Identification And Location Of Electronic Components Based On Convolutional Neural Networks

In the production,application and recycling of electronic components,the classification and positioning are very important basic work.With the development of science and technology,there are more and more types of electronic components,and they are developing in the direction of miniaturization and chipping.Artificial eye inspection,traditional image classification and image detection methods are no longer suitable for the current environment.At the same time,in order to liberate manpower and realize the automation of production,the current research hotspot convolutional neural network can meet this requirement and also perform well in the field of image feature learning.Therefore,this paper proposes a method of recognition and location of electronic components based on convolutional neural network.In order to identify and classify electronic components simply and efficiently,this paper combines the classification of electronic components with convolutional neural networks,and proposes a method for identifying electronic components based on convolutional neural networks.Compared with traditional image classification algorithms,this method requires only simple preprocessing of images and can be used as input for network model training.Moreover,convolutional neural network can reduce the number of parameters and the complexity of calculation.This paper builds a data set for training and testing,and the average accuracy of the final test set reached 92.20%.The experimental results show that the convolutional neural network model can automatically extract features even without complex preprocessing of images,and can identify multiple components with high precision and low complexity,which can overcome many shortcomings of traditional image classification algorithm.For the localization of multiple electronic components on a circuit board,this paper proposes two different segmentation methods based on two basic semantic segmentation models: cross-layer semantic feature fusion segmentation method and Shared semantic feature fusion segmentation method.These two methods improve the existing network model as follows:(1)The first method optimizes the network structure: rewrites the input mode of the training network,uses the transplant method to obtain the weight of the pre-training network training,and multi-scale the output of the multi-layer pooling layer.The training results show that the method can simultaneously locate a variety of electronic components,the average accuracy of electronic component positioning reaches 83.00%,the mean intersection over union reaches 0.78,and when the test picture is input into the trained network,the network can output a semantic segmentation picture with better segmentation effect.(2)The second method optimizes the detection frame by sharing the extracted convolution features with different convolutional neural networks and customizing the network training method,and then integrating the NMS and other algorithms,to optimize the result of the test box,implements the resistor,capacitor,IC chips of the effective positioning.By training the semantic segmentation network,the precision of these three electronic components reached 80.22%,67.54%,and 99.98%,respectively.During the testing phase,the average accuracy of these electronic components was above 90.00%.These two methods guarantee the real-time positioning while ensuring the accuracy.