A joint research group from Tokyo University of Science, Kanazawa University, and Kanagawa University used deep learning to determine whether a crab is male or female based on images of its shell.
In Hokkaido, the fishing of crab crabs is strictly restricted, and in particular, the capture of female crabs for research purposes is completely prohibited unless permission is obtained.It is easier to catch male and female crabs if you can tell them apart by their appearance, which is also important to prevent overfishing of females.Males and females can be easily determined visually from the ventral side due to the different shapes of their reproductive organs, but they are often captured with their carapaces facing up, making it difficult to visually determine the sexes.However, fishermen say it is easy to tell by looking at the dorsal side.
Therefore, in this research, we used deep learning to determine the sex of the crab from images of its carapace and ventral side.Deep learning (explainable AI) was applied to photographic images of 60 males and 60 females captured in Oshamanbe, Hokkaido.The deep learning algorithms used are AlexNet, VGG-16, and ResNet-50 neural networks.
As a result, the correct answer rate for VGG-16 was the highest.A heat map* showing the affected areas as a basis for judgment appeared strongly near the shape of the genitals on the abdominal side.The male heatmap highlighted the bottom of the carapace, while the female heatmap highlighted the top of the carapace.Image recognition of the shell side based on deep learning algorithms enables more accurate gender identification than inspection by the human eye.
Whale crabs are an important aquatic resource, and the ability to distinguish between sexes using deep learning is an important finding for protecting them, and it can also be used to develop efficient livestock farming techniques.Furthermore, the deep learning algorithm resulting from this research can be applied to other species of crabs, and can contribute to the establishment of efficient methods for determining the sexes of crabs.
*Editor's note: An analysis tool that visualizes the strength and weakness of data using colors.
Paper information:[Scientific Reports] Gender identification of the horsehair crab, Erimacrus isenbeckii (Brandt, 1848), by image recognition with a deep neural network