Automated acoustic method for counting and sizing farmed fish during transfer using DIDSON

Abstract

Counting and sizing large farmed fish such as tuna is often performed during their transfer from one net cage to another. Dual-frequency identification sonar (DIDSON) provides an automated fish counting and sizing tool. However, its counter and sizer are not suitable for measuring farmed fish because of net movements due to currents and subsequent frequent image breakups. This paper presents a fully automated acoustic method to count and size farmed fish during fish transfer by using DIDSON imaging. The background is subtracted from the image after being stabilized by an image phase-only correlation method. The segmentation of the fish is obtained by tracing the edges with a contour tracing method. To prevent recounting the same fish, a Kalman filter algorithm was designed and adapted to predict fish movements. Automated counting was performed by analyzing the spatiotemporal trajectory of the track. The separated fish images were searched for and body length was obtained by summing down the centerline segments from the head to the tail of the fish. The proposed system was verified using farmed yellowtail, Seriola quinqueradiata (mean total length 83.1 cm) to obtain a sizing error of mean total length within 2.4 cm.