<p>Underwater crack classification is a crucial task in the maintenance and repair of submerged structures, such as pipelines, bridges, and offshore platforms. Finding and classifying cracks correctly prevent catastrophic failure and ensure the longevity of these structures. In the past few years, techniques like image processing and deep learning algorithms have been suggested to automate classifying the cracks on different structures such as buildings, pavements, and bridges. But there is still a scope to automate the process of underwater crack detection. presence of water can cause cracks to appear differently than they would in air, which can make it more challenging to accurately identify and classify them. The goal of these methods is to make it easier and more accurate to find and classify cracks while reducing the need for human intervention. This paper proposed improved EfficientNet-based lightweight crack classification model and a novel image dataset for underwater crack detection. Model performance evaluated by the performance metrics which gives 99% accurate results with 99%, 99%, and 99% precision, recall, and F1-score respectively. Experimental results shown that proposed method gives better result compared to other state of arts methods.</p>