<p>This paper considers the identification and estimation of a high-dimensional matrix factor model with unknown breaks in both row and column factor loadings. We first develop the sequential method for estimating the break dates when the number of breaks is known. The key idea focuses on estimating the common break dates in factor loadings, which can be equivalently represented as the break dates within the second moments of the vectorized pseudo-factor matrix estimation. We can show that the distance between the proposed estimators and the true break dates is stochastically bounded under certain conditions. Secondly, we put forward a cross-validation approach utilizing an order-preserved sample-splitting strategy to determine the number of breaks in matrix factor models. We establish the consistency of our estimator for the number of changes under some mild conditions. Monte Carlo simulation results indicate the desired performance of the proposed methods in finite samples. A real data analysis is performed for illustration. Our method is implemented in the R package <Emphasis FontCategory="NonProportional">MSECV</Emphasis>.</p>