Research On PCB Fault Diagnosis Based On Tiny Object Detection

With the rapid development of science and technology and the advent of the information age,electronic devices have already been widely used in all aspects of People’s Daily life,the rapid increase in the number of components used in electronic equipment makes its internal circuit structure increasingly complex.As the basic parts of electronic equipment-printed circuit boards with the volume of electronic equipment is becoming smaller and more integrated,resulting in the probability of failure and detection difficulties are greatly increased.Therefore,in order to reduce the difficulty and cost of circuit board fault detection,it is necessary to explore and research new circuit board fault diagnosis methods.In this paper,we study the printed circuit board fault diagnosis based on the tiny object detection method,improve the model structure for circuit board tiny fault points,and optimize the data set for the improved model.(1)Improved CenterNet-based circuit board fault diagnosisThe circuit board fault diagnosis problem is considered as a tiny object detection problem,and the object detection model based on Anchor-free is introduced to build and implement the Center Net-based circuit board fault diagnosis model.In order to better meet the needs of circuit board fault diagnosis tasks,this paper introduces the attention-based mechanism of Res Ne St to improve the network feature extraction capability,and builds and implements a circuit board fault diagnosis model based on the fusion of Res Ne St and Center Net,which improves the detection accuracy of the model.(2)ESRGAN super-resolution-based circuit board fault diagnosisIn order to solve the problem that the whole area of the circuit board image to be detected is small and the distinction between foreground and background is not obvious compared with other object recognition,the enhanced Generative Adversarial Networks is introduced to construct and implement the circuit board fault diagnosis model based on ESRGAN super-resolution.The model obtained clearer edges and richer texture details by super-resolution reconstruction of the circuit board image data set,which made it easier for the network model to find tiny fault points.Then,Mosaic algorithm was used for data enhancement to improve the data diversity and ultimately improve the model accuracy.The experimental comparison proves that the ESRGAN super-resolution algorithm is effective in improving the network model recognition accuracy,which is better than other target detection models.