Hybrid Supervised Model for Defect Detection Based on CrossFormer Theory: Enhancing Accuracy and Efficiency in Industrial Manufacturing

Quality management is an important aspect of industrial manufacturing. Plant managers strive to ensure that workers accurately recognize defects. Deep learning is widely used in manufacturing research, particularly for image inspection, and offers advantages such as accurate identification of defect sites and correct classification. We propose a hybrid supervised model based on CrossFormer theory to improve the accuracy and efficiency of defect detection in industrial manufacturing processes. CrossFormer combines the local perception capability of convolutional neural networks with the global contextual understanding of the transformer model and is suitable for addressing long-distance dependency problems in image data. A hybrid supervised strategy fully utilizes labeled and unlabeled data to enhance the learning capability of the model and reduce dependence on large amounts of labeled data. Experimental results on the KSDD2 and STEEL datasets confirm that our model outperforms existing methods on defect detection tasks, particularly when using a small amount of labeled data. In addition, analysis of the performance of the model on different datasets demonstrates the advantages of CrossFormer in capturing complex image features, which provides an effective solution for industrial defect detection and offers new perspectives for the design and training of deep learning models.