ESTUDIO DE UN MÉTODO DE DETECCIÓN BASADO EN DEEP LEARNING DE PUNTOS CLAVE DE MEDICIONES NO RESTRINGIDAS DEL CAMARÓN BLANCO DEL PACÍFICO EN EL AGUA (STUDY ON DEEP LEARNING-BASED DETECTION METHOD of KEY POINTS of UNCONSTRAINED PACIFIC WHITE SHRIMP IN WATER)

In the aquaculture industry, aquatic seedlings are commonly known as “chips.” The phenotypic data of shrimp seedlings, which can reflect their growth, are important reference indexes for breeding purposes. In traditional shrimp culture, key points on the shrimp body are mainly determined through artificial means, and parameters are manually measured to obtain breeding-related phenotypic data. However, this manual approach is not only time-consuming and laborious but can also lead to human errors. Moreover, shrimp are highly sensitive to handling, which can easily cause physical harm, spread diseases, and lead to water contamination during manual measurements. To improve the speed and accuracy of shrimp phenotypic data collection, this study proposed a novel approach: a deep learning-based network for the automatic detection of shrimp key points. The proposed method minimized physical contact, prevented potential damage, and enabled the collection of more comprehensive phenotypic data. With Pacific white shrimp as the study object, deep learning was applied to detect key points of the shrimp body. The key point detection network could detect 23 key points of the top view and 10 key points of the side view, which could provide data support for subsequent measurements of phenotypic parameters and for modeling shrimp body growth. Results show that, compared with traditional methods, the proposed approach has advanced performance, speed, and robustness in terms of accuracy and efficiency. The accuracy of key points of the top view reaches 97.57%, with an average test time of 137.3 ms, and the side view verification set a key point detection accuracy of 98.61%, with an average test time of 49.3 ms. This study realizes the unconstrained measurement of shrimp with water, which can accurately and quickly obtain the key points of shrimp and meet the needs of multiple phenotypic data measurement. Keywords: phenotypic data, deep learning, Pacific white shrimp, key point detection, unconstrained measurement