A DEEP LEARNING MODEL OF DEFECT DETECTION METHOD AND SYSTEM FOR PLASTIC INJECTION MOLDING PRODUCTS

The present invention discloses a defect detection method with machine learning for plastic injection molding products, comprising the steps of: collecting image dataset (100) from product samples with data augmentation by adjusting environment and camera settings (100a), and by a computational algorithm (100c); training a model (102) using the collected image dataset; evaluating the model (104); and detecting defects (106) by using the evaluated model. The training of model (102) comprises the steps of: applying transfer learning (102b) to the deep learning neural network; applying model structure of YOLOv5 (102a) to the deep learning neural network; passing the image to a single focus layer (102c) in the beginning layers of the deep learning neural network; using cross-stage partial (CSP) networks (102e) to duplicate and merge the feature map when concatenating the image; and adopting path aggregation network (PAN) (102d). The method of training model (102) in the present invention improves the performance and accuracy of bounding box prediction for localization. The method of training model (102) further provides for the detection of at least one defect and at least one type of defect on at least one product of the same kind.