Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Automated Size Measurement and Weight Estimation of Body-Curved Grass Carp Based on Computer Vision

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan

Citation:  2021 ASABE Annual International Virtual Meeting  2100605.(doi:10.13031/aim.202100605)
Authors:   Congcong Lee, Jianping Li, Songming Zhu
Keywords:   Grass carp; Body-Curved; Size; Weight; Computer vision

Abstract. Automated size measurement and weight estimation is vitally important in sorting, grading grass carp and other aquaculture products. Traditionally, automated size measurement is to obtain morphological feature information by means of machine vision, relying on some specific and hand-designed rules or patterns. However, encountering body-curved fish, traditional machine vision method will expose its shortcomings of low robustness and inflexibility, resulting in the increase of its measurement error. In this study, grass carp, a major cultured fish in China, was selected as the research object. Size measurement, including total length (TL), eye-focal length (EFL) and body width (BW), were obtained by measure the distance among relevant key points, which is located by a Deep Learning algorithm called YOLOV5. Despite the fish curved its body, the average error rate in measurement of eye-focal length (EFL) and body width (BW), was 2.11%, 2.06% respectively with maximum error rate of 5.65%, 6.30% individually. Through statistical analysis and multivariate nonlinear regression, weight(W) can be expressed with EFL and BW, W= 0.0003597*(EFL^1.21254)*(BW^1.68798) (R² = 0.965). The average error rate and root mean square error were 5.49% and 0.52 g while maximum error rate and worst-case prediction error were 17.06% and 1.78 g. In addition, the present study confirmed that the proposed size measurement and weight estimation algorithm can process at approximately 72,000 grass carps per hour.

(Download PDF)    (Export to EndNotes)