Abstract: Tuna samples were processed using steaming, roasting, air frying, and low-temperature boiling methods. Changes in sensory attributes, color, texture, moisture content, pH, crude protein, and crude fat were systematically analyzed. Three prediction models—Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), and Back-Propagation Neural Network (BPNN)—were developed based on moisture content as the characteristic parameter, combined with spectral preprocessing techniques, to establish a non-destructive detection method using hyperspectral imaging technology. The results demonstrated that moist-heat treatments (low-temperature boiling and steaming) resulted in lower hardness, effectively preserved moisture, and yielded a soft texture. In contrast, dry-heat treatments (roasting and air frying) promoted the Maillard reaction and moisture evaporation, imparting a crisp texture and rich aroma. As the thermal processing intensity increased, the moisture content decreased significantly (P<0.05), while the relative contents of protein and fat increased due to the concentration effect. Among the three models developed, the BPNN model exhibited superior prediction accuracy and stability compared to the PLSR and SVR models, achieving optimal performance (R²=0.9239, RMSEP=2.0284). The BPNN model enables accurate and rapid non-destructive detection of moisture content in tuna, providing a highly reliable spectral analysis solution for quality monitoring of thermally processed tuna products.