A Robust Recognition Method for Automotive Manufacturing Based on Deep Neural Networks

Abstract

The method of object recognition based on deep learning has a wide range of applications in various fields. For the visual detection of components in automobile manufacturing, the object classification network based on deep learning has achieved good results. However, environmental factors, such as camera shaking, camera rotation, illumination and so on, etc., may cause the detection accuracy of the object classification network to decrease. By analyzing industrial data, this paper proposes a robust deep learning-based recognition method for automotive manufacturing. Through data set enhancement and model selection, robust detection performance and higher detection accuracy are achieved even in the harsh environments of industrial production line. Experimental results show that while ensuring real-time performance, this method has better recognition performance on automotive components. More than 2% improvement is achieved in industrial environment with camera shaking and rotation, compared with traditional classification networks.