Laparoscopic surgery is a minimally invasive surgical procedure that is commonly used with several advantages over traditional open surgery for a variety of medical procedures. However, in some cases, gauze is left in the abdominal cavity and may lead to serious inflammatory reactions caused by foreign bodies and intestinal obstruction by intraluminal migration, which will have catastrophic repercussions. Since laparoscopes use lenses to enter and illuminate the view inside the patient’s body, the images and videos could be analyzed by using Computer Vision Technology (CVT) to identify and track surgical instruments. This study focuses on reducing medical errors by enhancing gauze detection during laparoscopic surgery using three deep learning models, which are U-Net, U-Net++, and Trans-U-Net models, designed for image segmentation tasks. The performance of these models is evaluated using four metrics to compare the effectiveness of these three semantic recognition techniques. The results indicate that U-Net++ and Trans-U-Net have better overall performance and higher precision compared to U-Net, which is likely due to their improved feature extraction capabilities and ability to capture more complex patterns within surgical imagery. In addition, the Trans-U-Net model has the best performance according to the comprehensive evaluations. The findings of this research are expected to assist in reducing the mental workload of surgeons in monitoring and memorizing the location of gauze and enabling the surgeons to focus more on the surgical procedures, thus reducing the risk of adverse events and enhancing patient safety in laparoscopic surgery within clinical practice.