In recent years, composite materials have made remarkable progress in improving production efficiency, reducing manufacturing costs, and optimizing overall performance, and they have been successfully applied in engineering practice. However, during manufacturing and service, composites are prone to internal defects such as delamination, debonding, and cracking, which can lead to structural failure. These defects degrade material performance and pose serious safety concerns. Therefore, developing efficient and reliable Non-Destructive Testing (NDT) techniques is crucial to ensure the quality of composite materials. As an important optical NDT method, shearography can accurately capture the shape and location of defects. With advantages such as non-contact measurement, high sensitivity, and full-field detection, shearography shows great potential for detecting defects in composites.Due to significant differences in the physical properties of different composite materials—such as thermal conductivity and stiffness—their responses to external loads also vary considerably. As a result, the choice of loading strategy has a significant impact on detection accuracy and the recognition accuracy of defect fringe patterns. However, current loading strategies largely depend on the operator’s experience, lacking systematic parameter selection guidelines. This often leads to overloading or underloading, thus affecting the detection performance.To evaluate the influence of different loading strategies on defect detection effectiveness, this study selects two typical loading methods—thermal loading and vacuum loading—to perform quantitative loading experiments on various composite materials. Phase maps of defective regions are obtained using shearography. After filtering the results from both loading methods, a YOLOv9-based defect phase map recognition model is constructed to intelligently identify and analyze the detection images. By comparing the recognition rates of defect fringe maps under different loading strategies, the most suitable loading strategy for each material is determined.The experimental results are as follows for aluminum honeycomb panels, the highest defect detection rate under thermal loading is 38.26%, while under vacuum loading it reaches 91.3%; for carbon fiber-reinforced composites, the highest detection rate under thermal loading is 97.28%, while only 18.9% under vacuum loading; for aluminum skin-paper honeycomb panels, the detection rates are 38.26% for thermal loading and 91.27% for vacuum loading.These results demonstrate that for metal-based composites, due to their high thermal conductivity, thermal loading cannot effectively stimulate deformation differences between defective and intact areas, leading to poor detection results. In contrast, vacuum loading produces significantly better outcomes. On the other hand, for carbon fiber-reinforced composites, which possess low thermal conductivity and high stiffness, the temperature gradient between defective and intact regions is relatively large, enabling thermal loading to more effectively highlight defect features. In terms of loading magnitude, the experiments show that at the initial stages of loading, the amount of deformation is small, resulting in faint fringe patterns and low defect recognition rates. As the loading level increases, the fringe contrast becomes clearer, and recognition performance improves. However, when the loading becomes too high, the fringe density becomes excessive, which adversely affects recognition accuracy. Notably, the peak recognition rate under different loading levels varies across material types. For composites with lower stiffness, excessive loading can lead to overly dense fringes, reducing detection effectiveness. Therefore, it is essential to select appropriate loading magnitudes based on the stiffness characteristics of each material.The findings of this study provide both theoretical guidance and practical reference for selecting loading strategies in shearography-based NDT, and are of significant value in improving the accuracy and automation level of defect detection in composite materials.