Optimization of Mask R-CNN based Auto Parts Defect Detection System with Image Enhancement

The recycling and reuse of car parts aligns with current energy conservation and environmental protection trends. Reused car parts may exhibit scratches and dents, necessitating their detection before being reintroduced to the market. This study introduces an image measurement system that integrates artificial intelligence with automatic focusing control and high-frequency structured light for surface defect detection on car bodies. The practical test reveals that recycled car body parts often contain residual stains, leading to reduced detection accuracy and increased false detection rates. Additionally, the system demonstrated poor resistance to interference from fingerprints and a high false recognition rate for dent detection. To enhance detection accuracy and improve the system's resistance to interference, an optimization method based on high-frequency structured light and neural networks is proposed. This method enhances the structural characteristics of high-frequency structured light through image enhancement techniques. By optimizing the neural network structure and dataset, dent recognition accuracy by the neural network model is improved. Comparative statistics of the detection results between our method and original method indicate that when detecting car body shells with fingerprints, the precision rate of our method increased by 71%, the recall rate increased by 11%, and accuracy increased by 32%.