This study proposes a method for detecting microscopic defects in polymer materials based on convolutional neural networks (CNNs). By designing a deep convolutional network structure, incorporating attention mechanisms and feature pyramid networks, and optimizing the loss function and training strategy, the detection performance has been significantly improved. Experiments on a large-scale self-built dataset show that the model has achieved excellent results in both average precision and inference speed. Compared to traditional image processing methods and other deep learning models, this approach exhibits clear advantages in detection accuracy and speed, especially in detecting small target defects. The research provides an efficient and reliable solution for quality control in the polymer material manufacturing industry, with significant theoretical significance and practical value in automation and quality management, offering new ideas and methods for technological innovation and industrial upgrading in related fields.