Abstract: Aiming at the problems of low segmentation accuracy and slow detection speed in the segmentation and detection of infrared images of invisible methane gas, based on the YOLOv11-seg algorithm model, the C2PSA is replaced with the C2PSAiEMA network to enhance the ability to extract feature information from methane gas infrared images; the neck network is replaced with the Slim-neck structure, which reduces the amount of computational data of the model and accelerates the detection speed of the model; a dataset is created by collecting methane gas infrared images using a laboratory infrared camera, and the dataset is expanded through image enhancement methods such as contrast enhancement and noise addition to improve the quality of image segmentation and enhance the segmentation performance of the improved YOLOv11-seg. Tests conducted on the methane gas infrared image dataset show that the improved YOLOv11-seg achieves an average precision (AP) of 0.90 for detection boxes and an AP of 0.89 for mask segmentation, which are significantly improved compared with the original model, effectively enhancing the instance segmentation accuracy of methane gas infrared images. Moreover, the model's detection speed (FPS) is increased from 91.64 of the original version to 108.43, which effectively improves the real-time performance of the model.