The current work presents the collision integral data for N(4
S)–N(4
S, 2
D, 2
P) and O(3
P, 1
D, 1
S)–O(3
P, 1
D, 1
S) interactions in the temperature range of 500–50 000 K. The collision integrals are calculated based on high-quality potential energy curves (PECs) obtained from fitting the high-level
a
b
i
n
i
t
i
o
calculation data in a wide energy range to the neural network (NN) functions. In the construction of PECs, the diabatic PECs are adopted when avoided crossings exist because the diabatic paths are much more likely to be followed for such situations. Moreover, the nonadiabatic transition effects are estimated to be negligible for PECs crossings. The accuracy of traditional analytical formulas to fit PECs are also examined. It is found that the collision integral calculations are sensitive to the accuracy of PECs and the NN based PECs overwhelm the others. The contribution of inelastic excitation exchange processes to the diffusion collision integrals are also computed by using an accurate evaluation of the differences of PECs for gerade and ungerade pairs of excited atoms. Finally, based on the new collision integral data, we calibrate the collision model parameters suitable for the widely used particle simulation methods. The collision integrals and collision models developed in this work can be used to support high-confidence simulations of weakly ionized air plasma problems.