Chip needle scratch precise detection method based on semi-supervised self-evolution

Abstract

This invention discloses a semi-supervised self-evolutionary method for accurate chip pin mark detection. A basic chip pin mark detection network model is constructed, trained using supervised pin mark samples with sufficient source domain, and stable teacher model snapshots from different training stages are selected to construct a joint teacher model group. Several teacher models in the joint teacher model group perform collaborative inference on several unlabeled samples in the target domain, dividing the pin mark targets in the unlabeled samples into a high-reliability sample set and a drift sample set. Based on the high-reliability sample set and the drift sample set, the basic chip pin mark detection network model is trained again under supervision to obtain an updated student model. After obtaining the updated student model, the model parameters of the joint teacher model group are iteratively updated using a differentiated EMA method until the optimal student model is obtained. The optimal student model is then used to detect pin marks on the target domain chip image to obtain the pin mark detection results.

Images