In the field of concrete structure health monitoring, accurately and swiftly identifying damage characteristics stands as a pivotal task. To enhance the accuracy and efficiency of concrete damage identification, this research proposes an improved Self-Organizing Map algorithm based on visual sensing. By optimizing feature extraction and representation methods, introducing novel learning strategies, and incorporating spatial attention mechanisms, the model becomes adept at capturing and identifying concrete damage features more effectively. Additionally, employing stochastic gradient descent as an optimization algorithm enhances the model training efficiency. Experimental results showcase that the model exhibits a detection time of merely 0.8 seconds, while demonstrating outstanding fitting and clustering performance, achieving an actual accuracy of 98.2%. Compared to methods based on digital image monitoring and deep learning detection, it shows an improvement of 12.7% and 31.8%, respectively. The proposed enhanced model significantly augments the accuracy and efficiency of concrete damage identification, providing an effective solution for the health monitoring of concrete structures, particularly in scenarios requiring large-scale and real-time monitoring. This advancement elevates the practicality and convenience of concrete damage detection, propelling progress in the field of building safety.