Surface defect detection of automobile engine cylinder block based on artificial intelligence

In view of the problems such as sample scarcity, small target missing detection and poor adaptability to complex working conditions in the detection of automotive engine cylinder block surface defects, this study proposed an intelligent detection method based on improved Faster-RCNN and K-Means clustering. By integrating multi-scale feature pyramid network and dynamic anchor frame optimization mechanism, a defect detection framework for industrial scenarios is constructed. Firstly, LabelMe annotation tool is used to construct defect data set, and geometric transformation and pixel adjustment strategy are combined to alleviate the problem of data scarcity. Secondly, K-Means algorithm is introduced to perform adaptive cluster analysis of anchor frame size to optimize the generation mechanism of candidate regions. Finally, a two-stage cooperative detection network is designed to realize efficient feature reuse through parameter sharing between RPN and target detection network. Experiments show that after training with A1 enhanced data set, the average accuracy of the model reaches 97.18%, which is 18.56% higher than that of the baseline method. With the help of Adam optimizer and 30-iteration training strategy, the system still maintains 86.36% recall rate under complex illumination and oil pollution interference, which verifies the robustness and practicability of the algorithm in industrial detection scenarios.