Brief

Researchers have introduced PBD5K, a new large-scale benchmark designed for power battery detection (PBD) using industrial X-ray images. This benchmark addresses the critical need for quality inspection in electric vehicle batteries, where internal defects can lead to safety risks. PBD5K includes 5,000 images across nine battery types, featuring fine-grained annotations and eight types of visual interference to simulate real-world conditions. To tackle the challenges of dense plate endpoints, low contrast, and imaging artifacts, the team also developed MDCNeXt, a novel model that integrates multi-dimensional structural clues and employs state space modules for enhanced discrimination and density-aware refinement. AI