Research On Qualification Detection Of Cable Joint Solder Joint Based On Deep Learning

Most electronic products are produced in assembly-line mode,and each process will affect the quality and service life of the products.Among them,solder joint detection is an important production process.Whether the solder joint qualification detection can be completed accurately and efficiently directly determines the product quality and production efficiency.At present,most of the cable joint solder joint qualification inspections are conducted by artificial vision,which is subjective and has the problems of low accuracy and reliability.In order to solve the above problems,the paper mainly uses computer vision technology to collect solder joint images and uses deep learning technology to quickly detect solder joint eligibility,so as to realize automatic detection of solder joints.It is an important significance to ensure the quality and production efficiency of electronic products based on production lines.In this paper,the solder joint data image acquisition,processing,and solder joint qualification are studied to realize the automatic detection of welding spots.The specific research work is as follows:(1)Create a USB cable connector solder joint data set.As the application of depth learning technology to the detection of USB cable joint solder joints is a completely new field,there is no data set available.Therefore,this paper creates a data set of USB cable connector solder joint.In this paper,firstly,the data are cleaned,then the tilted images are effectively corrected by using Discrete Fourier Transform,Hough Line Detection,and Affine Transform.The experimental results show that the correction effect is good.Finally,the USB solder joint data set is created by using the Labelme image labeling tool.(2)Solder joint qualification detection based on DCNN.To solve the problem of the low recognition rate of traditional solder joint detection algorithms,this paper proposes a solder joint qualification detection model based on DCNN.In view of the insufficient features extracted by the shallow CNN networks,for example,the general shallow network can only extract some low-level features,and the detection results for lines,edges,and edges are not good enough to meet the requirements of factories.So the deep convolution neural network is used to detect the solder joints.Batch normalization+ Xavier initialization is adopted in DCNN to optimize the model.Experimental results show that the algorithm in this paper has a high recognition rate and fast detection speed.In addition,the problem of loss function instability is solved by adding batch standardization between all connection layers.(3)Improved Mask R-CNN solder joint qualification detection.For the sake of solving the problem that the DCNN model has poor detection effect on solder joints of six-core joints,this paper proposes a solder joint qualification detection method based on improved Mask R-CNN.This model optimizes the network structure and improves the ResNet50 feature extraction network,and FPN network is used to fuse the features of low-level and high-level to learn the features with different receptive fields.In addition,the RPN network is used to generate candidate frames,and each solder joint in the solder joint image is tested through the candidate frames so that the problem of low detection rate of six-core joints is solved.The experimental results prove that the improved Mask R-CNN algorithm has high recognition rate and better detection speed than the original Mask R-CNN.