Aiming at the single-view angle through-wall radar in the process of human behavior recognition there is a viewpoint blind area, resulting in the inability to accurately identify the movement state of the human target behind the wall. In this paper, based on the multi-view through-wall radar system, a multi-view through-wall human behavior recognition method based on attention fusion is proposed. Firstly, the features of the radar distance image are enhanced by the method of clutter suppression and compensated gain to, and the dataset is constructed in this way. Secondly, we use the improved DenseNet to extract features from the distance image data of the two viewpoints, and finally, we use the attention mechanism to calculate the weight information of the radar echo signals of the two viewpoints, and then we fuse the two action feature information according to this weight data, and input the classifier to get the behavior recognition results. The experimental results show that the behavior recognition method proposed in this paper achieves better results compared with the existing behavior recognition methods.