A Holistic Approach to IGBT Board Surface Fractal Object Detection Based on the Multi-Head Model

In industrial visual inspection, foreign matters are mostly fractal objects. Detailed detection of objects is difficult but necessary because better decision-making relies on more detailed and comprehensive information. This presents a challenge for applications. To solve this problem, we proposed holistic approach to object based multi-head model. We the IWS (Information Watch Study) module provide enhancement learning capabilities It increases dimension can perform detection. order realize portability module, it be easily quickly deployed existing advanced model achieve end-to-end FGI (Fine-Grained Information) Head, which used extract feature vectors from original base WST (Watch Study Tactic) Learner information processing adaptive class cluster centers. Using MRD (Multi-task Result Determination) strategy combine classification results results, final derived. experiment, were mounted three different models YOLO series. The experimental show that YOLO+IWS has good meet needs inspection. Moreover, ability objects, than other 11 competing methods. designed new evaluation index an adjustment mechanism weights make judgments balanced learning. Not only that, applied form brand system.