Abstract: In response to the shortcomings of traditional object detection frameworks in detecting weak infrared targets， such as low recall and large model parameters， a new lightweight infrared weak target detection algorithm， FCDN， is proposed with multiple attention mechanisms. Firstly， a lightweight network structure is proposed to reduce the loss of target information during the information flow transmission process. Secondly， a lightweight upsampling operator is used during the upsampling process to obtain a larger receptive field and improve the recall rate. Thirdly， detection heads is adopted with multiple attention mechanisms again to focus more on small target information from different levels for target regression and recognition. Finally， a new loss function is applied， which uses the maximum similarity bounding box metric to characterize the similarity between the real box and the predicted box. Experimental results show that on the infrared weak target detection dataset under complex backgrounds in comparison with the benchmark model YOLOv5， the accuracy reaches 99.4%， the recall rate increases by 5%， reaching 94.7%， the mAP50 increases by 4.3%， reaching 97.8%， and the parameter quantity is reduced by about 42% compared to original basic model. The model size is compressed to 8.66 MB. The detection results show that the proposed algorithm can meet the future detection needs of infrared weak targets， which has important value and reference significance for practical applications such as infrared tracking， early warning reconnaissance， missile guidance， etc.