Ensemble Neural Network Based Fish Species Identification for Emerging Aquaculture Application

Precision fish farming is a novel idea in the aquaculture research and development industry that integrates modern technology and pre-data processing methodologies inorder to allow data-driven decisions in precise fish farming. The objectives are to use contactless technologies to automate the process of monitoring fish, improve infrastructure, and marine environment. Individualized treatment, biomass measurement, and fish status assessment all depend on the ability to identify specific fish of same species within group frame. Recent studies have demonstrated that specific fish body patterns are used for individual species identification, but there is currently no system for automation of this process. We here have introduced a method to entirely automate identification of the fish species entirely based on their characteristics. Therefore, we adopted Ensemble of MobileNetV3 and NASNets, also known as (Ens_MobV3+NasNet), to excusively automate the process of classification of the species present in the Deep Vision trawl camera system images. Speeded-up robust features, which are based on realistic simulations of the Deep Vision images, are used to increase speed. Fish4Knowledge dataset is used for study, and it is discovered that the suggested Ens_MobV3+NasNet achieves 98.0% accuracy, 95% precision, 94.3% recall, and 91% F1-score.