The invention discloses a micro expression recognition method based on sparse projection learning. The method comprises the steps of: Step 1, collecting a micro expression sample, extracting LBP features P, Q, R of three orthogonal planes of the micro expression, and defining C, D, E as feature optimization variables of three orthogonal planes of XY, XT, YT respectively; constructing an optimization model; 2, setting the initial value and maximum value of iterative counting variable t and n and initializing the regularization parameter kappa, kappa max, and the scale parameter ruho; step 3, initializing the expression (shown in the description), calculating C, and updating T1 and Kappa; If the expression (shown in the description) converges or n> nmax, proceeding to step 4; step 4, initializing the expression (shown in the description), calculating D, and updating T2 and Kappa; and if the expression (shown in the description) converges or n>nmax, proceeding to step 5; step 5, initializing the expression (shown in the description), calculating E, and updating T3 and Kappa; and if the expression (shown in the description) converges or n> nmax, proceeding to step 6; step 6, makingt=t+1, if t <= tmax, returning to step 3, otherwise, outputting C, D, E; step 7, optimizing the LBP features of the three orthogonal plane by optimizing the variables C, D and E to obtain a new fusionfeature Ftest, and predicting the emotion category of the test sample through the trained SVM classifier for the fusion feature Ftest.