Visual steel surface detection plays a crucial role in enhancing the quality of steel sheet manufacturing. However conventional manual inspection-based strategies are subjective and does not comply with increasing quality standards, necessitating the deployment of more efficient automated tools leveraging advanced computer vision techniques. Recent advancements in machine learning models, particularly in Automated Visual Inspection (AVI) methods, have shown promise in detecting and classifying various surface defects using images. However, these classification models encounter challenges due to the limited availability of well-labeled datasets that encompass a wide range of surface defects. In this work, we examine the potential use of autoencoder, a type of self-supervised neural network, along with deployment scheme of decision tree to recognize various types of surface defects that occur during the manufacturing of hot rolled steel sheets. The proposed hierarchical tree structured autoencoder-based multi-classifier model achieves an overall recognition accuracy of 92.33% in distinguishing the various surface defect types in the images of steel sheet by analyzing the reconstruction loss of the images of the same. The experimental results based on the publicly accessible dataset in Kaggle repository demonstrate the effectiveness of the proposed model in comparison with other classification techniques, paving the way for developing autoencoder-based automated tools for similar surface defect recognition in various engineering materials.