To address the localization and the detection accuracy problems of the Faster R-CNN target-detection algorithm, a movable attention (MA) model that can be embedded in the algorithm and trained end-to-end is designed. First, to obtain more accurate spatial location information, MA uses two adaptive maximum pooling operations to aggregate features based on the horizontal and the vertical directions of the input feature and generates two independent directional-sensing feature maps. Second, to prevent model overfitting, the sigmoid activation function is used to increase network nonlinearity. Finally, to fully exploit the obtained spatial location information, the two nonlinear and input feature maps are multiplied successively to enhance the representational ability of the latter. The experimental results show that the improved Faster R-CNN target-detection algorithm based on MA can effectively enhance the network’s ability to locate the target of interest, as well as considerably improve the average detection accuracy.