Ultrasonic Image Recognition of Terminal Lead Seal Defects Based on Convolutional Neural Network

At present, high-voltage cables are widely used in urban power grid transmission projects. As an important part of high-voltage cable terminal accessories, lead seals at high-voltage cable terminals will have defects such as holes, cracks or debonding due to unqualified installation quality or external forces during operation, affecting the safe and stable operation of power systems. The traditional ultrasonic phased array detection method for lead seal defects is to process the ultrasonic defect image and observe it manually, which has low efficiency and accuracy. Traditional machine learning methods need to manually select the detection object features, lack of adaptability and robustness, and have low accuracy of target detection. In order to improve the intelligent level of lead seal defect detection, an ultrasonic image recognition method of lead seal defect based on convolutional neural network is proposed, which can automatically learn features from the ultrasonic image of lead seal defect and complete defect classification and recognition. The ultrasonic image sample library of four typical lead seal defects was established, and the ultrasonic image recognition model of lead seal defects was built. The model was trained and tested by using standardized ultrasonic image data. The results show that by adjusting the convolution neural network test parameters, different types of defects in lead seal can be quickly and accurately identified, and the accuracy rate can reach 100%. It shows that the method has good robustness, strong anti-interference ability and good detection performance for lead seal defects, and has a good application prospect in the actual terminal lead seal defect detection.