Sunflower seeds, recognized for their nutritional value and taste, are a well-loved snack. However, throughout their growth and storage, sunflower seeds can develop various defects that not only compromise their quality but also present potential health hazards. To address these issues and ensure adherence to safety standards, we investigate the use of THz spectroscopy and imaging techniques for non-destructive identification and classification of common defects in sunflower seeds. The study begins by analyzing spectroscopy features to identify defective seeds, particularly those affected by mildew. It establishes three qualitative discrimination models (support vector machine, random forest, and backpropagation neural networks), which achieve overall accuracies of 88.3%, 91.7%, and 95%, respectively. Furthermore, THz transmission imaging is employed as a quantitative method to visualize the internal structure of sunflower kernels and provide precise plumpness estimates. A noteworthy innovation is the analysis of time delays in reflected pulses at each pixel, enabling the extraction of valuable kernel thickness information. These data are then utilized to convert traditional two-dimensional scanning data into intricate three-dimensional (3D) images, facilitating direct measurements of both 3D plumpness and kernel weight. The findings have significant implications for improving the quality and safety of sunflower seeds and may extend to the assessment of other agricultural products, contributing to enhanced quality control in the food industry.