The
Phenotype Segmentation
pipeline utilizes advanced image-based segmentation techniques
to automate the extraction and analysis of phenotypic traits in
aquaculture species. By combining a binary level set method with
the Mumford-Shah functional, this tool accurately segments fish
bodies from background images, allowing precise measurements of
key traits such as size, shape, and growth characteristics.
Leveraging deep learning and mathematical optimization, the tool
ensures high-throughput, reliable, and consistent phenotyping,
significantly reducing manual labor and minimizing errors. This
automated phenotyping process provides breeders with accurate,
data-driven insights, enabling more informed decision-making in
breeding programs, thereby enhancing genetic selection and
improving breeding outcomes.
The ocean harbors abundant
biological resources, particularly aquatic species such as fish,
shrimp, sea cucumbers, and scallops. However, the underwater
environment is complex, and traditional image processing methods
struggle to detect objects that are distorted or have colors
similar to the surrounding environment. In this study, we
captured images of fish in their natural underwater habitat and
employed deep learning techniques to effectively delineate fish
boundaries for body region segmentation. Using the area ratio,
we developed a method to predict body weight, providing an
efficient, non-invasive, and high-throughput approach to
underwater feature measurement. Directly measuring weight in
underwater environments is nearly impossible; however, the area
ratio enables highly accurate biomass estimation, replacing
manual detection and measurement. This significantly enhances
the efficiency of biomass assessment.
