Abstract: UAV flying in complex battlefield environment, due to the similar shape and color of the enemy UAV, and the existing algorithms can not accurately identify and classify the UAV of the enemy, resulting in false detection or even mishandling attack. To solve this problem, a feature fusion algorithm based on the combination of the bottom visual features and high-level visual features was proposed to classify the UAV target objects. The algorithm first extracted the underlying visual features and high-level visual features of the target object by using visual feature descriptors and Sparse Auto-Encoder (SAE). Then, the principal component analysis (PAC) method was used to reduce the dimensionality of the global features. Finally, the global feature response was sent to the softmax regression model to complete the recognition and classification of the target object of the UAV. Experiments show that the new algorithm has higher accuracy and robustness compared with the traditional SAE algorithm and the traditional recognition algorithm based on the underlying visual features.