Using computer vision to identify limpets from their shells: a case study using four species from the Baja California peninsula
Using computer vision to identify limpets from their shells: a case study using four species from the Baja California peninsula
复制标题
DOI：
10.3389/fmars.2023.1167818
复制
发表时间：
2023-07
影响因子：
3.7
通讯作者：
Jack D. Hollister;Xiaohao Cai;T. Horton;B. Price;K. M. Zarzyczny;Phillip B. Fenberg
中科院分区：
文献类型：
--
作者：
Jack D. Hollister;Xiaohao Cai;T. Horton;B. Price;K. M. Zarzyczny;Phillip B. Fenberg
帽贝的外壳形态在物种内部和物种之间可能是神秘的且高度可变的。因此，即使对于专家来说，视觉识别物种也可能很麻烦。在这里，我们展示了计算机视觉模型作为辅助识别新方法的能力。我们研究了计算机从背侧和腹侧贝壳的数字图像中区分来自下加利福尼亚半岛（墨西哥）的四种和两个属帽贝的能力。总体而言，在预测同一组图像时，模型的表现 (97.9%) 略好于专家 (97.5%)，并且速度快了 240 倍。此外，我们利用热图系统来验证模型是否专注于标本，并查看模型用于区分物种和属的标本上的哪些特征。然后，我们聘请了专门识别巴哈半岛物种的帽贝生态学家的专业知识，以评论热图是否确实关注每个物种/属的特定形态特征。他们证实，在他们看来，大多数热图似乎突出显示了对于区分群体具有重要形态学意义的区域和特征。我们的研究结果表明，计算机视觉的尖端技术在增强分类学家和生态学家使用的物种识别技术方面具有巨大的潜力。它不仅为传统方法提供了补充方法，而且还为更详细地探索帽贝生物学和生态学开辟了新途径。
The shell morphology of limpets can be cryptic and highly variable, within and between species. Therefore, the visual identification of species can be troublesome even for experts. Here, we demonstrate the capability of computer vision models as a new method to assist with identifications. We investigate the ability of computers to distinguish between four species and two genera of limpets from the Baja California peninsula (Mexico) from digital images of shells from both dorsal and ventral orientations. Overall, the models performed marginally better (97.9%) than experts (97.5%) when predicting the same set of images and did so 240x faster. Moreover, we utilised a heatmap system to both verify that models are focussing on the specimens and to view which features on the specimens the models used to distinguish between species and genera. We then enlisted the expertise of limpet ecologists specialised in identification of species from the Baja peninsula to comment on whether the heatmaps are indeed focusing on specific morphological features per species/genus. They confirm that in their opinion, the majority of the heatmaps appear to be highlighting areas and features of morphological importance for distinguishing between groups. Our findings reveal that the cutting-edge technology of computer vision holds tremendous potential in enhancing species identification techniques used by taxonomists and ecologists. Not only does it provide a complementary approach to traditional methods, but it also opens new avenues for exploring the biology and ecology of limpets in greater detail.