Fish Species Detection and Recognition Using MobileNet v2 Architecture: A Transfer Learning Approach

Abstract

The classification of fish is essential for aquatic research and the preservation of biodiversity. This article suggests a fish detection model combining transfer and deep learning methods. Our method extracts hierarchical features from fish photos using the MobileNetV2 architecture, which has been trained on the Image dataset. We obtain precise fish species classification by adjusting the pre-trained model and adding extra dense layers. Performance criteria, including accuracy, precision, recall, and F1-score, are used to measure the model's efficacy, and it is trained and tested using a dataset of fish photos. The results show the dependability and durability of our fish detection model, with a test set accuracy of 99.94%. Additionally, a thorough evaluation of the model's precision, recall, and F1score highlights its capability to categorize fish species while validating its performance accurately. The proposed model has a lot of promise for use in various applications, such as ecological study, fish population monitoring, and conservation initiatives, which will further our knowledge of and help protect aquatic environments.