A comprehensive annotated image dataset for real-time fish detection in pond settings

Abstract

Fish is a vital food source, providing essential nutrients and playing a crucial role in global food security. In Tamil Nadu, fish is particularly important, contributing significantly to the local diet, economy, and livelihoods of numerous fishing communities along its extensive coastline. Our objective is to develop an efficient fish detection system in pond environments to contribute to small-scale industries by facilitating fish classification, growth monitoring, and other essential aquaculture practices through a non-invasive approach. This dataset comprises of Orange Chromide fish species (Etroplus maculatus) captured under several computer vision challenges, including occlusion, turbid water conditions, high fish density per frame, and varying lighting conditions. We present annotated images derived from underwater video recordings in Retteri Pond, Kolathur, Chennai, Tamil Nadu (GPS coordinates: Lat 13.132725, Long 80.212555). The footage was captured using an underwater camera without artificial lighting, at depths less than 4 m to maintain naturalness in underwater images. The recorded videos were converted to 2D images, which were manually annotated using the Roboflow tool. This carefully annotated dataset, offers a valuable resource for aquaculture engineers, marine biologists, and experts in computer vision, and deep learning, aiding in the creation of automated detection tools for underwater imagery.