Yolov5 polyp detection algorithm based on multi-branch convolution RFB module

Colorectal cancer remains a formidable global health concern, underscoring the pivotal role of timely detection and intervention in addressing colorectal polyps. This paper unveils the "YOLOv5 Optimized Polyp Detection Algorithm Based on the Multi-Branch RFB Module," offering innovative solutions to overcome existing challenges in manual interpretation and resource disparities. Key Innovations: Our approach introduces a groundbreaking adaptive Gamma transformation in image preprocessing, enhancing contrast and significantly elevating the algorithm's ability to identify colorectal polyps. This transformation contributes to a remarkable improvement in detection accuracy. Furthermore, the integration of the Multi-Branch RFB Convolution Module into the YOLOv5 model is a pioneering addition. This module strategically considers the interplay between receptive field size and eccentricity, resulting in a substantial increase in discriminative power and, consequently, a notable enhancement in detection accuracy. Method Flow Overview: Our algorithm modifies the YOLOv5 framework by incorporating three RBF multi-branch convolutional modules, strategically enhancing feature extraction at deep layers for precise and rapid object detection. Results and Discussion: Comprehensive evaluations on diverse datasets, including CVC-ColonDB and Kvasir-seg, underscore the algorithm's superior performance. Precision, Recall, F1-Score, and mAP@0.5 scores on CVC-ColonDB outperform alternative methods, affirming the effectiveness of our proposed algorithm. Conclusion: This study pioneers advancements in computer-aided colorectal polyp detection, seamlessly integrating adaptive image preprocessing and the multi-branch RFB module. The algorithm demonstrates heightened accuracy, efficiency, and adaptability across datasets, with real-time capabilities and an expanded receptive field contributing to improved diagnostic accuracy. Our findings suggest a promising solution for enhancing early colorectal cancer diagnosis, streamlining clinical workflows, and ushering in a new era of precision in medical imaging.