MemoryMamba is a novel memory-augmented state space model designed for defect recognition in industrial settings. It integrates state space models with memory augmentation mechanisms to effectively capture dependencies and intricate defect characteristics. The model uses coarse- and fine-grained memory networks to retain and access critical defect information from previously trained samples. A fusion module integrates features extracted from these memory networks, enhancing the model's capability. Optimization strategies based on contrastive learning and mutual information maximization are also proposed for the memory networks. The model was evaluated across four industrial datasets with diverse defect types and complexities. MemoryMamba consistently outperformed other methods, demonstrating its capability to adapt to various defect recognition scenarios. The architecture of MemoryMamba includes a Mem-SSM Block that combines memory encoding and selective scanning to extract robust representations of the input. The block processes input features through a linear layer, memory encoding modules, and a fusion module to generate fused features. These features are then processed through a 2D Selective Scan module and normalized using layer normalization before being combined with the original input features through a residual connection. The model's performance was evaluated using metrics such as accuracy, precision, recall, and F1 score. Results showed that MemoryMamba achieved the highest scores in all metrics across the four datasets, particularly outperforming in challenging defect detection scenarios. The integration of coarse- and fine-grained memory encoding significantly enhances the model's ability to capture detailed contextual information, improving its effectiveness in complex visual pattern recognition tasks. The ablation study confirmed the importance of the memory networks and fusion module in the model's performance. The choice of similarity metric and memory size also significantly influenced the model's performance, highlighting the need for careful tuning in different scenarios. Overall, MemoryMamba demonstrates superior performance in defect recognition, outperforming existing technologies such as CNNs and Vision Transformers.MemoryMamba is a novel memory-augmented state space model designed for defect recognition in industrial settings. It integrates state space models with memory augmentation mechanisms to effectively capture dependencies and intricate defect characteristics. The model uses coarse- and fine-grained memory networks to retain and access critical defect information from previously trained samples. A fusion module integrates features extracted from these memory networks, enhancing the model's capability. Optimization strategies based on contrastive learning and mutual information maximization are also proposed for the memory networks. The model was evaluated across four industrial datasets with diverse defect types and complexities. MemoryMamba consistently outperformed other methods, demonstrating its capability to adapt to various defect recognition scenarios. The architecture of MemoryMamba includes a Mem-SSM Block that combines memory encoding and selective scanning to extract robust representations of the input. The block processes input features through a linear layer, memory encoding modules, and a fusion module to generate fused features. These features are then processed through a 2D Selective Scan module and normalized using layer normalization before being combined with the original input features through a residual connection. The model's performance was evaluated using metrics such as accuracy, precision, recall, and F1 score. Results showed that MemoryMamba achieved the highest scores in all metrics across the four datasets, particularly outperforming in challenging defect detection scenarios. The integration of coarse- and fine-grained memory encoding significantly enhances the model's ability to capture detailed contextual information, improving its effectiveness in complex visual pattern recognition tasks. The ablation study confirmed the importance of the memory networks and fusion module in the model's performance. The choice of similarity metric and memory size also significantly influenced the model's performance, highlighting the need for careful tuning in different scenarios. Overall, MemoryMamba demonstrates superior performance in defect recognition, outperforming existing technologies such as CNNs and Vision Transformers.