Improving Fisheries Management through Deep learning based Automated fish counting

The process of quantifying fish populations holds significant value in the realm of fisheries management, as it enables a precise evaluation of population sizes and facilitates comprehension of the current state of the fish stock. Nonetheless, the process of manually counting fish is demanding in terms of labor, time, and susceptible to inaccuracies. In order to tackle this issue, there have been advancements in automated fish counting techniques utilising computer vision and deep learning algorithms. The present study introduces an automated fish counting system based on deep learning, which employs a Convolutional neural network (CNN) to identify and enumerate the fish present in an image. The system under consideration has been assessed on a dataset comprising of underwater images that encompass diverse fish species. The findings of the evaluation indicate that the system attains a mean absolute error of 0.5 fish per image. The system under consideration exhibits a high degree of precision in quantifying fish populations across diverse settings, thereby presenting a viable avenue for enhancing fisheries governance. The system exhibits the ability to identify distinct fish species, rendering it appropriate for employment in fisheries management contexts, including stock evaluation and species categorization. In summary, the present study presents empirical support for the efficacy of the deep learning-based automated fish counting system, which has the capability to accurately quantify the quantity of fish in an image. This technology holds significant potential for enhancing fisheries management practices.