BD-freshwater-fish: An image dataset from Bangladesh for AI-powered automatic fish species classification and detection toward smart aquaculture.

There are about 33,000 different species of fish and they are visually identified using variety of traits, i.e., size and shape of body, head's size and shape, skin pattern, fin pattern, mouth pattern, scale pattern, and eye pattern etc. In traditional manner, identifying these fish species is always difficult with necked eye. Identification and detection of fish species from images using deep learning and computer vision based techniques is challenging topic among researchers worldwide as an interesting problem. Automatic fish species classification and detection has practical importance for both smart aquaculture and fish industry. AI powered deep learning and computer vision based automatic fish species recognition and sorting system becoming significant factor for making aquaculture industry more productive and sustainable. However, the performance of machine learning classifier greatly depends on the size of image dataset and the quality of the images in the dataset. This article demonstrate , an image dataset contain 4389 images of 12 different species captured in natural environment using HD mobile camera from local fish market of Sylhet and Jessore district of Bangladesh. Twelve (12) different data classes are: Rohu (), Catla (), Mrigal (), Grass Carp (, Common Carp (), Mirror Carp ( var. specularis), Black Rohu (), Silver Carp ( Striped Catfish (), Nile Tilapia (), Long-whiskered Catfish (), Freshwater Shark () has been included in the dataset with a different number of images of different species. The dataset is hosted by Department of Computer Science and Engineering mutually with the help of the Department of Aquaculture, Sylhet Agricultural University, Sylhet, Bangladesh.