In order to extract the information characterization of landmark spectrum out of TM image, the functional relation of recognition of landmark spectrum compositions was constructed and the general principle of landmark recognition out of TM image was proposed. Genetic algorithm(GA) was applied to optimize the coefficients of functional relation of landmark spectrum composition. It is very easy to recognize each landmark from TM image, according to the correspondent relation between each landmark and the ranges of function values. The analysis of cases shows that this recognition model of landmark spectrum composition has the properties of simplicity, practicability and differentiation.