Research on Infrared Dim and Small Target Recognition Algorithm based on Improved YOLO v5s

In response to the demand for high-precision infrared small target detection in complex backgrounds, a target intelligent recognition algorithm based on YOLO v5s is proposed. This algorithm introduces the SKAttention module into the backbone network of the YOLO v5s model to enhance its ability to extract features of targets at different scales. Additionally, the SimAM module is incorporated into the connection layer to help the model focus more effectively on the target region, without increasing the number of model parameters. Moreover, SIOU is used in place of EIOU to achieve more accurate target localization. Various image processing algorithms are employed to preprocess the infrared SIRST dataset, augmenting the number of images and completing model training and comparative experiments. Experimental results demonstrate that the improved YOLO v5s model outperforms the original YOLO v5s model by a 2.6% increase in accuracy, and exhibits superior target recognition performance compared to competing algorithms, effectively identifying infrared small targets in diverse background scenarios.