3D imaging and machine learning system aims to improve weight estimates for frozen skipjack tuna

3D imaging and machine learning system could reduce labor in seafood processing

Researchers in Japan have developed a 3D imaging and machine learning system that could improve non-contact weight estimation of frozen skipjack tuna, with potential applications in seafood processing and fisheries resource management.

In a proof-of-concept study, the system’s weight estimates agreed more closely with the fish’s measured weight classes than classifications made by experienced market graders, according to researchers at the University of Tsukuba.

The system uses a three-dimensional time-of-flight camera to scan frozen skipjack tuna on a conveyor belt. The camera measures distance using reflected infrared light and captures the surface of each fish as dense 3D point-cloud data.

Researchers developed the approach in part to address challenges associated with measuring frozen skipjack tuna. Skipjack caught in distant-water fisheries are typically frozen onboard, and frost on their surfaces strongly reflects light, making accurate shape measurements difficult with conventional two-dimensional imaging methods.

Using the 3D scans, researchers reconstructed the contours of frost-covered fish and extracted three measurements: body width, fork length and body height.

Body width was particularly important for improving the accuracy of the weight estimates, according to the study. Researchers said the measurement has been difficult to obtain using conventional imaging methods.

The team then combined the 3D measurements with machine learning analysis to generate non-contact estimates of fish body weight.

The resulting estimates agreed more closely with the fish’s measured weight classes than classifications made by experienced market graders, according to the researchers.

The technology is not yet fully automated. In the proof-of-concept study, collection of the 3D point-cloud data was automated, but researchers manually extracted the body measurements used for the weight estimates.

The researchers said further development toward automated operation could help reduce labor demands at fisheries and seafood-processing facilities while supporting more efficient and consistent management of marine resources.

Read the full study here.