Abstract: In object recognition based on feature fusion, the key issues are the correlation of selected features and the strategy of feature fusion.For military target recognition with rapidness and high accuracy, we motivate the object recognition algorithm of time-frequency domain feature integration based on covariant-matrix, the new method with lower dimensions and more de-correlated among features compared to traditional method with single feature.The proposed method comprises three steps:firstly, extracting several sets of features with the same pattern and establishing each feature vectors;then, constructing co-variable matrix with these vectors;and finally, doing pattern recognition using improved normalized Fisher Linear Discrminant Analysis.At last, we demonstrated our improved algorithm on coil-20 image library, Colombia University, and obtained 99.54% average recognition accuracy.Our experimental results validated the wonderful robustness of our method.