Algorithm for Detecting Small Targets Based on Upgraded YOLOv5s

Abstract

Small targets are discovered despite the challenges posed by the limited number of features, low resolution, difficulty in extracting discriminative characteristics, and ease of interference from external influences. A modest target detection technique based on enhanced YOLOv5s is suggested in this research. First, a test storey dedicated to tiny objects is appended to the characteristic fusion part of the algorithm to help the model efficiently acquire features with smaller sensory fields; second, to help the model find features more correctly, a cooperative attention method module is integrated into the characteristic extraction section to eliminate trait redundancy without sacrificing feature information; finally, the Bi-FPN network construction is combined to help the model to locate features more accurately. With the Bi-FPN construction, the characteristic combination process is modified to strengthen the examining capability of small objectives with a characteristic combination. The experimental results demonstrate that the enhanced algorithm increases average detection accuracy over the baseline YOLOv5s method by 6.3%, while also increasing small target detection accuracy and decreasing false and missed detection situations.