Refining CNN-based Heatmap Regression with Gradient-based Corner Points for Electrode Localization
Lin Wu
TL;DR
This work tackles the challenge of precisely localizing electrode poles in high-resolution X-ray images of lithium-ion batteries, where low contrast and multi-layer overlap hinder traditional methods. It proposes a joint framework that first uses gradient-based corner points (via OFAST) to identify a region of interest, then applies CNN-based heatmap regression (HRNet) to estimate pole coordinates within that ROI, and finally refines these estimates with a post-correction step that fuses CNN predictions with corner priors through a confidence-driven fusion. Key contributions include ROI estimation from corner points, HRNet-based pole regression within the ROI, and a corner-prior–driven post-correction mechanism, evaluated with NME, PCK, and PCS metrics showing notable improvements, especially with larger corner sets (e.g., N=512). The approach demonstrates that combining traditional pixel-gradient analysis with modern heatmap-based keypoint detection can enhance both accuracy and efficiency, with practical impact on manufacturing QA and safety-critical battery applications.
Abstract
We propose a method for detecting the electrode positions in lithium-ion batteries. The process begins by identifying the region of interest (ROI) in the battery's X-ray image through corner point detection. A convolutional neural network is then used to regress the pole positions within this ROI. Finally, the regressed positions are optimized and corrected using corner point priors, significantly mitigating the loss of localization accuracy caused by operations such as feature map down-sampling and padding during network training. Our findings show that combining traditional pixel gradient analysis with CNN-based heatmap regression for keypoint extraction enhances both accuracy and efficiency, resulting in significant performance improvements.