Deep Learning Architecture for Real-Time Polyps Detection on Endoscopic Video

To detect polyps on images of endoscopic examinations of the intestine we suggested algorithms based on the YOLOv5 architecture. We conducted tests using an open set of images K vasir-SEG, and obtained the best result for the proposed neural network algorithm DLPD
YL
, for which the value of the AP@[0.25.0.75] metric 92.5% exceeded the known analogues by 5% and more. However its computational complexity exceeded three times the ColonSegN et algorithm known from the literature is a disadvantage. The studies are the basis for building a video stream analysis module in a real endoscopic system.