Common object detection algorithms have too many backgrounds, duplicated boxes, and a high miss ratio in detecting rotated objects, which limit their applications on the industrial site. To address these problems, this paper proposed an improved YOLOv5 to detect rotated objects. First, this paper used the K-means clustering algorithm to develop clustering analysis for trained datasets to confirm more proper anchor boxes to reduce the training time. Then this paper transferred the angle issue to a classification problem. This paper also learned angles in the original loss function combined with the circular smooth label (CSL) algorithm, thus avoiding the periodicity of angle regression. Last, this paper selected one from different detection results of an object to increase the accuracy of the detection results. The experiment showed that the proposed algorithm had a higher detection precision than other methods in the public dataset DOTA. When the proposed algorithm detected rotated objects in the dataset collected on the industrial site, its mAP reached 94.35%. This value was 8.20% higher than that of YOLOv5, satisfying the detection requirement on the industrial site.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.