Transmission Defects Localization Network: Towards Wrong Assembly in the Transmission Assembly Process

Transmission valve component is an important part of automobile manufacture. The success the assembly transmission directly related to driving safety vehicles. While localizing defects particularly in component. As image processing problem, real-time images are adopted determine whether correct or wrong. these have small, severe reflection, and sparse properties, which increases difficulty detection. Therefore, this paper proposes a localization network based on Siamese for improving performance components. In our model, we establish similarity evaluation with designed multi-scale features fusion approach. Furthermore, order reduce intra-class spacing by similar samples action, improved binary cross entropy focal loss function discovered feature re-processing. Finally, experimental results real-world dataset indicate that proposed approach outperforms other compared methods.