Abstract: In thermal buckling experiments of thin plates under nonuniform temperature rise and complex boundary constraints, stress fields cannot be measured directly, and the thermoelastic parameters cannot be inferred from full-field digital image correlation (DIC) displacements alone. This study proposes a structural-level equivalent thermoelastic-parameter inversion method that combines high-temperature DIC full-field displacements with membrane weak-form equilibrium constraints. First, the mid-plane membrane strain is calculated from the three-dimensional DIC displacement field, and the nonuniform temperature field is reconstructed by combining thermocouple measurements. Then, the elastic modulus and thermal expansion coefficient are expressed as continuous functions of a normalized thermal-loading state variable and are differentiably represented by a two-stage feedforward neural network. On this basis, a membrane thermoelastic trial stress field under the plane-stress condition is constructed from the observed membrane strain field and temperature field, and the weak-form equilibrium residual of effective internal nodes is taken as the minimization objective to invert the thermoelastic parameters varying with the thermal-loading state. A trend-consistency term is included, while the parameter functions are represented in a continuously differentiable form by tanh feedforward neural networks; no separate smoothness penalty is imposed. A nonuniform-heating thermal buckling experiment of a four-edge-clamped thin plate is conducted for method assessment. Loading steps 1-24 are used for parameter inversion, whereas loading steps 25-30 are reserved for assessment at subsequent loading states along the same monotonic heating path. Single-parameter-fixed ablation and local sensitivity analyses further show that the two parameters have clearly coupled but non-collinear effects on the equilibrium residual, while ten independent inversions demonstrate good numerical repeatability. The results show that, compared with the constant-parameter model, the identified parameter functions reduce the average membrane weak-form equilibrium residual over the considered loading process by 20.95% and 35.53% in the inversion interval and the subsequent-loading interval, respectively. The center-point deflection and full-field out-of-plane displacement morphology obtained based on the inverted parameters can reproduce the main evolutionary characteristics of the experimental thermal buckling process. The proposed method therefore provides a full-field-observation-based route for structural-level equivalent thermoelastic-parameter inversion without stress labels or boundary-reaction data.