Recognition of weld defects from X-ray images based on improved convolutional neural network

When convolutional neural network (CNN) is used for welding defect detection image recognition, the recognition result will be affected by many factors such as human factors, the activation function is sensitive to input parameters, and the edge features are weakened. In order to overcome the above problems, the methods include image processing, exponential linear unit (ELU) activation function and improved pooling model are used. According to the experiment, the image processing method can effectively segment the weld and defects, and the defect location in the weld image can be located. Using the ELU activation function in the CNN model can improve the robustness of the neural network to the input parameters and increase the sparsity of the network to increase the model's convergence speed. The improved pooling method based on grayscale adaptation can increase the extraction range of weld defect features and reduce the impact of noise, and has certain dynamic adaptability to the defect features. The result shows that the improved convolutional neural network(ICNN) method can effectively improve the accuracy of recognition in weld image recognition, and the overall recognition rate can reach 98.13%.

Automated summary

In this paper, methods such as image processing, ELU activation function, and improved pooling model are proposed to address the challenges in weld defect detection image recognition using convolutional neural network (CNN). The experiments show that these methods can effectively improve the accuracy and convergence speed of weld image recognition.