Research On Deep Learning And Its Application On The Electronic Components Counting

With the rapid development of science and technology,electronic products have become indispensable in production and daily life.In the production and processing of electronic products,real-time accurate electronic components counting is important to improving automaticity.However,electronic components have a great variety in types and shapes,and most of them are small in size,which makes the traditional manual electronic components counting complicate,time-consuming and easily influenced by workers' working state.In order to overcome the inefficiency of manual means,an electronic components counting algorithm based on image processing is proposed.But the experimental results indicate that the accuracy of algorithm highly relies on manual adjustment of the parameters,and the detection is slow because the process requires a large amount of access to the image pixels.Convolution Neural Network(CNN)has ability to learn features from samples automatically because it simulates the visual processing mechanism of human.So CNN outperforms other traditional algorithms in the field of object recognition.The Regions with CNN(R-CNN)model and its improved model Fast R-CNN and Faster R-CNN proposed in recent years have good performance in the field of object recognition.To overcome the disadvantages of electronic components counting algorithm based on image processing and the miss detection of small-size,intensively distributed or tilted electronic components using Faster R-CNN,this paper proposes an improved model based on Faster R-CNN which is optimized in three aspects including multi-convolutional information fusion,context extension and position-sensitive convolution and pooling.Our experimental results show that the optimized algorithm has faster detection speed and can improve the accuracy of counting from 87.8% to 98.3%,compared with the algorithm based on image processing.At last,this paper implements an electronic components counting software system based on the proposed improved algorithm,which has great value in practice.