<p>To address wood product paint defect detection challenges, we propose MSIEGNet - a Multiscale Information-Enhanced Global Network that enhances defect saliency in complex backgrounds. Firstly, an innovative multi-scale edge feature fusion mechanism is integrated into the backbone network. This architecture effectively extracts defect edge features while enhancing detail sensitivity through multi-scale perception. Secondly, a feature channel attention mechanism is implemented during upsampling to preserve structural details. This approach enables fine-grained feature weight allocation across feature maps. Thirdly, a global dynamic perception network is developed to strengthen feature extraction. This architecture propagates context-rich multi-level features across detection scales, maintaining robust performance in complex environments and small target detection. Finally, an enhanced shared re-parameterizable convolutional detection head is introduced to address accuracy degradation caused by traditional heads in processing multi-level features. We compiled a wood finish defect dataset comprising four characteristic defect types: bubbles, cracks, holes, and scratches. Extensive comparative experiments demonstrate the network’s superior performance, achieving 97.4% accuracy and 90.8% mAP₅₀ in wood defect detection, surpassing state-of-the-art methods. The network also exhibits strong generalization capabilities, showing 68.5% mAP₅₀ on public wood defect datasets, confirming its broad applicability in complex scenarios.</p>