Small Sample Containment Vessel Cracks Detection Algorithm Based the Latent Diffusion Model and Improved Faster-RCNN

Nuclear energy is a clean, efficient, and sustainable energy widely used in power generation. In recent years, nuclear power has experienced rapid development worldwide. As a result, the safety of nuclear power, particularly the containment vessel that serves as the final barrier to avoid the release of radioactive materials into the surrounding atmosphere, has received increasing attention. The surface cracks of the containment vessel are an important factor threatening its lifespan. However, the current detection of surface cracks in the containment vessel mainly relies on inefficient and dangerous manual detection. This paper proposes a small samples containment vessel cracks detection algorithm using the latent diffusion model data expansion and improved Faster-RCNN. There are primarily two parts to this approach. the first part uses the latent diffusion model to expand data and constructs the DD training dataset, solving the problem of small samples being difficult to train and easily overfitting in deep learning object detection methods. The other part uses clustering algorithms to obtain anchor boxes that adapt to the geometric features of cracks, improving the detection performance of Faster-RCNN in containment vessel cracks detection tasks. In addition, this article compares different detection methods to evaluate the accuracy of the proposed method. This method has been applied to the detection of containment vessel cracks in practical sites.