CN109918523A - A kind of circuit board element detection method based on YOLO9000 algorithm - Google Patents
A kind of circuit board element detection method based on YOLO9000 algorithm Download PDFInfo
- Publication number
- CN109918523A CN109918523A CN201910114795.1A CN201910114795A CN109918523A CN 109918523 A CN109918523 A CN 109918523A CN 201910114795 A CN201910114795 A CN 201910114795A CN 109918523 A CN109918523 A CN 109918523A
- Authority
- CN
- China
- Prior art keywords
- circuit board
- yolo9000
- classification
- electronic component
- picture
- 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
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of circuit board element detection methods based on YOLO9000 algorithm, comprising the following steps: S1: building includes the database of circuit board picture；S2: database training YOLO9000 algorithm model is utilized, trained YOLO9000 algorithm model is obtained；S3: circuit board picture to be detected is obtained；S4: circuit board picture to be detected is detected using trained YOLO9000 algorithm model；S4: output result.The present invention trains YOLO9000 algorithm model, it can carry out the model classification identification of electronic component, positioning on picture position is carried out to electronic component, reach the model classification of detection electronic component, position, whether weld mistake, simultaneously as on circuit board the electronic component of polarized welding direction, classification problem is converted by direction discernment problem, realizes the direction discernment to the electronic component of polarized.
Description
Technical field
The present invention relates to computer vision fields, more particularly, to a kind of circuit board member based on YOLO9000 algorithm
Device inspection method.
Background technique
It is essential link that detection is carried out in industrial production electronic product, after welding electronic component, and present
Detection mode be borrow electronic instrument according to being detected, waste manpower and material resources to a certain extent.And the anti-polarized of weldering
Electronic component will cause bigger loss when being powered detection.Deep learning in 2012 in Krizhevsky et al. since mentioning
The depth convolutional neural networks (DCNN) for being called AlexNet are gone out, it is in large-scale visual identity challenge (ILSRVC)
In realize record-breaking image classification accuracy rate.Subsequent deep learning has development at full speed, YOLO in object detection field
The algorithm of series constantly emerges in large numbers.YOLO9000 has very excellent performance on recognition accuracy, and in medical image identification, text
It is widely used in word identification, recognition of face.Existing realization electronic component sorting technique disadvantage is as follows:
1, the direction of polarized component cannot be identified；
2, manual extraction feature is needed, can only realize simple classification, can not achieve complicated classification；
3, the component on picture cannot be classified and is positioned simultaneously.
Summary of the invention
The present invention provides a kind of circuit board element detection method based on YOLO9000 algorithm, is able to carry out electronics member device
The classification identification of part can also carry out the positioning on picture position to it, reach all component models in detection circuit board
Whether classification, direction, position weld mistake.
In order to solve the above technical problems, technical scheme is as follows:
A kind of circuit board element detection method based on YOLO9000 algorithm, comprising the following steps:
S1: building includes the database of circuit board picture, is further divided into four classes to the electronic component of polarized, wherein with circuit
On the basis of 12 o'clock of plate picture direction, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of upper direction；With 3 points of circuit board picture
On the basis of clock direction, 45 degree counterclockwise to 45 degree of ranges clockwise are right direction classification；Using 6 o'clock of circuit board picture direction as base
Standard, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of lower direction；On the basis of 9 o'clock of circuit board picture direction, counterclockwise
45 degree are that left direction is classified to 45 degree of ranges clockwise；
S2: database training YOLO9000 algorithm model is utilized, trained YOLO9000 algorithm model is obtained；
S3: circuit board picture to be detected is obtained；
S4: circuit board picture to be detected is detected using trained YOLO9000 algorithm model；
S5: output result.
In above scheme, YOLO9000 algorithm is capable of the feature of oneself study perception object, does not need manual extraction feature
Hundreds of classification is detected simultaneously, and obtains good effect, the even up to effect of real-time detection, while being capable of oneself study
The feature of object is perceived, while detecting hundreds of classification, and be able to detect position of the object in picture.
Preferably, database of the building including circuit board picture in step S1, comprising the following steps:
S1.1: the information list of circuit board element is extracted；
S1.2: it is shot according to component of the information list to corresponding model, generates the first data set；
S1.3: the enhancing of first time data is carried out to the first data set, generates the second data set；
S1.4: second of data enhancing is carried out to the first data set and the second data set, generates third data set；
S1.5: the classification and position of electronic component in the first data set, the second data set and third data set are marked；
S1.6: the data set marked is divided into training set, verifying collection and test set, as database.
Preferably, the enhancing of first time data is carries out data enhancing using generation confrontation network, due to photographed data collection sample
This is limited, and the training of YOLO9000 algorithm needs a large amount of data set, to the first data set with generate confrontation network (GAN) into
The enhancing of row data, generates the second data set,.
Preferably, second of data enhancing is to be rotated by 90 ° to the first data set and the second data set, rotate 180
Degree, 270 degree of rotation, flip horizontal, flip vertical and horizontal vertical overturn six kinds of modes and carry out data enhancing.
Preferably, to the class of electronic component in the first data set, the second data set and third data set in step S1.5
It is not marked with position, four classes is further divided into the electronic component of polarized, wherein be with 12 o'clock of circuit board picture direction
Benchmark, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of upper direction；On the basis of 3 o'clock of circuit board picture direction, the inverse time
45 degree of needle are that right direction is classified to 45 degree of ranges clockwise；On the basis of 6 o'clock of circuit board picture direction, 45 degree counterclockwise to suitable
45 degree of ranges of hour hands are the classification of lower direction；On the basis of 9 o'clock of circuit board picture direction, 45 degree counterclockwise to 45 degree of models clockwise
It encloses for left direction classification；
Since the component welding on circuit board is almost the two horizontal and vertical directions, polarized electronic component is come
There are four saying, it may be assumed that upwards, to the left, downwards, to the right.Since there are no algorithms can specifically identify an object for deep learning
Orientation angle, so direction discernment problem is converted into classification problem.
Preferably, database training YOLO9000 algorithm model is utilized in step S2, is obtained trained YOLO9000 and is calculated
Method model, comprising the following steps:
S2.1: training set input YOLO9000 algorithm model is subjected to propagated forward, exports calculated result；
S2.2: loss calculating is carried out according to calculated result, and judges whether the loss average value of nearest 50 iteration is less than and sets
Whether fixed value, judgement reach greatest iteration step number, if the two have one is, deconditioning exports trained
YOLO9000 algorithm model executes step S2.4；If not being, S2.3 is thened follow the steps:
S2.3: the parameter and return step S2.1 of each neural network in YOLO9000 algorithm model are updated according to the value of loss；
S2.4: trained YOLO9000 algorithm model is tested using test set, exports all kinds of accuracy of the mean AP
And average AP value MAP；
S2.5: judging whether all kinds of accuracy of the mean AP and average AP value MAP reach requirement, if reached, exports final
Otherwise parameter is adjusted in YOLO9000 algorithm model, return to S2.1 re -training YOLO9000 algorithm model.
Preferably, circuit board picture to be detected is obtained in S3, comprising the following steps:
S3.1: it in the case where the position that focal length, the distance apart from circuit board, circuit board are placed is fixed, is imaged using industry
Head shoots circuit board to be detected；
S3.2: detection circuit board profile extracts the picture of circuit board section, obtains original picture
S3.3: to extract circuit board section picture be rotated by 90 °, rotate 180 degree, rotation 270 degree, flip horizontal, vertically
Overturning, horizontal vertical turning operation obtain seven pictures including original picture, are to improve test with plurality of pictures test purpose
Accuracy rate；
S3.4: output picture.
Preferably, step S4 examines circuit board picture to be detected using trained YOLO9000 algorithm model
It surveys, comprising the following steps:
S4.1: the template of circuit board to be detected is read；
S4.2: seven pictures after step S3.3 operation are sequentially input to trained YOLO9000 algorithm model；
S4.3: be sequentially output detection as a result, include electronic component model classification, position, confidence level；
S4.4: the testing result of six pictures through being rotated, being overturn in step S3.3 is mapped back original picture, wherein polarized
Electronic component also corresponding change classification；
S4.5: testing result is obtained.
Preferably, testing result is obtained in step S4.5, comprising the following steps:
S4.5.1: setting IOU threshold value is superimposed the testing result of remaining six picture based on original picture testing result respectively,
Every picture is successively detected, is same if the IOU with some band of position in original picture testing result is greater than IOU threshold value
One band of position, increase record the corresponding electronic component model classification in the band of position and confidence level；If examined with original picture
The IOU for surveying some electronic component in result is less than IOU threshold value, then is judged as in new position detection to electronic component type
Number classification, on the basis of the testing result of original picture plus the location information of the band of position, electronic component model classification and
Confidence level；
S4.5.2: setting confidence threshold value after being sequentially overlapped the testing result of 6 pictures, is differentiated according to record information:
If 1) seven pictures detection same position region is all the same electronic component model classification, result is the electronics
Component model classification；
2) if seven pictures detection same position region is k different electronic component model classification, confidence level highest is chosen
Electronic component model classification of the classification as the band of position；
There is corresponding k electronic component model classification if 3) be discontented with seven pictures and detect the same band of position, that is, has
Picture does not detect that there is electronic component in the region, but other pictures have, but testing result is different, and it is highest to choose confidence level
Electronic component model classification of the electronic component model classification as the band of position；
4) if being discontented with 7 pictures detects that there are corresponding 1 electronic component model classification, that is, the figure having in some band of position
Piece does not detect that there is thing in the region, but other pictures have and testing result is the same, with its confidence level compared with confidence threshold value,
As this band of position as a result, by the electronic component model classification band of position when confidence level is higher than confidence threshold value
Electronic component model classification；
5) other situations are then judged as erroneous judgement, judge the position without electronic component.
Preferably, result is exported in S5, comprising the following steps:
S5.1: it will test result and be compared with the template of the circuit board；
S5.2: information is recorded if the electronic component model classification of some position in circuit board and template are inconsistent, and is chosen
It elects manual identified and corrects, return step S4；If the electronic component model classification in circuit board is consistent with template, then
Execute step S5.3；
S5.3: display result information.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
Training YOLO9000 algorithm model, can carry out electronic component model classification identification, to electronic component into
Positioning of the row on picture position reaches the model classification of detection electronic component, position, whether welds mistake, simultaneously as
Direction discernment problem is converted classification problem by the welding direction of the electronic component of polarized on circuit board, realizes to there is pole
The direction discernment of the electronic component of property.
Detailed description of the invention
Fig. 1 is a kind of circuit board element detection method flow chart based on YOLO9000 algorithm.
Fig. 2 is polarized component classification schematic diagram in circuit board.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent；
In order to better illustrate this embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent the ruler of actual product
It is very little；
To those skilled in the art, the omitting of some known structures and their instructions in the attached drawings are understandable.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment 1
The present embodiment provides a kind of circuit board element detection method based on YOLO9000 algorithm, such as Fig. 1, including following step
It is rapid:
S1: building includes the database of circuit board picture, is further divided into four classes to the electronic component of polarized, wherein with circuit
On the basis of 12 o'clock of plate picture direction, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of upper direction；With 3 points of circuit board picture
On the basis of clock direction, 45 degree counterclockwise to 45 degree of ranges clockwise are right direction classification；Using 6 o'clock of circuit board picture direction as base
Standard, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of lower direction；On the basis of 9 o'clock of circuit board picture direction, counterclockwise
45 degree are that left direction is classified to 45 degree of ranges clockwise；
S2: database training YOLO9000 algorithm model is utilized, trained YOLO9000 algorithm model is obtained；
S3: circuit board picture to be detected is obtained；
S4: circuit board picture to be detected is detected using trained YOLO9000 algorithm model；
S4: output result.
Building includes the database of circuit board picture in step S1, comprising the following steps:
S1.1: extracting the information list of circuit board element, four classes are further divided into the electronic component of polarized, wherein with electricity
On the basis of 12 o'clock of the plate picture direction of road, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of upper direction；With circuit board picture 3
On the basis of o'clock direction, 45 degree counterclockwise to 45 degree of ranges clockwise are right direction classification；It is with 6 o'clock of circuit board picture direction
Benchmark, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of lower direction；On the basis of 9 o'clock of circuit board picture direction, the inverse time
45 degree of needle are that left direction is classified to 45 degree of ranges clockwise；
S1.2: it is shot according to component of the information list to corresponding model, generates the first data set；
S1.3: the enhancing of first time data is carried out to the first data set, generates the second data set；
S1.4: second of data enhancing is carried out to the first data set and the second data set, generates third data set；
S1.5: the classification and position of electronic component in the first data set, the second data set and third data set are marked；
S1.6: the data set marked is divided into training set, verifying collection and test set, as database.
The enhancing of first time data carries out data enhancing to utilize generation to fight network.
Second of data enhancing is to be rotated by 90 ° to the first data set and the second data set, rotate 180 degree, rotation
270 degree, flip horizontal, flip vertical and horizontal vertical overturn six kinds of modes and carry out data enhancing.
Classification and position in step S1.5 to electronic component in the first data set, the second data set and third data set
It is marked, four classes, such as Fig. 2 is further divided into the electronic component of polarized, wherein using 12 o'clock of circuit board picture direction as base
Standard, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of upper direction；On the basis of 3 o'clock of circuit board picture direction, counterclockwise
45 degree are that right direction is classified to 45 degree of ranges clockwise；On the basis of 6 o'clock of circuit board picture direction, 45 degree counterclockwise to up time
45 degree of ranges of needle are the classification of lower direction；On the basis of 9 o'clock of circuit board picture direction, 45 degree counterclockwise to 45 degree of ranges clockwise
For left direction classification.
Database training YOLO9000 algorithm model is utilized in step S2, obtains trained YOLO9000 algorithm model,
The following steps are included:
S2.1: training set input YOLO9000 algorithm model is subjected to propagated forward, exports calculated result；
S2.2: loss calculating is carried out according to calculated result, and judges whether the loss average value of nearest 50 iteration is less than and sets
Whether fixed value, judgement reach greatest iteration step number, if the two have one is, deconditioning exports trained
YOLO9000 algorithm model executes step S2.4；If not being, S2.3 is thened follow the steps:
S2.3: the parameter and return step S2.1 of each neural network in YOLO9000 algorithm model are updated according to the value of loss；
S2.4: trained YOLO9000 algorithm model is tested using test set, exports all kinds of accuracy of the mean AP
And average AP value MAP；
S2.5: judging whether all kinds of accuracy of the mean AP and average AP value MAP reach requirement, if reached, exports final
Otherwise parameter is adjusted in YOLO9000 algorithm model, return to S2.1 re -training YOLO9000 algorithm model.
Circuit board picture to be detected is obtained in S3, comprising the following steps:
S3.1: it in the case where the position that focal length, the distance apart from circuit board, circuit board are placed is fixed, is imaged using industry
Head shoots circuit board to be detected；
S3.2: detection circuit board profile extracts the picture of circuit board section, obtains original picture
S3.3: to extract circuit board section picture be rotated by 90 °, rotate 180 degree, rotation 270 degree, flip horizontal, vertically
Overturning, horizontal vertical turning operation obtain seven pictures including original picture；
S3.4: output picture.
Step S4 detects circuit board picture to be detected using trained YOLO9000 algorithm model, including with
Lower step:
S4.1: the template of circuit board to be detected is read；
S4.2: seven pictures after step S3.3 operation are sequentially input to trained YOLO9000 algorithm model；
S4.3: be sequentially output detection as a result, include electronic component model classification, position, confidence level；
S4.4: the testing result of six pictures through being rotated, being overturn in step S3.3 is mapped back original picture, wherein polarized
Electronic component also corresponding change classification；
S4.5: testing result is obtained.
Testing result is obtained in step S4.5, comprising the following steps:
S4.5.1: setting IOU threshold value is superimposed the testing result of remaining six picture based on original picture testing result respectively,
Every picture is successively detected, is same if the IOU with some band of position in original picture testing result is greater than IOU threshold value
One band of position, increase record the corresponding electronic component model classification in the band of position and confidence level；If examined with original picture
The IOU for surveying some electronic component in result is less than IOU threshold value, then is judged as in new position detection to electronic component type
Number classification, on the basis of the testing result of original picture plus the location information of the band of position, electronic component model classification and
Confidence level；
S4.5.2: setting confidence threshold value after being sequentially overlapped the testing result of 6 pictures, is differentiated according to record information:
If 1) seven pictures detection same position region is all the same electronic component model classification, result is the electronics
Component model classification；
If 2) seven pictures detection same position region is k different electronic component model classification, k chooses confidence less than 4
Spend electronic component model classification of the highest classification as the band of position；
There is corresponding k electronic component model classification if 3) be discontented with seven pictures and detect the same band of position, selection is set
Electronic component model classification of the highest electronic component model classification of reliability as the band of position；
4) if being discontented with 7 pictures detects that there is corresponding 1 electronic component model classification in some band of position, with its confidence
Spend the result compared with confidence threshold value, as this band of position；
5) other situations are then judged as erroneous judgement, judge the position without electronic component.
Result is exported in S5, comprising the following steps:
S5.1: it will test result and be compared with the template of the circuit board；
S5.2: information is recorded if the electronic component model classification of some position in circuit board and template are inconsistent, and is chosen
It elects manual identified and corrects, return step S4；If the electronic component model classification in circuit board is consistent with template, then
Execute step S5.3；
S5.3: display result information.
The same or similar label correspond to the same or similar components；
The terms describing the positional relationship in the drawings are only for illustration, should not be understood as the limitation to this patent；
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be to this hair
The restriction of bright embodiment.For those of ordinary skill in the art, it can also do on the basis of the above description
Other various forms of variations or variation out.There is no necessity and possibility to exhaust all the enbodiments.It is all in the present invention
Spirit and principle within made any modifications, equivalent replacements, and improvements etc., should be included in the guarantor of the claims in the present invention
Within the scope of shield.
Claims (10)
1. a kind of circuit board element detection method based on YOLO9000 algorithm, which comprises the following steps:
S1: building includes the database of circuit board picture, is divided into four classes to the electronic component of polarized, wherein with circuit board
On the basis of 12 o'clock of picture direction, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of upper direction；With 3 o'clock of circuit board picture
On the basis of direction, 45 degree counterclockwise to 45 degree of ranges clockwise are right direction classification；Using 6 o'clock of circuit board picture direction as base
Standard, 45 degree counterclockwise to 45 degree of ranges clockwise are the classification of lower direction；On the basis of 9 o'clock of circuit board picture direction, counterclockwise
45 degree are that left direction is classified to 45 degree of ranges clockwise；
S2: database training YOLO9000 algorithm model is utilized, trained YOLO9000 algorithm model is obtained；
S3: circuit board picture to be detected is obtained；
S4: circuit board picture to be detected is detected using trained YOLO9000 algorithm model；
S5: output result.
2. the circuit board element detection method according to claim 1 based on YOLO9000 algorithm, which is characterized in that step
Building includes the database of circuit board picture in rapid S1, comprising the following steps:
S1.1: the information list of circuit board element is extracted；
S1.2: it is shot according to component of the information list to corresponding model, generates the first data set；
S1.3: the enhancing of first time data is carried out to the first data set, generates the second data set；
S1.4: second of data enhancing is carried out to the first data set and the second data set, generates third data set；
S1.5: the classification and position of electronic component in the first data set, the second data set and third data set are marked；
S1.6: the data set marked is divided into training set, verifying collection and test set, as database.
3. the circuit board element detection method according to claim 2 based on YOLO9000 algorithm, which is characterized in that institute
The enhancing of first time data is stated to carry out data enhancing using generation confrontation network.
4. the circuit board element detection method according to claim 3 based on YOLO9000 algorithm, which is characterized in that institute
Second of data enhancing is stated as 180 degree, rotation 270 degree, water are rotated by 90 °, rotated to the first data set and the second data set
Flat overturning, flip vertical and horizontal vertical overturn six kinds of modes and carry out data enhancing.
5. the circuit board element detection method according to claim 4 based on YOLO9000 algorithm, which is characterized in that step
The classification of electronic component in the first data set, the second data set and third data set and position are marked in rapid S1.5,
Four classes are further divided into the electronic component of polarized, wherein on the basis of 12 o'clock of circuit board picture direction, 45 degree counterclockwise extremely
45 degree of ranges clockwise are the classification of upper direction；On the basis of 3 o'clock of circuit board picture direction, 45 degree counterclockwise to 45 degree clockwise
Range is right direction classification；On the basis of 6 o'clock of circuit board picture direction, 45 degree counterclockwise to 45 degree of ranges clockwise are lower section
To classification；On the basis of 9 o'clock of circuit board picture direction, 45 degree counterclockwise to 45 degree of ranges clockwise are left direction classification.
6. the circuit board element detection method according to claim 5 based on YOLO9000 algorithm, which is characterized in that step
Database training YOLO9000 algorithm model is utilized in rapid S2, obtains trained YOLO9000 algorithm model, including following step
It is rapid:
S2.1: training set input YOLO9000 algorithm model is subjected to propagated forward, exports calculated result；
S2.2: loss calculating is carried out according to calculated result, and judges whether the loss average value of nearest 50 iteration is less than and sets
Whether fixed value, judgement reach greatest iteration step number, if the two have one is, deconditioning exports trained
YOLO9000 algorithm model executes step S2.4；If not being, S2.3 is thened follow the steps:
S2.3: the parameter and return step S2.1 of each neural network in YOLO9000 algorithm model are updated according to the value of loss；
S2.4: trained YOLO9000 algorithm model is tested using test set, exports all kinds of accuracy of the mean AP
And average AP value MAP；
S2.5: judging whether all kinds of accuracy of the mean AP and average AP value MAP reach requirement, if reached, exports final
Otherwise parameter is adjusted in YOLO9000 algorithm model, return to S2.1 re -training YOLO9000 algorithm model.
7. the circuit board element detection method according to claim 6 based on YOLO9000 algorithm, which is characterized in that S3
It is middle to obtain circuit board picture to be detected, comprising the following steps:
S3.1: it in the case where the position that focal length, the distance apart from circuit board, circuit board are placed is fixed, is imaged using industry
Head shoots circuit board to be detected；
S3.2: detection circuit board profile extracts the picture of circuit board section, obtains original picture
S3.3: to extract circuit board section picture be rotated by 90 °, rotate 180 degree, rotation 270 degree, flip horizontal, vertically
Overturning, horizontal vertical turning operation obtain seven pictures including original picture；
S3.4: output picture.
8. the circuit board element detection method according to claim 7 based on YOLO9000 algorithm, which is characterized in that step
Rapid S4 detects circuit board picture to be detected using trained YOLO9000 algorithm model, comprising the following steps:
S4.1: the template of circuit board to be detected is read；
S4.2: seven pictures after step S3.3 operation are sequentially input to trained YOLO9000 algorithm model；
S4.3: be sequentially output detection as a result, include electronic component model classification, position, confidence level；
S4.4: the testing result of six pictures through being rotated, being overturn in step S3.3 is mapped back original picture, wherein polarized
Electronic component also corresponding change classification；
S4.5: testing result is obtained.
9. the circuit board element detection method according to claim 8 based on YOLO9000 algorithm, which is characterized in that step
Testing result is obtained in rapid S4.5, comprising the following steps:
S4.5.1: setting IOU threshold value is superimposed the testing result of remaining six picture based on original picture testing result respectively,
Every picture is successively detected, is same if the IOU with some band of position in original picture testing result is greater than IOU threshold value
One band of position, increase record the corresponding electronic component model classification in the band of position and confidence level；If examined with original picture
The IOU for surveying some electronic component in result is less than IOU threshold value, then is judged as in new position detection to electronic component type
Number classification, on the basis of the testing result of original picture plus the location information of the band of position, electronic component model classification and
Confidence level；
S4.5.2: setting confidence threshold value after being sequentially overlapped the testing result of 6 pictures, is differentiated according to record information:
If seven pictures detection same position region is all the same electronic component model classification, result is electronics member
Device type classification；
If it is k different electronic component model classification that seven pictures, which detect same position region, k is greater than 1, chooses confidence level
Electronic component model classification of the highest classification as the band of position；
If discontented seven pictures, which detect the same band of position, corresponding k electronic component model classification, confidence is chosen
Spend electronic component model classification of the highest electronic component model classification as the band of position；
If discontented 7 pictures detect that there is corresponding 1 electronic component model classification in some band of position, with its confidence level
Result compared with confidence threshold value, as this band of position；
Other situations are then judged as erroneous judgement, judge the position without electronic component.
10. the circuit board element detection method according to claim 9 based on YOLO9000 algorithm, which is characterized in that
Result is exported in S5, comprising the following steps:
S5.1: it will test result and be compared with the template of the circuit board；
S5.2: information is recorded if the electronic component model classification of some position in circuit board and template are inconsistent, and is chosen
It elects manual identified and corrects, return step S4；If the electronic component model classification in circuit board is consistent with template, then
Execute step S5.3；
S5.3: display result information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910114795.1A CN109918523B (en) | 2019-02-14 | 2019-02-14 | Circuit board component detection method based on YOLO9000 algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910114795.1A CN109918523B (en) | 2019-02-14 | 2019-02-14 | Circuit board component detection method based on YOLO9000 algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109918523A true CN109918523A (en) | 2019-06-21 |
| CN109918523B CN109918523B (en) | 2021-03-30 |
Family
ID=66961512
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910114795.1A Active CN109918523B (en) | 2019-02-14 | 2019-02-14 | Circuit board component detection method based on YOLO9000 algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109918523B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110808033A (en) * | 2019-09-25 | 2020-02-18 | 武汉科技大学 | Audio classification method based on dual data enhancement strategy |
| CN112287896A (en) * | 2020-11-26 | 2021-01-29 | 山东捷讯通信技术有限公司 | Unmanned aerial vehicle aerial image target detection method and system based on deep learning |
| CN112560836A (en) * | 2020-11-09 | 2021-03-26 | 联想(北京)有限公司 | Component identification method and device and computer readable storage medium |
| CN113077453A (en) * | 2021-04-15 | 2021-07-06 | 华南理工大学 | Circuit board component defect detection method based on deep learning |
| CN113804704A (en) * | 2020-06-11 | 2021-12-17 | 广东美的白色家电技术创新中心有限公司 | Circuit board detection method, visual detection equipment and device with storage function |
| DE102021206323A1 (en) | 2021-06-21 | 2022-12-22 | Robert Bosch Gesellschaft mit beschränkter Haftung | Method of analyzing an electrical circuit |
| CN118115488A (en) * | 2024-04-02 | 2024-05-31 | 辰穗(广州)网络科技有限公司 | Computed imaging method, device, computer equipment and medium for circuit board detection |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106940887A (en) * | 2017-03-09 | 2017-07-11 | 中国科学院遥感与数字地球研究所 | A kind of satellite sequence image clouds of GF 4 and shadow detection method under cloud |
| CN107368847A (en) * | 2017-06-26 | 2017-11-21 | 北京农业信息技术研究中心 | A kind of crop leaf diseases recognition methods and system |
| CN108921123A (en) * | 2018-07-17 | 2018-11-30 | 重庆科技学院 | A kind of face identification method based on double data enhancing |
| CN109063594A (en) * | 2018-07-13 | 2018-12-21 | 吉林大学 | Remote sensing images fast target detection method based on YOLOv2 |
| CN109255769A (en) * | 2018-10-25 | 2019-01-22 | 厦门美图之家科技有限公司 | The training method and training pattern and image enchancing method of image enhancement network |
-
2019
- 2019-02-14 CN CN201910114795.1A patent/CN109918523B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106940887A (en) * | 2017-03-09 | 2017-07-11 | 中国科学院遥感与数字地球研究所 | A kind of satellite sequence image clouds of GF 4 and shadow detection method under cloud |
| CN107368847A (en) * | 2017-06-26 | 2017-11-21 | 北京农业信息技术研究中心 | A kind of crop leaf diseases recognition methods and system |
| CN109063594A (en) * | 2018-07-13 | 2018-12-21 | 吉林大学 | Remote sensing images fast target detection method based on YOLOv2 |
| CN108921123A (en) * | 2018-07-17 | 2018-11-30 | 重庆科技学院 | A kind of face identification method based on double data enhancing |
| CN109255769A (en) * | 2018-10-25 | 2019-01-22 | 厦门美图之家科技有限公司 | The training method and training pattern and image enchancing method of image enhancement network |
Non-Patent Citations (2)
| Title |
|---|
| 宿鸣明: ""电路板元器件的检测与识别"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 高宗: ""基于机器视觉的电子元器件质量检测技术研究"", 《万方》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110808033A (en) * | 2019-09-25 | 2020-02-18 | 武汉科技大学 | Audio classification method based on dual data enhancement strategy |
| CN110808033B (en) * | 2019-09-25 | 2022-04-15 | 武汉科技大学 | Audio classification method based on dual data enhancement strategy |
| CN113804704A (en) * | 2020-06-11 | 2021-12-17 | 广东美的白色家电技术创新中心有限公司 | Circuit board detection method, visual detection equipment and device with storage function |
| CN112560836A (en) * | 2020-11-09 | 2021-03-26 | 联想(北京)有限公司 | Component identification method and device and computer readable storage medium |
| CN112287896A (en) * | 2020-11-26 | 2021-01-29 | 山东捷讯通信技术有限公司 | Unmanned aerial vehicle aerial image target detection method and system based on deep learning |
| CN113077453A (en) * | 2021-04-15 | 2021-07-06 | 华南理工大学 | Circuit board component defect detection method based on deep learning |
| CN113077453B (en) * | 2021-04-15 | 2022-09-13 | 华南理工大学 | Circuit board component defect detection method based on deep learning |
| DE102021206323A1 (en) | 2021-06-21 | 2022-12-22 | Robert Bosch Gesellschaft mit beschränkter Haftung | Method of analyzing an electrical circuit |
| CN118115488A (en) * | 2024-04-02 | 2024-05-31 | 辰穗(广州)网络科技有限公司 | Computed imaging method, device, computer equipment and medium for circuit board detection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109918523B (en) | 2021-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109918523A (en) | A kind of circuit board element detection method based on YOLO9000 algorithm | |
| CN108764257A (en) | A kind of pointer instrument recognition methods of various visual angles | |
| CN106874894A (en) | A kind of human body target detection method based on the full convolutional neural networks in region | |
| CN103718175B (en) | Detect equipment, method and the medium of subject poses | |
| CN103530590B (en) | DPM Quick Response Code recognition system | |
| CN109447076A (en) | A kind of vehicle VIN code recognition detection method for vehicle annual test | |
| CN107103298A (en) | Chin-up number system and method for counting based on image procossing | |
| CN109190446A (en) | Pedestrian's recognition methods again based on triple focused lost function | |
| CN105740780A (en) | Method and device for human face in-vivo detection | |
| CN108305243A (en) | A kind of magnetic tile surface defect detection method based on deep learning | |
| CN103528617A (en) | Cockpit instrument automatic recognizing and detecting method and device | |
| CN107977656A (en) | A kind of pedestrian recognition methods and system again | |
| CN102657532A (en) | Height measuring method and device based on body posture identification | |
| CN103154972A (en) | Text-based 3D augmented reality | |
| CN113597614B (en) | Image processing method and device, electronic equipment and storage medium | |
| TW201317904A (en) | Tag detecting system, apparatus and method for detecting tag thereof | |
| CN110136130A (en) | A kind of method and device of testing product defect | |
| CN103455826B (en) | Efficient matching kernel body detection method based on rapid robustness characteristics | |
| CN109740654A (en) | A kind of tongue body automatic testing method based on deep learning | |
| CN108830179A (en) | Merge the pedestrian detection algorithm of Color Image Edge and depth direction histogram | |
| CN110335242A (en) | A kind of tire X-ray defect detection method based on multi-model fusion | |
| CN110321869A (en) | Personnel's detection and extracting method based on Multiscale Fusion network | |
| CN109492573A (en) | A kind of pointer read method and device | |
| CN103927517A (en) | Motion detection method based on human body global feature histogram entropies | |
| CN106446837B (en) | A kind of detection method of waving based on motion history image |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Chen Wenshuai Inventor after: Ren Zhigang Inventor after: Wu Zongze Inventor after: Chen Xiaocong Inventor before: Chen Wenshuai Inventor before: Ren Zhigang Inventor before: Wu Zongze Inventor before: Fu Min Yue Inventor before: Chen Xiaocong |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |