General-purpose subgraph mining algorithms are seldom used in real-world applications due to the high complexity of the mining process mostly based on isomorphism tests and countless expansion possibilities during the search. Efficient algorithms have been proposed for special cases for which there exist polynomial-time algorithms for graph isomorphism such as, for example, plane graphs, geometric graphs, and outerplanar graphs under the block-and-bridge preserving constraint. In this thesis, we proposed a new mining algorithm dedicated to find frequent patterns in spatio-temporal grids : GriMA. Use a regular grids allow our algorithm to reduce the complexity of isomorphism tests. Two applications are proposed to experimentally evaluate our algorithm : image classification for 2D grid mining and cellular automata prediction for 2D+t grid mining.