Multimodal data fusion-enhanced surface defect detection of body reinforcement components under uncertain illumination conditions

The surface quality of automotive body components directly affects the appearance and safety of automobiles. In industrial defect detection scenarios, differences in imaging conditions and changes in illumination environments make existing methods difficult to meet requirements of robust and efficient detections. To address this challenge, this study proposes a domain-adaptive multimodal object detection algorithm for automotive body components under real industrial conditions with uncertain illumination. Based on the classical You-Only-Look-Once-Version-8 (YOLOv8) algorithm, a dual-branch architecture is constructed to extract features from visible-light and infrared images, respectively, achieving information complementarity via an intermediate fusion strategy. To further improve the model’s adaptability to domain changes under uncertain illumination, a hybrid domain perturbation augmentation strategy integrated with a Robust-Source-Domain-Random-Transformation (RDT) module is introduced to expand the feature space, and combined with an improved adaptive instance normalization module, which effectively suppresses domain-related feature interference, thereby significantly improving the model's generalization performance in unknown illumination environments without the need for labeled target domain data. Experimental results show that the proposed algorithm outperforms the original YOLOv8 and other mainstream algorithms on the automobile body reinforcement dataset under various illumination uncertainty scenarios, with the mean average precision (mAP@0.5) improved by approximately 10%, verifying the effectiveness and robustness of the algorithm under complex illumination conditions. In addition, through a typical case analysis, it is further confirmed that the method can improve the detection accuracy of surface defects in automobile body structure reinforcements, providing a new technical approach for industrial quality traceability and risk prevention and control.