MV-DEIM: a multi-view information fusion method for zinc electrolysis contact anomaly detection

Electrode plate contact abnormalities are one of the key factors affecting current efficiency in the zinc electrolysis process. Recent studies on infrared image-based detection of electrode plate contact abnormalities have demonstrated promising performance under conventional scenarios. However, in real industrial environments, the perspective projection effect leads to poor image quality for plates at the edge of the field of view. The resulting scale reduction and loss of local details pose a major bottleneck to enhancing detection accuracy. In addition, the acid mist formed by the evaporation of the electrolyte will weaken the effective features of the infrared image, further increasing the detection difficulty. To alleviate the limitations of single-view detection in edge regions, this paper proposes a multi-view information fusion detection method, termed MV-DEIM. The method first extracts shallow features from different viewpoints using two independent backbone networks. Then, a multi-view feature fusion module is introduced to realize efficient interaction and fusion of multi-scale features across different views. A multi-scale synergistic parallel fusion module is further constructed to enhance the expression of the fused features. Finally, a decoder is used to generate the detection results. Experiments on real-world zinc electrolysis data confirm the effectiveness of the proposed method, which achieves 87.4% mAP50, 78.0% mAP75, and 62.5% mAP.