There has been a constant need for the classification of fish species for a better understanding of the underwater ecological balance. Identifying the characteristics of different fish species plays a significant role in knowing the insights of marine ecology and is a great deal to many fisheries and industries. Manually classifying fish species is time-consuming and requires high sampling efforts. The behaviour of fishes can be well understood using an automated system that accurately classifies various fish species effectively. The classification of underwater images has difficulties like background noise interruption, image disruption, lower quality of images, occlusion. The proposed model lights up on the assortment of fish species using Alexnet. The knowledge of the previously trained model is given to the alexnet for improving the system. The performance of our improved model is demonstrated with real-world data from a research organization called Kaggle. CNN has used several layers trained for precise identification of the distinct features of a species and classify them accordingly. This paper ensures increased accuracy than the existing systems.