With the rapid development of industrial automation, large-scale production of automated pipelines has become a trend. In the production process of industrial products, the quality detection of industrial products is an essential means to avoid economic losses and ensure personal safety. Improving industrial product quality detection accuracy and speed has been an important research topic in this field. In recent years, one-stage and two-stage deep learning object detection methods have been applied in defect detection. However, problems such as significant changes in object scale, high similarity, and the balance of speed and accuracy in industrial detection scenarios result in poor performance of common object detection algorithms. To address these issues, this paper proposes a Glass Bottle Bottom Mold Sequence Recognition Data Set and designs a novel dual collaborative paths feature network (DCP- Net) for industrial defect detection. The DCP-Net is designed to extract object features with varied scales from two branches, which can effectively focus on multi-scale industrial product defect features. Moreover, this paper presents the feature filter to obtain fine-grained information about objects, significantly improving similar objects' classification. In the experiments of the open data set of steel defects and the point sequence data set of glass bottle bottom mold collected in this paper, this method achieves higher accuracy than other methods.