Abstract: Safflower is a high-value crop with multiple economic uses. However, the natural air-drying process alters the mechanical properties of safflower filaments, leading to damage and shedding during mechanical harvesting and consequently resulting in loss. Numerical simulation studies of the interaction between filaments and harvesting machinery can optimize component design and reduce losses. This research measured and calibrated the intrinsic and contact parameters of dried safflower filaments, providing data support for a discrete element simulation model. The discrete element parameter of the dried safflower filaments was determined by combining the physical and simulation tests. Five-point sampling of the dried safflower filaments was conducted using a roller-brush harvester. The filaments were categorized into the intact, petal-, stalk-, and dual-damaged filaments, according to the damage targets. Three-view images were captured by an electron microscope. A three-dimensional model of the filaments was constructed in conjunction with the three-view images after measuring the dimensions of each part of the filament. A discrete element filling model was established using an automatic filling method, according to different smoothness levels and three-dimensional coordinate particle distributions of the filaments. The average mass of a single filament was measured at 8.915×10^-4 g using an electronic balance. A super-depth-of-field microscope was then employed to obtain the filament volume. The density of dried safflower filaments was then determined to be 88.923 kg/m³. The collision recovery coefficient and the friction coefficient of the filaments were measured using the free-fall and the modified inclined plane. The filament plates and columns were employed in the static and rolling friction coefficient tests of the modified inclined plane. Results showed that the collision recovery coefficient between the filaments and stainless steel ranged from 0.046 to 0.187, the static friction coefficient was from 0.229 to 0.322, while the rolling friction coefficient was from 0.077 to 0.091. Furthermore, the impact recovery coefficient between filaments was from 0.031 to 0.228, while the static friction coefficient was from 0.301 to 0.743, and the rolling friction coefficient was from 0.085 to 0.122. In the free-fall validation test, the rebound heights recorded for the steel plate and filament sample plate were 5.306 and 4.858 mm, respectively, while the simulated values were 5.086 and 4.686 mm. The relative errors between the simulation and actual results were 4.15% and 3.54%, indicating reliable measurements of the crash recovery coefficient. In the modified inclined plane validation test, the measured inclination angles for the steel and filament sample plates were 17.8° and 31.4°, compared to simulated values of 17.58° and 31.32°, leading to relative errors of 1.24% and 0.25%. Additionally, the measured horizontal rolling distances were 138.900 and 35.669 mm for the steel and filament plates, with simulated distances of 134.206 and 34.840 mm, resulting in relative errors of 3.38% and 2.32%. These results confirm the accuracy of the friction coefficient measurements. The angle of repose was determined to be 46.618° using MATLAB software. The influencing factors were screened after Plackett-Burman optimizations with the steepest slope and Box-Behnken experiments. Ultimately, a combination of the optimal parameters was calibrated: a coefficient of static friction between filaments of 0.347, a coefficient of rolling friction of 0.086, and a coefficient of restitution of 0.211. The stacking angle was simulated as 46.172°, with a relative error of only 0.96%. Coupled gas-solid simulations were then conducted on the roller-brush device for the dried safflower filament harvesting using the DEM-CFD method. Simulation tests confirmed that the filament loss rate was 4.78% in the harvesting mechanism, with a relative error of 8.43%, compared with the field trials. Both values were below the 10% error threshold in a typical simulation. Both values were below the 10% error threshold in a typical simulation. A discrete element model can also provide a theoretical basis for the material properties of the dried safflower filaments and the mechanical harvesting.