Steel surfaces are highly esteemed due to their exceptional strength, durability, and versatility, rendering them extensively used in many industries such as automobiles, marine, electronics, aerospace, and construction. However, defects on steel surfaces can cause severe risks and financial damage if these defects remain undetected. Numerous research works have been conducted to address this issue, drawing upon CNN as a primary source. However, the current algorithms using Convolutional Neural Networks (CNNs) also suffer from low accuracy problems, leaving plenty of room for further improvement. This research introduces a novel methodology for classifying steel surface defects with high accuracy. The proposed approach uses a newly developed Convolutional Neural Network (CNN) architecture of 81 layers. Results achieved demonstrate the effectiveness and robustness of the proposed model in classifying steel surface defects compared to existing approaches in the field.