 MSINET is a lightweight detection network for infrared target detection. It uses asymmetric convolutions to reduce the number of parameters and improve detection performance. Additionally, a down sampling module called DPP reduces information loss from pooling down sampling. A feature fusion structure called LRFPN is used to shorten the information transmission path and reduce noise in the process of feature fusion. Coordinate attention, CA, is integrated into the LRFPN to provide more expressive feature information. The network is compared against other state-of-the-art methods on the FLIR on-board infrared image dataset and demonstrates its superior detection performance. This article was authored by Jiminyu, Shunli, Shangguajo, and others.