In complex industrial scenes, the 6D pose estimation of weakly textured objects is characterized by difficult feature extraction and a high cost of dataset production. These challenges make it difficult to ensure the accuracy and reliability of the 6D pose estimation. To address these issues, a weakly textured object pose estimation method driven by decoupling analysis and algorithm fusion strategy is proposed in this paper. In this approach, the 6D pose estimation is decoupled. Firstly, a synthetic dataset is created using rendering and sampling techniques, and the relevant initial pose information is recorded. Then, a CBAM-CDAE network is proposed, which incorporates the Convolutional Block Attention Module (CBAM) into the Convolutional Denoising Autoencoder (CDAE). This network suppresses irrelevant features and enhances the expressive power of the network. Additionally, the hierarchical network structure and sliding window operation from the Swin Transformer network are introduced to improve the overall performance. The trained CBAM-CDAE network is used to generate potential vector codebooks corresponding to the template samples. The trained M-ST instance segmentation network is employed to obtain object bounding box and mask information, which are utilized to calculate the initial pose estimation. Extensive experiments are conducted on the T-LESS dataset to validate the proposed method. The proposed method accomplishes highly accurate while maintaining detection speed, the method achieves improved accuracy and reliability. The experimental results provide strong evidence of the method's performance and highlight its potential for practical applications.