A Large-Scale Dataset for Fish Segmentation and Classification

Assessing the quality of seafood both in retail and during packaging at the production side must be carried out minutely in order to avoid spoilage which causes severe human health problems and also economic loss. Since the illnesses and decay in seafood presents distinct symptoms in different species, primarily the classification of species is required. In this field, the inadequacy of the current laborious and slow traditional methods can be overcome with systems based on machine learning and image processing, which present fast and precise results. In order design such systems, practical and suitable datasets are required. However, most of the publicly available datasets are not fit for the mentioned purpose. These datasets either contain images taken underwater or consist of seafood which is generally not (widely) consumed. In this study, a practical and large dataset containing nine distinct seafood widely consumed in the Aegean Region of Turkey is formed. Additionally, comprehensive experiments based on different classification approaches are performed to analyze the usability of this collected dataset. Experimental results demonstrate very promising outcomes; therefore, this dataset will be made publicly available for further investigations in this research domain.