<![CDATA[Graphical Abstract Highlight Research 1. Provide insights into the development, methodologies and significant advancements in fish classification research utilizing CNNs. 2. The review discusses the progression of CNN architectures, beginning with earlier models such as AlexNet and advancing to more sophisticated frameworks like ResNet and GoogLeNet. 3. Examines the extensive use of benchmark datasets across many studies and discusses collaborative research efforts that refine models and enhance reproducibility. 4. Guide future research by underscoring best practices, acknowledging less explored areas and encouraging interdisciplinary approaches to monitoring fish biodiversity and managing aquatic ecosystems through deep learning. Abstract Recent advancements in deep learning have substantially improved the classification of fish species. These innovations present a contemporary and dependable alternative to conventional methodologies, such as image processing and manual identification. Implementing Convolutional Neural Networks (CNNs) has significantly enhanced accuracy, flexibility and scalability within aquatic ecosystems. This comprehensive review evaluates 81 scholarly articles published from January 2015 to October 2025. The analyses were guided by the VICO framework, supported by defined inclusion and exclusion criteria, data extraction and synthesis and the PRISMA process to ensure systematic selection of relevant studies. It aims to provide insights into the development, methodologies, and significant advancements in fish classification research utilizing CNNs. The review discusses the progression of CNN architectures, beginning with earlier models such as AlexNet and advancing to more sophisticated frameworks like ResNet and GoogLeNet, as well as transformer and hybrid CNN model such as the Vision Transformer (ViT) and the Convolutional Vision Transformer (ConViT). It highlights the adoption rates, training performance and contexts of use. Additionally, it examines the extensive use of benchmark datasets across many studies and discusses collaborative research efforts that refine models and enhance reproducibility. A thorough comparison of classification accuracy, dataset composition and trends in model choice offers a clearer picture of the current impact of deep learning in this area. Furthermore, this review identifies crucial challenges, including the lack of data for rare species, issues related to low-resolution image recognition and the need for standardization in model evaluation. The insights offered aim to guide future research by underscoring best practices, acknowledging less explored areas and encouraging interdisciplinary approaches to monitoring fish biodiversity and managing aquatic ecosystems through deep learning.]]>
