithm to the preset server through the plurality of mobile terminals in a relay type, so as to transmit the image detection result to the preset server, and the transmission of the image detection result between different mobile terminals is realized through a bluetooth network;
the preset selection method includes the steps of,
the total energy consumption of each group of alternative mobile terminals is calculated according to the following formula (1)
In the above formula (1), ni is the total energy consumption of the current group of alternative mobile terminals, T is the total number of mobile terminals in the current group of alternative mobile terminals, M is the data amount of the image detection result of the cv algorithm, and Q elec For the energy consumption, mu, of the Bluetooth module in each alternative mobile terminal fs D, for the energy consumption of the signal amplifying circuit corresponding to the Bluetooth module i,i+1 Alpha is the distance between the ith alternative movable terminal and the (i+1) th alternative movable terminal mov D, a mobile energy consumption factor for each alternative movable terminal i The total moving distance of the ith alternative movable terminal in the corresponding time period from the time when the image detection result is received to the time when the image detection result is transmitted to the (i+1) th alternative movable terminal;
finally, the corresponding group of alternative mobile terminals with the smallest total energy consumption is selected as the plurality of mobile terminals.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911407242.1A CN111177434B (en) | 2019-12-31 | 2019-12-31 | Data reflow method for improving accuracy of cv algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911407242.1A CN111177434B (en) | 2019-12-31 | 2019-12-31 | Data reflow method for improving accuracy of cv algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111177434A CN111177434A (en) | 2020-05-19 |
| CN111177434B true CN111177434B (en) | 2023-09-05 |
Family
ID=70650555
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911407242.1A Active CN111177434B (en) | 2019-12-31 | 2019-12-31 | Data reflow method for improving accuracy of cv algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111177434B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764371A (en) * | 2018-06-08 | 2018-11-06 | Oppo广东移动通信有限公司 | Image processing method, device, computer readable storage medium and electronic equipment |
| CN110443141A (en) * | 2019-07-08 | 2019-11-12 | 深圳中兴网信科技有限公司 | Data set processing method, data set processing unit and storage medium |
| CN110490238A (en) * | 2019-08-06 | 2019-11-22 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10810465B2 (en) * | 2017-06-30 | 2020-10-20 | Datalogic Usa, Inc. | Systems and methods for robust industrial optical character recognition |
-
2019
- 2019-12-31 CN CN201911407242.1A patent/CN111177434B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764371A (en) * | 2018-06-08 | 2018-11-06 | Oppo广东移动通信有限公司 | Image processing method, device, computer readable storage medium and electronic equipment |
| CN110443141A (en) * | 2019-07-08 | 2019-11-12 | 深圳中兴网信科技有限公司 | Data set processing method, data set processing unit and storage medium |
| CN110490238A (en) * | 2019-08-06 | 2019-11-22 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111177434A (en) | 2020-05-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110245213A (en) | Questionnaire generation method, device, equipment and storage medium | |
| CN109120429B (en) | Risk identification method and system | |
| CN105553769A (en) | Data collecting-analyzing system and method | |
| CN110348519A (en) | Financial product cheats recognition methods and the device of clique | |
| CN112001274A (en) | Crowd density determination method, device, storage medium and processor | |
| CN108389070A (en) | A kind of customer action characteristic analysis method, server and system | |
| CN108579094A (en) | A kind of user interface detection method and relevant apparatus, system and storage medium | |
| CN110475124A (en) | Video cardton detection method and device | |
| CN111090822A (en) | Business object pushing method and device | |
| CN105373293A (en) | Data acquisition method and apparatus | |
| CN108334444A (en) | Various dimensions dynamic combined shunts method of servicing, device, terminal and storage medium | |
| CN111931809A (en) | Data processing method and device, storage medium and electronic equipment | |
| CN107911397A (en) | A kind of intimidation estimating method and device | |
| CN111652661B (en) | Mobile phone client user loss early warning processing method | |
| CN110209551A (en) | A kind of recognition methods of warping apparatus, device, electronic equipment and storage medium | |
| CN110610169A (en) | Picture labeling method and device, storage medium and electronic device | |
| CN105991722A (en) | Downloader recommendation method, application server, terminal and system | |
| CN109710827B (en) | Picture attribute management method and device, picture server and business processing terminal | |
| CN111177434B (en) | Data reflow method for improving accuracy of cv algorithm | |
| CN108197050B (en) | Equipment identification method, device and system | |
| CN112215509A (en) | Resource parameter determination method, device and equipment | |
| CN106651183A (en) | Communication data security audit method and device for industrial control system | |
| CN111092764A (en) | Real-time dynamic intimacy relationship analysis method and system | |
| CN104462422A (en) | Object processing method and device | |
| CN106681803A (en) | Task scheduling method and server |
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 |