Research on Quality Control of Gigabit Network Transformers Based on Improved Transformer Neural Networks

The structure of Gigabit network transformers is complex, with significant scale variations in defects such as pin skew and missing magnetic cores. This leads to low efficiency and high false detection rates in traditional manual inspection. To address this challenge and meet the industrial production line's demand for high speed and high precision, this study optimizes the YOLOv5s model with targeted improvements. First, to enhance detection speed, the original SPP module in the network is replaced with the more efficient simSPPF module, accelerating feature fusion and inference. Second, to improve the model's feature extraction capability for complex defects and its understanding of global context, some Conv and C3 modules in the backbone network are replaced with Transformer modules, thereby effectively enhancing the recognition accuracy for small targets and defects with variable morphologies. Experiments on the Gigabit network transformer dataset show that the optimized model significantly outperforms the baseline model in both detection accuracy and inference speed, achieving a balance between precision and efficiency. This provides an effective solution for the intelligent quality inspection of related electronic components.