The invention discloses a method for optimizing maximum MI waveforms of the cognition radar based on an IGA-NP algorithm. The method comprises steps of determining the cognition radar, wherein target signals exist in a detection range of the cognition radar, carrying out calculation to obtain target signal spectrum variance, clutter spectrum variance and a noise power spectrum and setting a to-be-optimized waveform power spectrum; determining an energy constraint condition of the to-be-optimized waveforms and carrying out problem modeling; setting iteration times and carrying out integar algebras of non-linearity programming optimization; setting the number of chromosomes, the crossover probability and the mutation probability; calculating an initial population Code; carrying out initialization: making h represent the hth iteration, wherein h belongs to {1, 2,..., G}, and G represents the set biggest evolution algebra and calculating the iterated k-opt searching population shown in the description and the optimal chromosome after the hth iteration; and finishing the iteration until the h=G, and using the corresponding optimal chromosome when the iteration is stopped as the global optimal chromosome, wherein the global optimal chromosome is the to-be-optimized waveform power spectrum.