CN114782437B - Computer mainboard quality detection method and system based on artificial intelligence - Google Patents
Computer mainboard quality detection method and system based on artificial intelligence Download PDFInfo
- Publication number
- CN114782437B CN114782437B CN202210698021.XA CN202210698021A CN114782437B CN 114782437 B CN114782437 B CN 114782437B CN 202210698021 A CN202210698021 A CN 202210698021A CN 114782437 B CN114782437 B CN 114782437B
- Authority
- CN
- China
- Prior art keywords
- feature map
- mainboard
- loss
- image
- sampling
- 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.)
- Expired - Fee Related
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 145
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 15
- 238000005070 sampling Methods 0.000 claims abstract description 76
- 238000000034 method Methods 0.000 claims abstract description 25
- 238000010586 diagram Methods 0.000 claims abstract description 18
- 238000012545 processing Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 12
- 238000000605 extraction Methods 0.000 claims description 26
- 238000011176 pooling Methods 0.000 claims description 15
- 238000007689 inspection Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 230000007246 mechanism Effects 0.000 claims description 9
- 230000000087 stabilizing effect Effects 0.000 claims description 7
- 206010037180 Psychiatric symptoms Diseases 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 description 40
- 230000004913 activation Effects 0.000 description 13
- 238000010606 normalization Methods 0.000 description 10
- 239000011159 matrix material Substances 0.000 description 6
- 238000002372 labelling Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000005286 illumination Methods 0.000 description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 230000002349 favourable effect Effects 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 239000003990 capacitor Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000004313 glare Effects 0.000 description 1
- 229910052742 iron Inorganic materials 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
A computer mainboard quality detection method and system based on artificial intelligence are disclosed, the method includes the following steps: collecting an image of a mainboard to be detected; constructing a mainboard quality detection model, and training the mainboard quality detection model; processing the image of the mainboard to be detected by using the trained mainboard quality detection model, and judging whether the fan of the mainboard to be detected corresponding to the image is correctly connected and whether the type of the mainboard fixing screw is correct; the main board quality detection model is used for carrying out four times of down sampling and three times of up sampling on the image of the main board to be detected, fusing a feature diagram obtained by the third time of up sampling and a feature diagram obtained by the first time of down sampling, and obtaining a first feature diagram with the detection scale of 160 multiplied by 160. According to the invention, on the basis of the traditional Yolov5 model, the first characteristic diagram with the detection scale of 160 multiplied by 160 is added, so that small targets such as fixed screws can be better detected, and the accuracy of the mainboard quality detection model is effectively improved.
Description
Technical Field
The invention relates to the field of computer mainboard quality detection, in particular to a computer mainboard quality detection method and system based on artificial intelligence, which are used for realizing automatic detection on the quality of a computer mainboard in the computer production process.
Background
The traditional mainboard detection assembly line needs to go through a series of complex and high error rate processes such as manual sampling, diagnosis card detection, manual report filling and the like, wastes time and labor, is difficult to summarize and analyze historical detection results, and cannot improve the error rate of each production link in a targeted manner. In recent years, the research of an object detection algorithm based on a convolutional neural network is rapidly developed, and an effective detection means is provided for improving the detection efficiency and accuracy of the quality of a computer mainboard.
The two algorithms of the R-CNN series algorithm and the YOLO series algorithm are the mainstream algorithms for object detection at present. Compared with the R-CNN series algorithm, the YOLO series algorithm has the advantages of high speed, more general image characteristics and prediction based on the information of the whole picture.
The main defects of the computer mainboard detection assembly line detection include whether the fan is correctly connected and whether the type of the mainboard fixing screw is correct. Except for the fan and the screw, the computer mainboard is also provided with a plurality of components and parts such as slots, and the three-detection-scale detection layer network adopted by the traditional Yolov5 model easily loses smaller target area information in the model training process, so that the false detection rate and the missing detection rate are increased.
Disclosure of Invention
The invention aims to provide a computer mainboard quality detection method based on artificial intelligence, which can effectively improve the connection and fusion of low-level semantics and deep-level semantics, improve the small detection capability of a model, strengthen the detection capability of computer mainboards with various shapes, various types and numerous small components by introducing a detection scale of 160 multiplied by 160 on a detection layer network and forming a four-detection-scale detection layer network together with 20 multiplied by 20, 40 multiplied by 40 and 80 multiplied by 80, can quickly, efficiently and accurately detect whether a mainboard fan is correctly connected and the type of a mainboard fixing screw is correct, and greatly reduce the false detection and the omission ratio.
The invention is realized by the following technical scheme:
a computer mainboard quality detection method based on artificial intelligence comprises the following steps:
collecting an image of a mainboard to be detected;
constructing a mainboard quality detection model, and training the mainboard quality detection model;
processing the image of the mainboard to be detected by using the trained mainboard quality detection model, and judging whether the fan of the mainboard to be detected corresponding to the image of the mainboard to be detected is correctly connected and whether the type of the mainboard fixing screw is correct;
the main board quality detection model is used for carrying out four times of down sampling and three times of up sampling on an image of a main board to be detected, fusing a feature map obtained by the third time of up sampling and a feature map obtained by the first time of down sampling, and obtaining a first feature map with the detection scale of 160 multiplied by 160.
Yolov5 is a commonly used detection algorithm for object detection at present, and detection layer networks with three detection scales of 20 × 20, 40 × 40 and 80 × 80 are generally adopted. However, the computer motherboard not only includes components with larger volume such as the fan and the hard disk slot, but also includes components with smaller volume such as the fixing screw, the button cell, the resistor, the capacitor and the inductance element. If the existing Yolov5 model is directly adopted to detect the quality of the mainboard, target area information with small loss is easy to occur, and the false detection and the missing detection rate are improved.
In order to solve the above problems, in the technical solution, a mainboard quality detection model is constructed based on a traditional Yolov5 model, and the mainboard quality detection model performs four times of down-sampling and three times of up-sampling on an input image of a mainboard to be detected. Performing convolution layer processing, batch normalization processing, activation of an activation function and other processing after down sampling to extract features; after each upsampling, the feature map obtained by a certain downsampling is fused, and features are extracted from the fused feature map.
In the technical scheme, in order to obtain a detection layer with a smaller detectable volume, the feature map obtained by the third upsampling, namely the last upsampling, is fused with the feature map obtained by the first downsampling, so that the first feature map with the detection scale of 160 × 160 can be obtained, and thus the first feature map and the traditional detection scales of 20 × 20, 40 × 40 and 80 × 80 form a four-detection-scale detection layer network together, so that small targets such as fixed screws can be better detected, and the accuracy of the mainboard quality detection model is effectively improved.
And after the mainboard quality detection model is established, training the mainboard quality detection model. In a partially preferred embodiment, 2 sample images can be mixed at random in proportion, a partial area of one image is cut out, the other image is cut out and filled in the cut area, so that a new image and a corresponding label thereof are generated, data labeling is performed by using a website makesense.
After the mainboard quality detection model is trained, the collected images corresponding to the mainboards to be detected can be processed, and whether the fan of the mainboard corresponding to the images of the mainboards to be detected is correctly connected and the mainboard fixing screw model is correct is judged according to the feature extraction recognition result.
Further, in some embodiments, the number of the mainboard with the correct connection and the incorrect connection of the fan and the number of the mainboard with the correct and the incorrect type of the fixing screws of the mainboard are counted to generate a statistical report. According to the judgment result of the mainboard quality detection model, the number of the mainboard with wrong fan connection and the number of the mainboard with wrong fixing screw model in each detection batch can be counted, the correct connection rate of the mainboard fan, the wrong connection rate of the fan, the correct rate of the fixing screw model of the mainboard and the wrong fixing screw model of the mainboard in each detection batch are counted based on the number, and a statistical report is generated so as to be convenient for a client to check.
In a preferred embodiment of the present invention, the main board quality detection model is configured to perform down-sampling on an image of a main board to be detected four times to obtain a first down-sampling feature map, a second down-sampling feature map, a third down-sampling feature map, and a fourth down-sampling feature map, where the fourth down-sampling feature map is subjected to up-sampling for the first time and then fused with the third down-sampling feature map to obtain a first fused feature map, the first fused feature map is subjected to feature extraction and up-sampling for the second time and then fused with the second down-sampling feature map to obtain a second fused feature map, the second fused feature map is subjected to feature extraction and up-sampling for the third time and then fused with the first down-sampling feature map to obtain a third fused feature map, and the third fused feature map is subjected to feature extraction to obtain the first feature map.
In the technical scheme, the first to fourth downsampling feature maps are obtained by four downsampling processes, wherein after the first three downsampling processes, feature extraction is carried out through convolutional layer processing, batch normalization processing and activation function activation. In the up-sampling process, the fourth down-sampling feature map is fused with the third down-sampling feature map after the first up-sampling, the second up-sampling is carried out after the feature extraction, the feature map obtained by the second up-sampling is fused with the feature map obtained by the second down-sampling to obtain a second fused feature map, the second fused feature map is subjected to the feature extraction, the third up-sampling is carried out, the third fused feature map is finally fused with the first down-sampling feature map to obtain a third fused feature map, and the third fused feature map is subjected to the feature extraction to obtain the first feature map of 160 multiplied by 255.
Further, the mainboard quality detection model is configured to fuse the second fused feature map and the third fused feature map to obtain a fourth fused feature map, where the fourth fused feature map is subjected to feature extraction to obtain a second feature map with a detection scale of 80 × 80, and is configured to fuse the fourth fused feature map, the third downsampled feature map, and the first fused feature map to obtain a fifth fused feature map, where the fifth fused feature map is subjected to feature extraction to obtain a third feature map with a detection scale of 40 × 40, and is configured to fuse the fifth fused feature map and the fourth downsampled feature map to obtain a sixth fused feature map, and the sixth fused feature map is subjected to feature extraction to obtain a fourth feature map with a detection scale of 20 × 20.
Further, target prediction is carried out on the first feature map, the second feature map, the third feature map and the fourth feature map, and the anchor point configuration of the target prediction comprises anchor points [5,6,8,14,15,11 ]. The anchor points [5,6,8,14,15,11] of the traditional Yolov5 model used for target prediction are usually 9, in the technical scheme, anchor points [5,6,8,14,15,11] are added on the basis of the traditional 9 anchor points to form 12 different anchor points for target prediction, so that the detection capability of the model on small-size objects can be further improved, and the model is favorable for identifying fan connecting lines and fixing screw models.
As a preferred embodiment of the present invention, a loss function is used to calculate a target prediction loss of the motherboard quality inspection model, and the formula of the loss function is:
wherein,loss rect in order to predict the total loss of the box,loss obj in order to account for the total loss of confidence,loss clc to classify the total loss;loss rect160 、loss obj160 、loss clc160 respectively representing the prediction frame loss, confidence coefficient loss and classification loss of the first feature map;loss rect80 、loss obj80 、loss clc80 respectively representing the prediction frame loss, confidence coefficient loss and classification loss of the second feature map;loss rect40 、loss obj40 、loss clc40 the prediction frame loss, the confidence coefficient loss and the classification loss of the third feature map are respectively;loss rect20 、loss obj20 、loss clc20 the predicted frame loss, the confidence coefficient loss and the classification loss of the fourth feature map are respectively;a 1 、a 2 、a 3 、a 4 the weights of the first to fourth characteristic diagrams in turn, wherein,a 1 >a 2 =a 3 >a 4 and is anda 1 <a 2 +a 3 。
in the technical scheme, for a 160 × 160 grid, the loss function calculation expressions of each type are as follows:
wherein,maskthe method comprises the steps that a mask matrix is adopted, 160 × 160 bool values (Boolean variables) correspond to 160 × 160 prediction frames one by one, whether a target exists in each prediction frame is judged according to label information and a certain rule, if yes, the value of the corresponding position in the mask matrix is set to true, otherwise, the value is set to false, and only rectangular frame loss and classification loss of the prediction frame with the mask matrix as true are calculated;
is a BCE (binary cross entropy) loss function, is used for solving the loss of a single class,
not only is the overlapping area considered for CIoU loss, but also the center point distance and aspect ratio are included,l noobj in order to be a confidence loss for a label as background,l obj in order to be a loss of confidence in the labeling of the object,aindicating the weight of the calculated classification loss.
In the present technical solution, in order to increase the attention degree to the loss function in the screw detection process, i.e. the attention degree to the small target, the weight of the loss function of the 160 × 160 detection layer in the overall loss function needs to be the highest, and the weight of the loss function of the 20 × 20 detection layer is the lowest. Meanwhile, in the process of detecting the main board, not only various errors of screws but also errors of fans, wiring and the like of the main board are required to be detected, so that in order to consider both components with larger sizes such as fans and components with smaller sizes such as screws, wiring and the like, the weights of the loss functions of the 80 × 80 detection layer and the 40 × 40 detection layer are required to be kept the same, and the sum of the weights of the loss functions and the weight of the loss functions is required to be greater than that of the 160 × 160 detection layer.
In one or more embodiments, after the predicted loss of the model is calculated by the loss function, the gradient descent adjustment parameters are performed by the optimizer. 150 iteration rounds are trained, and every 15 iteration rounds are validated by the validation set and the optimal model is saved. And finally, obtaining an optimal model after the training is finished.
Further, the mainboard quality detection model is used for slicing the image of the mainboard to be detected and then performing feature extraction to obtain first to fourth feature maps. According to the technical scheme, the image of the mainboard to be detected is input to a Focus module for slicing, four complementary pictures are obtained, no information loss is ensured, then the sliced pictures are subjected to convolution operation on the characteristic diagram through a series of CSP1_ X and/or CSP2_ X, and the down-sampling operation is realized by utilizing spatial pyramid pooling. Specifically, in CSP1_ X, the input is split into two branches, one branch is convolved once again through convolution operation, batch normalization and FRelu (funnel activation function), and then through multiple residual structures; the other branch is directly convoluted; and then fusing the two branches, connecting the two characteristic graphs, carrying out batch normalization, carrying out a funnel activation function again, and finally carrying out convolution, batch normalization and funnel activation function operation once. In CSP2_ X, the input is split into two branches, one branch is convolved once again by convolution operation, batch normalization and FRelu (funnel activation function), and then by 2 × X, i.e. even number of convolutions, batch normalization and funnel activation function operations; the other branch is directly convoluted; and then the two branches are fused to connect the two characteristic graphs, batch normalization is carried out, then a funnel activation function is carried out once, and finally convolution, batch normalization and funnel activation function operation are carried out once.
Further, the mainboard quality detection model comprises a classification correction module, the classification correction module is used for weighting the output channel prediction constant weight, and the formula of the classification correction module is as follows:
wherein, the
Is the weighted sum of the pixel points and is,Hthe height of the pooling window is the height of the pooling window,Win order to be the width of the pooling window,F sq (pixel i ) Indicating that a global weighted average pooling operation is performed, and s is an output value of the pooling window.
In the technical scheme, a constant weight is predicted and weighted for each output channel to highlight a characteristic diagram with high result gain, so that the classification effect of the model is further improved.
As a preferred embodiment of the present invention, a light source for collecting an image of a motherboard to be detected is located directly above a fan of the motherboard to be detected, after a motherboard quality detection model determines whether the fan of the motherboard to be detected is correctly connected according to a collected first image, if it is determined that a motherboard fixing screw model in the first image is incorrect, the motherboard to be detected is rotated by 180 ° and a second image is collected, it is determined whether a motherboard fixing screw model in the second image is correct, if not, the motherboard fixing screw model is determined to be incorrect, and if so, the motherboard fixing screw model is determined to be correct.
Among this technical scheme, aim at the fan that waits to detect the mainboard with the light source of shooing with the position that shows fan and connecting wire clearly as far as possible in the image of shooing to avoid thin and small connecting wire to shelter from because of components and parts effectively, perhaps the regional misidentification of shadow is the connecting wire fracture in the model identification in-process mistake.
Although the light source is opposite to the fan, the illumination intensity of the fan and the connecting line area can be greatly improved, and the accuracy of identifying whether the fan is connected correctly is improved. However, the fan is often not located at the center of the motherboard, so the fixing screws far away from the fan may be shielded by the shadow of other components on the motherboard. Therefore, in the technical scheme, at most two images can be collected for the same main board to be detected. The first image is an image shot by the fan with the light source facing the fan, and if the screw type in the first image is correct, only the first image is shot for the mainboard; however, if the screw model in the first image is wrong, the problem that the model is not easy to identify the fixed screw or the fixed screw is identified incorrectly may be caused by shadow occlusion. At this moment, the main board to be detected can be rotated by 180 degrees and then a second image is acquired, in the second image, the fixing screw area originally far away from the light source is located below the light source, and therefore false identification caused by shadow influence is remarkably reduced. If the type of the mainboard fixing screw in the second image is incorrect, the type of the mainboard fixing screw is judged to be incorrect, otherwise, the type of the mainboard fixing screw is judged to be correct.
The principle in the technical scheme is as follows: compared with a fixed screw, the fan connecting line with smaller size and more complex arrangement is easier to be misjudged by the model and is also easier to be shielded by the fan with large size, so that the problem of misjudgment of the connecting line can be well solved by facing the light source to the fan and the connecting line thereof in the technical scheme. The size of the fixing screw is larger, no connecting line is strong for the requirement of light intensity, and the fixing screw is not easy to be shielded and misjudged relative to the connecting line, so that the fixing screw can be positioned in an area far away from the light source in the first image; if the fixed screw identifies that the model is wrong, the fixed screw can be rotated to a position below the light source, the light intensity is increased, the shadow is reduced, the second image is shot to improve the accuracy of model identification, and the judgment result of the model of the fixed screw of the second image is used as the final judgment result.
Another objective of the present invention is to provide a computer motherboard quality detection system based on artificial intelligence, which includes:
the image acquisition unit is used for acquiring an image of the mainboard to be detected;
the analysis unit is used for constructing and training a mainboard quality detection model, processing the image of the mainboard to be detected by using the trained mainboard quality detection model, and judging whether the fan of the mainboard to be detected corresponding to the image of the mainboard to be detected is correctly connected and whether the model of a mainboard fixing screw is correct, wherein the mainboard quality detection model is used for carrying out down sampling and up sampling for four times on the image of the mainboard to be detected, and fusing a feature map obtained by the up sampling for the third time and a feature map obtained by the down sampling for the first time to obtain a first feature map with the detection scale of 160 x 160;
and the output unit is used for generating a statistical form according to the judging structure, wherein the statistical form comprises the number of the mainboard with correct and wrong fan connection and the number of the mainboard with correct and wrong mainboard fixing screw type.
In one or more embodiments, the image acquisition unit comprises a housing, the housing is a darkroom, the conveyor belt carries a containing plate to pass through the darkroom, a main board of the computer to be detected is placed in the center of the containing plate, a light source and an industrial camera are connected to the center of the inner side of the top of the closed darkroom, and when the main board to be detected is conveyed to the position under the industrial camera by the conveyor belt, the industrial camera takes a picture and transmits picture data to a client computer through a data line. The client computer uploads the picture data to the server through the picture uploading service on the server. And judging the error type of the mainboard by using the trained mainboard quality detection model through an algorithm analysis service on the server, wherein the target detection result comprises the position of a target area in the image to be classified and the type corresponding to each target area, and the types of the target areas are correct fan connection, incorrect fan connection, correct mainboard fixing screw type and incorrect mainboard fixing screw type and are recorded in a database. And reading the error type of the main board in the database through a statistical form generating service on the server, and displaying the statistical form of the fault type and the error times by using a visualization tool Echarts in software.
In one or more embodiments, the light source may also be disposed off-center to face the area of the motherboard to be inspected where the fan is located.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. on the basis of the traditional Yolov5 model, the invention adds the first characteristic diagram with the detection scale of 160 multiplied by 160, and forms a four-detection-scale detection layer network together with the traditional three detection scales of 20 multiplied by 20, 40 multiplied by 40 and 80 multiplied by 80 so as to better detect small targets such as fixed screws and effectively improve the accuracy of the mainboard quality detection model;
2. in order to increase the attention degree of the loss function in the screw detection process, the weight of the loss function of the 160 × 160 detection layer in the overall loss function is the highest, and meanwhile, in order to take account of elements with larger sizes such as fans and elements with smaller sizes such as screws and connecting lines, the weights of the loss functions of the 80 × 80 detection layer and the 40 × 40 detection layer need to be kept the same, and the sum of the weights of the loss functions of the 80 × 80 detection layer and the 40 × 40 detection layer is larger than the weight of the 160 × 160 detection layer;
3. the invention adds anchor points [5,6,8,14,15,11] on the basis of the traditional 9 anchor points to form 12 different anchor points for target prediction, can further improve the detection capability of the model on small-size objects, and is beneficial to identifying fan connecting lines and fixing screw models;
4. the invention can collect at most two images aiming at the same main board to be detected, the light source in the first image is over against the fan and the area where the connecting line of the fan is located, the second image is the image collected after the first image is turned over for 180 degrees, the fixed screw area can be rotated to the position below the light source, the light intensity is increased, the shadow is reduced, the accuracy of the model for identifying the fixed screw is improved, and the judgment result of the fixed screw model of the second image is taken as the final judgment result.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention. In the drawings:
FIG. 1 is a block flow diagram of a detection method in an embodiment of the invention;
FIG. 2 is a block diagram of a flow chart of a method for constructing a motherboard quality inspection model according to an embodiment of the present invention;
FIG. 3 is a block diagram of a detection system in an embodiment of the invention;
FIG. 4 is a schematic structural diagram of an image capturing unit according to an embodiment of the present invention;
FIG. 5 is a schematic view, partially in section, of a detection system in accordance with an embodiment of the present invention;
FIG. 6 is a schematic view of a containment plate according to an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of a holding plate and a position limiting member according to an embodiment of the present invention;
FIG. 8 is a schematic cross-sectional view of a telescoping mechanism in an embodiment of the invention;
fig. 9 is a schematic view of an inner surface of a position-limiting member according to an embodiment of the invention.
Reference numbers and corresponding part names in the drawings:
1-shell, 2-conveyor belt, 3-containing plate, 31-roller, 32-sleeve, 33-telescopic rod, 4-main board to be detected, 41-fan, 5-light source, 6-camera, 7-data line, 8-client, 9-server, 10-driving device, 11-limiting piece, 111-sensor, 112-arc groove and 113-diffuse reflection layer.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to examples and accompanying drawings, and the exemplary embodiments and descriptions thereof are only used for explaining the present invention and are not meant to limit the present invention.
In the description of the present invention, it is to be understood that the terms "front", "rear", "left", "right", "upper", "lower", "vertical", "horizontal", "high", "low", "inner", "outer", etc. indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed in a particular orientation, and be operated, and therefore, should not be taken as limiting the scope of the invention.
Example 1:
as shown in fig. 1, a method for detecting the quality of a computer motherboard based on artificial intelligence comprises the following steps:
collecting an image of a mainboard to be detected;
constructing a mainboard quality detection model, and training the mainboard quality detection model;
processing the image of the mainboard to be detected by using the trained mainboard quality detection model, and judging whether the fan of the mainboard to be detected corresponding to the image of the mainboard to be detected is correctly connected and whether the type of the mainboard fixing screw is correct;
the main board quality detection model is used for carrying out four times of down sampling and three times of up sampling on an image of a main board to be detected, fusing a feature map obtained by the third time of up sampling and a feature map obtained by the first time of down sampling, and obtaining a first feature map with the detection scale of 160 multiplied by 160.
In this embodiment, after the mainboard quality detection model is constructed, the mainboard quality detection model is trained. After the mainboard quality detection model is trained, the collected images corresponding to the mainboards to be detected can be processed, and whether the fan of the mainboard corresponding to the images of the mainboards to be detected is correctly connected and the mainboard fixing screw model is correct is judged according to the feature extraction recognition result.
In some embodiments, the number of the mainboard with correct connection and wrong connection of the fan and the number of the mainboard with correct and wrong type of the mainboard fixing screws are counted to generate a statistical report. In one or more embodiments, software in the client computer is used for reading the error type of the main board in the database, and visualization tools such as Echarts are used for displaying the statistical report of the fault type and the error times.
In order to obtain the first feature map for detecting the small size, in a partially preferred embodiment, the main board quality detection model is configured to perform down-sampling on an image of a main board to be detected four times to obtain a first down-sampling feature map, a second down-sampling feature map, a third down-sampling feature map and a fourth down-sampling feature map, where the fourth down-sampling feature map is subjected to up-sampling for the first time and then fused with the third down-sampling feature map to obtain a first fused feature map, the first fused feature map is subjected to feature extraction and up-sampling for the second time and then fused with the second down-sampling feature map to obtain a second fused feature map, the second fused feature map is subjected to feature extraction and up-sampling for the third time and then fused with the first down-sampling feature map to obtain a third fused feature map, and the third fused feature map is subjected to feature extraction to obtain the first feature map.
In one or more embodiments, after the convolutional layer processing and batch normalization processing, the learu function is adopted to replace the Leaky ReLU activation function, the Relu function can be extended to the visual parameter activation function with the pixel-level modeling capability, and a fine spatial layout of an object is extracted.
In some embodiments, the mainboard quality detection model is configured to fuse the second fused feature map and the third fused feature map to obtain a fourth fused feature map, where the fourth fused feature map is subjected to feature extraction to obtain a second feature map with a detection scale of 80 × 80, and is configured to fuse the fourth fused feature map, the third downsampled feature map, and the first fused feature map to obtain a fifth fused feature map, where the fifth fused feature map is subjected to feature extraction to obtain a third feature map with a detection scale of 40 × 40, and is configured to fuse the fifth fused feature map and the fourth downsampled feature map to obtain a sixth fused feature map, and where the sixth fused feature map is subjected to feature extraction to obtain a fourth feature map with a detection scale of 20 × 20.
In one or more embodiments, the motherboard quality inspection model includes a classification and rectification module, where the classification and rectification (SENet) module is configured to weight and weight a prediction constant of an output channel, and a formula of the classification and rectification module is:
wherein, the
Is the weighted sum of the pixels, and the weighted sum is,Hin order to be the height of the pooling window,Wthe width of the pooling window is the width of the pooling window,F sq (pixel i ) Indicating that a global weighted average pooling operation is performed, and s is an output value of the pooling window.
Example 2:
on the basis of embodiment 1, as shown in fig. 2, the method for constructing the motherboard quality inspection model includes the following steps:
slicing an image of a mainboard to be detected to obtain four complementary pictures;
carrying out a series of convolution operations on the sliced picture, and carrying out downsampling operation;
fusing the feature maps to respectively obtain a first feature map, a second feature map, a third feature map and a fourth feature map to form a four-detection-scale detection layer network;
adopting an anchor point to carry out target prediction and adjusting a prediction frame;
calculating the prediction loss of the mainboard quality detection model by using a loss function, adjusting parameters by gradient descent through an optimizer, training 150 iteration rounds, performing verification set verification every 15 iteration rounds, and storing an optimal model;
and obtaining an optimal model after the training is finished.
In some embodiments, when the mainboard quality detection model is trained, 2 sample images are mixed in proportion at random, a partial area of one image is cut out, the other image is cut out and filled in the cut area, so that a new image and a corresponding label are generated, data labeling is performed by using website makesense.
In one or more embodiments, a coordinate file and a sample image file are imported to train the model, and the trained optimal weight file is loaded to a Yolov5 model with a four-detection-scale detection layer network, so that the trained mainboard quality detection model is obtained. In one or more embodiments, the motherboard quality inspection model is trained using an optimizer whose formula includes:
momentum update formula:
v(t)= γv(t-1)+ g t
learning ratelrIs updated by
Updating parameterswThe formula is as follows:
w ← w - v(t) * lr
wherein, gamma is a momentum preservation value,mit is the learning rate increase amount that is,nis the rate of decay of the learning rate,g t is the gradient value calculated by the parameter at the current moment,g t-1 is the gradient value calculated from the parameter at the previous time, v (t) is the velocity at that time, and v (t-1) is the velocity at the previous time.
In one or more embodiments, after the target prediction is performed on the feature map, redundant prediction boxes are removed using a non-maximum suppression algorithm. Since the non-maximum suppression algorithm suppresses all prediction frames having an IOU value greater than a given threshold from the preselected prediction frame in each iteration, missed detection and false detection of the target are easily caused. In some preferred embodiments, the IOU calculation method is changed to be one with the larger ratio of the intersection area to the area of the two prediction boxes, the local position in one category is mainly filtered and repeatedly detected, and the threshold is properly increased to avoid the adjacent prediction boxes from being filtered; and filters out results with low confidence.
Example 3
On the basis of the above embodiment, target prediction is performed on the first feature map, the second feature map, the third feature map and the fourth feature map, and the anchor point configuration of the target prediction includes anchor points [5,6,8,14,15,11 ].
The anchor points [5,6,8,14,15,11] of the traditional Yolov5 model used for target prediction are usually 9, in the technical scheme, anchor points [5,6,8,14,15,11] are added on the basis of the traditional 9 anchor points to form 12 different anchor points for target prediction, so that the detection capability of the model on small-size objects can be further improved, and the model is favorable for identifying fan connecting lines and fixing screw models.
Example 4
On the basis of the above embodiment, a loss function is used to calculate the target prediction loss of the motherboard quality inspection model, and the formula of the loss function is as follows:
wherein,loss rect in order to predict the total loss of the box,loss obj for the purpose of the overall loss of confidence,loss clc to classify the total loss;loss rect160 、loss obj160 、loss clc160 respectively representing the prediction frame loss, confidence coefficient loss and classification loss of the first feature map;loss rect80 、loss obj80 、loss clc80 respectively representing the prediction frame loss, confidence coefficient loss and classification loss of the second feature map;loss rect40 、loss obj40 、loss clc40 the prediction frame loss, the confidence coefficient loss and the classification loss of the third feature map are respectively;loss rect20 、loss obj20 、loss clc20 the predicted frame loss, the confidence coefficient loss and the classification loss of the fourth feature map are respectively obtained;a 1 、a 2 、a 3 、a 4 the weights of the first to fourth characteristic diagrams are sequentially set, wherein,a 1 >a 2 =a 3 >a 4 and is made ofa 1 <a 2 +a 3 。
In this embodiment, for a 160 × 160 grid, the loss function calculation expressions of each type are as follows:
wherein,maskthe method comprises the steps of representing a mask matrix, enabling 160 multiplied by 160 boost values (Boolean variables) to correspond to 160 multiplied by 160 prediction frames one by one, judging whether a target exists in each prediction frame according to label information and a certain rule, setting the value of the corresponding position in the mask matrix as true if the target exists, otherwise, setting the value as false, and only calculatingThe mask matrix is the rectangular frame loss and the classification loss of the prediction frame of true;
is a BCE (binary cross entropy) loss function, is used for solving the loss of a single class,
for the CIoU loss, not only the overlap area is considered, but also the center point distance and the aspect ratio are included,l noobj to be a loss of confidence in the label as background,l obj in order to be a loss of confidence in the labeling of the object,aindicating the weight of the calculated classification loss.
In this embodiment, in order to increase the attention to the loss function in the screw detection process, the weight of the loss function of the 160 × 160 detection layer in the overall loss function needs to be the highest, while the weight of the loss function of the 20 × 20 detection layer is the lowest, and meanwhile, for considering the components with larger size such as the fan and the components with smaller size such as the screw and the connecting line, the weights of the loss functions of the 80 × 80 detection layer and the 40 × 40 detection layer need to be kept the same, and the sum of the two weights should be larger than the weight of the 160 × 160 detection layer.
In one or more embodiments of the present invention,a 1 、a 2 、a 3 、a 4 the values of (A) are 0.35, 0.25 and 0.15 in sequence.
Example 5
On the basis of the embodiment, a light source used for collecting an image of the mainboard to be detected is located right above a fan of the mainboard to be detected, after the mainboard quality detection model judges whether the fan of the mainboard to be detected is connected correctly according to the collected first image, if the mainboard fixing screw model in the first image is judged to be incorrect, the mainboard to be detected is rotated by 180 degrees and a second image is collected, whether the mainboard fixing screw model in the second image is correct is judged, if not, the mainboard fixing screw model is judged to be incorrect, and if so, the mainboard fixing screw model is judged to be correct.
In this embodiment, at most two images can be collected for the same main board to be detected. The first image is an image shot by the fan with the light source facing the fan, and if the screw type in the first image is correct, only the first image is shot for the mainboard; however, if the screw model in the first image is wrong, the problem that the model is not easy to identify the fixed screw or the fixed screw is identified incorrectly may be caused by shadow occlusion. At this moment, the main board to be detected can be rotated by 180 degrees and then a second image is acquired, in the second image, the fixing screw area originally far away from the light source is located below the light source, and therefore false identification caused by shadow influence is remarkably reduced. If the type of the mainboard fixing screw in the second image is incorrect, the type of the mainboard fixing screw is judged to be incorrect, otherwise, the type of the mainboard fixing screw is judged to be correct.
Example 6:
as shown in fig. 3, a computer motherboard quality detection system based on artificial intelligence includes:
the image acquisition unit is used for acquiring an image of the mainboard to be detected;
the analysis unit is used for constructing and training a mainboard quality detection model, processing the image of the mainboard to be detected by using the trained mainboard quality detection model, and judging whether the fan of the mainboard to be detected corresponding to the image of the mainboard to be detected is correctly connected and whether the model of a mainboard fixing screw is correct, wherein the mainboard quality detection model is used for carrying out down sampling and up sampling for four times on the image of the mainboard to be detected, and fusing a feature map obtained by the up sampling for the third time and a feature map obtained by the down sampling for the first time to obtain a first feature map with the detection scale of 160 x 160;
and the output unit is used for generating a statistical form according to the judging structure, wherein the statistical form comprises the number of the mainboard with correct and wrong fan connection and the number of the mainboard with correct and wrong mainboard fixing screw type.
In some embodiments, as shown in fig. 4, the image capturing unit includes a housing 1, and a conveyor belt 2 carrying a plurality of holding plates 3 passes through the housing 1, where the holding plates 3 are used for fixing and placing a main board 4 to be detected. The shell 1 is a darkroom, the top of the darkroom is provided with a light source 5 and a camera 6 used for shooting the mainboard 4 to be detected, the camera 6 is connected to a client 8 through a data line 7, and the client 8 is connected with a server 9 through the internet. When image acquisition is carried out, when the mainboard 4 to be detected passes through the camera under, the camera 6 shoots the mainboard 4 to be detected, and the image of the mainboard to be detected is acquired for subsequent identification and judgment.
In one or more embodiments, the housing 1 is a 1000mm x 1000mm square iron container, and the conveyor belt 2 is preferably a TF-SSD conveyor belt. In one or more embodiments, the light source 5 is an LED tube and the industrial camera 6 is a CCD industrial camera.
In some embodiments, as shown in fig. 5 to 9, when the camera 6 is aligned with the center of the main board 4 to be detected, the light source 5 is eccentrically disposed and aligned with the fan 41 on the main board 4 to be detected, so as to collect the first image, determine whether the fan and its connecting line area are correct, and record. The bottom of the containing plate 3 is provided with a driving device 10, and the driving device 10 is used for rotating the containing plate 3. When the type of the main board fixing screw in the first image is wrong, the driving device 10 rotates the containing plate by 180 degrees, so that the area far away from the light source 5 in the first image is positioned under the light source 5, the camera 6 collects a second image for judging whether the type of the main board fixing screw is correct or not, and the judgment result of the second image is used as the standard. In one or more embodiments, the drive means 10 is preferably a stepper motor or a servo motor.
In some embodiments, as shown in fig. 6 and 7, a limiting member 11 is further disposed on an outer side of the containing plate 3, one or more sliding stabilizing mechanisms are fixed on the containing plate 3, the number of the sliding stabilizing mechanisms is preferably even, and each sliding stabilizing mechanism is uniformly distributed along a circumferential direction of the containing plate 3. As shown in fig. 8, the sliding stabilization mechanism includes a sleeve 32 fixedly installed on the sidewall of the containing plate 3, a spring 34 is installed in the sleeve 32, an expansion link 33 capable of sliding along the sleeve 32 is connected to the spring 34, and a roller 31 is installed on the expansion link 33. In the process of rotating the containing plate 3 relative to the limiting member 11, the spring 34 is in a compressed state, and the roller 31 abuts against the inner wall of the limiting member 11.
In some embodiments, as shown in fig. 7, an arc-shaped groove 112 is further disposed on an inner wall of the limiting member 11, and a sensor 111 is disposed in the arc-shaped groove 112, where the sensor 111 may be a distance sensor or a pressure sensor. When the sliding stabilizing mechanism passes through the arc-shaped groove 112, the roller 31 slides into the arc-shaped groove 112, meanwhile, the spring 34 extends and pushes the telescopic rod 33, the roller 31 is enabled to abut against the arc-shaped groove 112, the arc-shaped groove 112 plays a positioning role, and after the distance or the pressure signal collected by the sensor 111 in the arc-shaped groove 112 is greater than a threshold value, an electric signal is sent to the controller of the camera 6 to control the camera 6 to shoot a second image.
In one or more embodiments, as shown in fig. 9, a diffuse reflection layer is further disposed on the limiting member 11 and located above a contact point between the sliding stabilizing mechanism and the limiting member 11, the diffuse reflection layer can further improve the illumination intensity of an image acquisition area and reduce shadows, and can prevent problems such as highlight, glare and overexposure caused by excessive local illumination intensity, so that the light intensity of the image acquisition area is more uniform, which is beneficial to improving the quality of the acquired image to be detected.
As used herein, "first," "second," etc. merely distinguish the corresponding components for clarity of description and are not intended to limit any order or to emphasize importance, etc. Further, the term "connected" used herein may be either directly connected or indirectly connected via other components without being particularly described.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (9)
1. A computer mainboard quality detection method based on artificial intelligence is characterized by comprising the following steps:
collecting an image of a mainboard to be detected;
constructing a mainboard quality detection model, and training the mainboard quality detection model;
processing the image of the mainboard to be detected by using the trained mainboard quality detection model, and judging whether the fan of the mainboard to be detected corresponding to the image of the mainboard to be detected is correctly connected and whether the type of the mainboard fixing screw is correct;
the main board quality detection model is used for carrying out four times of down-sampling and three times of up-sampling on an image of a main board to be detected, fusing a feature map obtained by the third time of up-sampling and a feature map obtained by the first time of down-sampling, and obtaining a first feature map with the detection scale of 160 multiplied by 160;
the mainboard quality detection model judges whether the mainboard fan is connected correctly according to the collected first image, if the mainboard fixing screw model in the first image is judged to be incorrect, the mainboard is rotated by 180 degrees and a second image is collected, whether the mainboard fixing screw model in the second image is correct is judged, if not, the mainboard fixing screw model is judged to be incorrect, if so, the mainboard fixing screw model is judged to be correct.
2. The method as claimed in claim 1, wherein the computer motherboard quality inspection method based on artificial intelligence, the main board quality detection model is used for carrying out four times of down-sampling on an image of a main board to be detected to respectively obtain a first down-sampling feature map, a second down-sampling feature map, a third down-sampling feature map and a fourth down-sampling feature map, the fourth down-sampling feature map is fused with the third down-sampling feature map after the first up-sampling to obtain a first fused feature map, the first fusion characteristic diagram is fused with a second down-sampling characteristic diagram after characteristic extraction and second up-sampling to obtain a second fusion characteristic diagram, and the second fused feature map is subjected to feature extraction, third upsampling and then fused with the first downsampling feature map to obtain a third fused feature map, and the third fused feature map is subjected to feature extraction to obtain the first feature map.
3. The computer motherboard quality inspection method based on artificial intelligence of claim 2, wherein the motherboard quality inspection model is configured to fuse the second fused feature map and the third fused feature map to obtain a fourth fused feature map, the fourth fused feature map is subjected to feature extraction to obtain a second feature map with a detection scale of 80 × 80, the fourth fused feature map is configured to fuse the fourth fused feature map, the third downsampled feature map and the first fused feature map to obtain a fifth fused feature map, the fifth fused feature map is subjected to feature extraction to obtain a third feature map with a detection scale of 40 × 40, and the fifth fused feature map and the fourth downsampled feature map are fused to obtain a sixth fused feature map, and the sixth fused feature map is subjected to feature extraction to obtain a fourth feature map with a detection scale of 20 × 20.
4. The method of claim 3, wherein the first, second, third, and fourth feature maps are subjected to target prediction, and the anchor point configuration of the target prediction comprises anchor points [5,6,8,14,15,11 ].
5. The method according to claim 4, wherein a loss function is used to calculate the target prediction loss of the motherboard quality inspection model, and the formula of the loss function is:
wherein,loss rect in order to predict the total loss of the box,loss obj for the purpose of the overall loss of confidence,loss clc to classify the total loss;loss rect160 、loss obj160 、loss clc160 respectively representing the prediction frame loss, the confidence coefficient loss and the classification loss of the first feature map;loss rect80 、loss obj80 、loss clc80 respectively representing the prediction frame loss, confidence coefficient loss and classification loss of the second feature map;loss rect40 、loss obj40 、loss clc40 the prediction frame loss, the confidence coefficient loss and the classification loss of the third feature map are respectively;loss rect20 、loss obj20 、loss clc20 the predicted frame loss, the confidence coefficient loss and the classification loss of the fourth feature map are respectively obtained;a 1 、a 2 、a 3 、a 4 the weights of the first to fourth characteristic diagrams in turn, wherein,a 1 >a 2 =a 3 >a 4 and is made ofa 1 <a 2 +a 3 。
6. The computer motherboard quality detection method based on artificial intelligence of claim 3, wherein the motherboard quality detection model is used for slicing an image of a motherboard to be detected and then performing feature extraction to obtain first to fourth feature maps.
7. The artificial intelligence based computer motherboard quality inspection method of claim 3, wherein the motherboard quality inspection model comprises a classification correction module, the classification correction module is used for weighting the output channel prediction constant weight, and the formula of the classification correction module is as follows:
wherein, the
Is the weighted sum of the pixels, and the weighted sum is,Hin order to be the height of the pooling window,Win order to be the width of the pooling window,F sq (pixel i ) Indicating that a global weighted average pooling operation is performed, and s is an output value of the pooling window.
8. The method as claimed in claim 1, wherein the statistical report is generated by counting the number of correctly and incorrectly connected mainboards of the fan and the number of incorrectly and correctly fixed screws of the mainboards.
9. The utility model provides a computer motherboard quality detection system based on artificial intelligence which characterized in that includes:
the image acquisition unit is used for acquiring an image of the mainboard to be detected;
the analysis unit is used for constructing and training a mainboard quality detection model, processing the image of the mainboard to be detected by using the trained mainboard quality detection model, and judging whether the fan of the mainboard to be detected corresponding to the image of the mainboard to be detected is correctly connected and whether the model of a mainboard fixing screw is correct, wherein the mainboard quality detection model is used for carrying out down sampling and up sampling for four times on the image of the mainboard to be detected, and fusing a feature map obtained by the up sampling for the third time and a feature map obtained by the down sampling for the first time to obtain a first feature map with the detection scale of 160 x 160;
the output unit is used for generating a statistical form according to the judgment structure, wherein the statistical form comprises the number of the mainboard with correct and wrong fan connection and the number of the mainboard with correct and wrong mainboard fixing screw types;
the image acquisition unit comprises a shell (1), a light source (5) and a camera (6) are arranged at the top of the shell (1), a conveying belt (2) penetrates through the shell (1), a plurality of containing plates (3) are loaded on the conveying belt (2), the containing plates (3) are used for fixing and placing a main board (4) to be detected, a driving device (10) is installed at the bottom of each containing plate (3), a limiting piece (11) is further arranged on the outer side of each containing plate (3), one or more sliding stabilizing mechanisms are fixed on each containing plate (3), each sliding stabilizing mechanism comprises a sleeve (32) fixedly installed on the side wall of each containing plate (3), a spring (34) is arranged in each sleeve (32), a telescopic rod capable of sliding along the corresponding sleeve (32) is connected onto each spring (34), and a roller (31) is arranged on each telescopic rod (33), in the process that the containing plate (3) rotates relative to the limiting piece (11), the spring (34) is in a compressed state, and the roller (31) is abutted against the inner wall of the limiting piece (11).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210698021.XA CN114782437B (en) | 2022-06-20 | 2022-06-20 | Computer mainboard quality detection method and system based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210698021.XA CN114782437B (en) | 2022-06-20 | 2022-06-20 | Computer mainboard quality detection method and system based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114782437A CN114782437A (en) | 2022-07-22 |
| CN114782437B true CN114782437B (en) | 2022-09-02 |
Family
ID=82420820
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210698021.XA Expired - Fee Related CN114782437B (en) | 2022-06-20 | 2022-06-20 | Computer mainboard quality detection method and system based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114782437B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115346172B (en) * | 2022-08-16 | 2023-04-21 | 哈尔滨市科佳通用机电股份有限公司 | Method and system for detecting lost and broken hook lifting rod reset spring |
| CN115546187A (en) * | 2022-10-28 | 2022-12-30 | 北京市农林科学院 | Agricultural pest and disease detection method and device based on YOLO v5 |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111292305A (en) * | 2020-01-22 | 2020-06-16 | 重庆大学 | Improved YOLO-V3 metal processing surface defect detection method |
| CN112257849A (en) * | 2020-10-23 | 2021-01-22 | 西南石油大学 | Intelligent detection method, system and device based on deep learning and application thereof |
| CN112697806A (en) * | 2021-01-22 | 2021-04-23 | 南京英诺森软件科技有限公司 | Defect recognition device and system based on image recognition technology |
| CN112710671A (en) * | 2021-01-08 | 2021-04-27 | 广州南沙珠江啤酒有限公司 | Intelligent visual inspection device and inspection method for defects of trolley overhead rail trolley line |
| CN114240939A (en) * | 2022-02-24 | 2022-03-25 | 苏州浪潮智能科技有限公司 | Method, system, equipment and medium for detecting appearance defects of mainboard components |
| CN114529799A (en) * | 2022-01-06 | 2022-05-24 | 浙江工业大学 | Aircraft multi-target tracking method based on improved YOLOV5 algorithm |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104406984A (en) * | 2014-12-16 | 2015-03-11 | 浪潮电子信息产业股份有限公司 | Server mainboard material consistency image detection method |
| TWI697004B (en) * | 2018-10-17 | 2020-06-21 | 大陸商合肥沛睿微電子股份有限公司 | Power quality detecting system and power quality detecting module |
| US11216932B1 (en) * | 2021-03-26 | 2022-01-04 | Minds AI Technologies Ltd | Electronic substrate defect detection |
| CN113591790A (en) * | 2021-08-16 | 2021-11-02 | 上海铂端科技有限公司 | System, method and device for realizing production line assembly flow behavior monitoring based on computer vision, processor and storage medium thereof |
| CN113688723B (en) * | 2021-08-21 | 2024-03-19 | 河南大学 | Infrared image pedestrian target detection method based on improved YOLOv5 |
| CN114494908A (en) * | 2022-02-11 | 2022-05-13 | 南通大学 | Improved YOLOv5 power transmission line aerial image defect detection method |
-
2022
- 2022-06-20 CN CN202210698021.XA patent/CN114782437B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111292305A (en) * | 2020-01-22 | 2020-06-16 | 重庆大学 | Improved YOLO-V3 metal processing surface defect detection method |
| CN112257849A (en) * | 2020-10-23 | 2021-01-22 | 西南石油大学 | Intelligent detection method, system and device based on deep learning and application thereof |
| CN112710671A (en) * | 2021-01-08 | 2021-04-27 | 广州南沙珠江啤酒有限公司 | Intelligent visual inspection device and inspection method for defects of trolley overhead rail trolley line |
| CN112697806A (en) * | 2021-01-22 | 2021-04-23 | 南京英诺森软件科技有限公司 | Defect recognition device and system based on image recognition technology |
| CN114529799A (en) * | 2022-01-06 | 2022-05-24 | 浙江工业大学 | Aircraft multi-target tracking method based on improved YOLOV5 algorithm |
| CN114240939A (en) * | 2022-02-24 | 2022-03-25 | 苏州浪潮智能科技有限公司 | Method, system, equipment and medium for detecting appearance defects of mainboard components |
Non-Patent Citations (1)
| Title |
|---|
| 基于RetinaNet的手机主板缺陷检测研究;马美荣等;《计算机工程与科学》;20200430;第42卷(第4期);第2节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114782437A (en) | 2022-07-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114782437B (en) | Computer mainboard quality detection method and system based on artificial intelligence | |
| CN111325713B (en) | Neural network-based wood defect detection method, system and storage medium | |
| KR102022496B1 (en) | Process management and monitoring system using vision image detection and a method thereof | |
| CN105812746B (en) | A kind of object detection method and system | |
| CN111476827B (en) | Target tracking method, system, electronic device and storage medium | |
| CN111937034A (en) | Learning data set creation method and device | |
| CN113723157B (en) | Crop disease identification method and device, electronic equipment and storage medium | |
| CN112633255A (en) | Target detection method, device and equipment | |
| CN116228741A (en) | PCBA (printed circuit board assembly) component defect detection method and device | |
| CN116245882A (en) | Circuit board electronic element detection method and device and computer equipment | |
| CN115909157A (en) | Machine vision-based identification detection method, device, equipment and medium | |
| CN117726991B (en) | High-altitude hanging basket safety belt detection method and terminal | |
| CN114462469B (en) | Training method of target detection model, target detection method and related device | |
| CN115880260A (en) | Method, device and equipment for detecting base station construction and computer readable storage medium | |
| CN113205511B (en) | Electronic component batch information detection method and system based on deep neural network | |
| CN112561885B (en) | YOLOv 4-tiny-based gate valve opening detection method | |
| CN116580026B (en) | Automatic optical detection method, equipment and storage medium for appearance defects of precision parts | |
| CN115984593B (en) | Method, device, equipment and storage medium for detecting defects of cigarette filter sticks | |
| CN113076889A (en) | Container lead seal identification method and device, electronic equipment and storage medium | |
| CN110826473A (en) | Neural network-based automatic insulator image identification method | |
| CN115330997A (en) | YOLOv4 neural network-based control box assembly defect detection method, device and medium | |
| CN115170829A (en) | System and method for monitoring and identifying foreign matters in generator rotor vent hole | |
| CN111985334B (en) | Gun detection method, system, device and storage medium | |
| KR102022494B1 (en) | System for automatic generating of documents using vision image detection and a method thereof | |
| CN113642565B (en) | Object detection method, device, equipment and computer readable storage medium |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220902 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |