Images captured by UAVs can be disturbed by various factors, such as weather conditions, equipment limitations, etc., resulting in image quality degradation. However, traditional image filtering methods do not perform well in dealing with images with noise or artifacts. For this reason, this study proposes a UAV image fusion filtering method based on full convolutional twin network. After constructing the full convolutional twin network structure, the model is trained. The trained model is applied to the UAV image fusion filtering process. In the experiment, the feasibility of this method was verified from both subjective and objective testing perspectives. It was found that after applying this method, the quality of drone images was significantly improved. The average gradient of the fused filtered image was 10.58, the spatial frequency was 21.612, the information entropy was 4.90, and the structural similarity was 0.571. The visual information fidelity can reach 0.258.