Automated AI Approach for Noninvasive Shrimp Length and Weight Estimation Using Underwater Imaging and Feeding Induction

Abstract

We aim to develop an objective method for measuring shrimp length in traditional white shrimp farming using images from an automated underwater camera system, reducing reliance on subjective assessment. Traditional manual methods often suffer from subjectivity and inaccuracies, leading to inefficient feed management strategies. Computer vision techniques were used for object detection and image preprocessing. A deep learning network was used to classify completeness and assess the length and weight of shrimps. We also analyzed the correlation between the body length and weight of shrimps. The dataset consists of 8401 images categorized as measurable (3112) and visible (5289). An accuracy of 95.0% was obtained with an average error rate of 8.4%, highlighting the effectiveness of the proposed method. The weight estimation exhibited an average error rate of 22%. This system can optimize feed management and enhance aquaculture sustainability, reducing both resource waste and operational costs. Furthermore, we utilized underwater cameras as sensing devices, combined with specially designed feeding platforms and observation materials, to enable real-time image-based monitoring and noninvasive shrimp measurement.