Researchers have created a deep learning model that surpasses human fishermen in accurately determining the gender of horsehair crabs.
Deep learning models help tell apart male and female horsehair crabs. The heatmaps highlight the specific parts of the shell that the proposed artificial intelligence algorithms “focus on” when classifying male and female crabs. These models can vastly outperform humans in this task and may find applications in crab aquaculture and responsible fishing. Image Credit: Shin-ichi Satake from Tokyo University of Science, Japan.
In Japan, as winter descends upon the northern regions, fishermen eagerly embark on the quest to harvest one of the season's most eagerly awaited delicacies—the horsehair crab. Referred to locally as “kegani” and scientifically identified as Erimacrus isenbeckii, this sought-after crustacean holds a special place in culinary preferences throughout the country.
To safeguard the horsehair crab population from the perils of overfishing, both the Japanese government and prefectural authorities have implemented various restrictions on their capture. Notably, in Hokkaido, where kegani is particularly abundant, the consumption of females is strictly forbidden.
In adherence to these regulations, seasoned fishermen have acquired the skill of distinguishing between male and female crabs through visual examination. While discerning the gender by inspecting the underside (abdomen) of the crabs is a relatively straightforward task, doing so by examining their shell side proves to be considerably more challenging.
Regrettably, when crabs are captured and settled on board a ship, they tend to position themselves with their shell side facing upward. Individually picking them up and flipping them to ascertain their sex becomes a time-consuming endeavor.
Might artificial intelligence (AI) prove adept at yet another task? Addressing this query, a recent study led by a Japanese research team, including Professor Shin-ichi Satake from Tokyo University of Science (TUS), delved into the realm of deep learning.
Published in the esteemed journal Scientific Reports, their latest paper boasts co-authorship by Associate Professor Yoshitaka Ueki and Professor Ken Takeuchi from TUS, as well as Assistant Professor Kenji Toyota and Professor Tsuyoshi Ohira from Kanagawa University.
Employing three deep convolutional neural networks based on established image classification algorithms—AlexNet, VGG-16, and ResNet-50—the researchers trained and tested these models using 120 images of horsehair crabs from Hokkaido, evenly split between males and females.
A notable attribute of these models lies in their "explainable AI" nature. In simpler terms, when presented with a crab image, it becomes possible to discern the specific regions crucial for the algorithm's classification decision.
This feature unveils subtle distinctions between males and females, potentially aiding manual classification efforts.
Encouragingly, the test results demonstrated promising levels of accuracy and performance metrics.
Even though gender classification was virtually impossible by human visual inspection on the shell side, the proposed deep learning models enabled male and female classification with high precision, achieving an F-1 measure of approximately 95% and similarly high accuracy values.
Shin-ichi Satake, Professor, Tokyo University of Science
This implies that the AI approach significantly outperformed human capabilities, delivering consistent and reliable classification. Interestingly, upon examining the heatmaps representing the regions of focus for the models during classification, the research team identified noteworthy distinctions between the sexes.
Notably, the heatmap exhibited heightened activity near the genitalia shape on the abdomen side.
When classifying males, the algorithms concentrated on the lower section of the carapace. In contrast, when dealing with females, the algorithms directed their attention to the upper part of the carapace.
This revelation not only holds implications for the development of future AI-based sex classification models for crabs but also offers insights into how seasoned fishermen distinguish between male and female crabs, even when examining their shell side.
Considering the significant stress that being captured imposes on crabs, the ability to swiftly identify females without flipping them before release could play a crucial role in averting health or reproductive issues for these crustaceans. Consequently, deep learning stands as a potentially vital tool for augmenting conservation and farming efforts.
The fact that deep learning can discriminate male and female crabs is an important finding not only for the conservation of these important marine resources but also for the development of efficient aquaculture techniques.
Shin-ichi Satake, Professor, Tokyo University of Science
Significantly, the integration of AI classification techniques directly on ships has the potential to minimize manual labor and enhance the cost-effectiveness of crab fishing. Furthermore, the models proposed in this study can be retrained and adapted for gender classification in other crab species, including the blue crab or the Dungeness crab.
In essence, this research exemplifies how AI can be ingeniously employed to not only streamline human tasks but also contribute directly to the positive impact on conservation, responsible fishing practices, and the sustainability of crab aquaculture.
Journal Reference:
Ueki, Y., et al. (2023) Gender identification of the horsehair crab, Erimacrus isenbeckii (Brandt, 1848), by image recognition with a deep neural network. Scientific Reports. doi.org/10.1038/s41598-023-46606-x.