Anomaly Detection with Random Forest for Flexible Integrated Circuit Packaging Substrates

Due to the complex manufacturing process of Flexible Integrated Circuit Packaging Substrates (FICS), existing detection methods are difficult to detect all of defects. In this paper we propose an improved Speeded-Up Robust Features (SURF) and random forest (RF) algorithm for selective extraction of anomalous parts from the original image. Firstly, we improved the SURF algorithm by applying the radius and intensity of feature points to feature training, thereby increasing the information density of feature data. Next, in order to further improve the detection accuracy, we improved the random forest algorithm using data preprocessing methods. Finally, our method has significantly improved accuracy and reduced training time.