Aiming at the problems of low contrast, blurred edge and easy to be disturbed by noise of infrared targets in complex background, an infrared target detection algorithm based on deep learning is proposed. At first, the image is preprocessed by image smoothing and Gamma transform. And then, the local patch network (LPNet) with global concern is used to detect the small target by considering the global and local characteristics of the infrared small target image. From a global perspective, it is a supervised attention module trained by a small target diffusion graph to suppress most background pixels unrelated to small target features. From a local perspective, local patches are separated from global features and share the same convolutional weights with each other in the patch network. By utilizing both global and local characteristics, the data-driven framework of the proposed algorithm incorporated with multi-scale features is used to small target detection. The experimental results on SIRST dataset show that the proposed algorithm can effectively realize infrared target detection under complex background, which verifies the effectiveness and robustness of the proposed algorithm.