SVM Classification for Sealing Ring Counting in Occlusion and Multi-Pose Scenarios

With the continuous development of industrial automation, efficient and accurate counting of small objects such as sealing rings has become a significant challenge in the production process. This paper proposes a visual counting algorithm for sealing rings based on Support Vector Machine (SVM) classification, aimed at addressing the packaging counting problem of sealing rings under multi-pose and occlusion conditions. The algorithm segments sealing ring images using connected domain analysis and extracts four features: connected domain area, corner points, convexity, and hole count. It trains the model with 32,000 labeled and categorized sealing ring images, classifying them into five categories corresponding to the number of sealing rings (1 to 5) within a connected domain. In the classification phase, the SVM classifier accurately classifies each connected domain, thus enabling the counting of sealing rings in the entire image. Experimental results show that the SVM-based counting algorithm significantly outperforms traditional methods such as template matching, watershed segmentation, and connected domain area methods in terms of accuracy in scenes with occlusion and multi-pose sealing rings. Particularly, the F1 score for detecting a connected domain containing a single sealing ring under occlusion interference reaches 96.5%. Moreover, the SVM classification algorithm exhibits a significantly lower false detection rate compared to the other three traditional algorithms, demonstrating superior robustness and accuracy.