Infrared target detection and recognition based on deep learning algorithm is an important field in academic research. In the background of high precision in infrared target detection and recognition and lightweight algorithms, and based on YOLOv5n network model, at first, the C3 module in the network is replaced with dilated residual convolution (DWR) to achieve lightweight of the network and enable the network to flexibly extract features of different scales. And then, in response to the low resolution and blurred details of infrared images, AF-FPN is used to replace the original FPN structure to improve the ability of multi-scale infrared image target recognition. At last, the iRMB attention mechanism is inserted into the detection layer, making the model lightweight while still maintaining detection accuracy similar to the original YOLOv5n. Experimental results show that the network value of the improved model has increased by 0.8% compared to the original YOLOv5n network, the model volume has reduced by 17%, and lightweight model without affecting the accuracy of model detection is achieved, which meets the requirements of small and lightweight size and is suitable for deployed on embedded devices.