A Dataset of Ecological Images for Automatic species-identification of Butterflies

The automatic species identification of butterflies can provide strong technical assistance to do ecological environment monitoring, biodiversity protection, pest control, and reduce the strong dependence on expert’s recognition to butterfly species. However, all the available datasets used for the studies of butterfly auto-identification are based on specimen images, and lack of ecological butterfly images. Therefore we set up a dataset of butterfly images from the ecological environment in China and the corresponding annotation files in PASCAL VOC 2007 format. The dataset contains 721 butterfly images, covering 94 butterfly species. Each image corresponds to an annotation file, which contains the size of the image, the taxonomic information and the location of the butterfly in the image. This dataset could support researches on automatic recognition, object detection, and insect taxonomy.

蝴蝶物种自动识别技术可为生态环境监测、生物多样性保护以及害虫防治提供强有力的技术支撑，同时可大幅降低对蝴蝶物种专家鉴定的依赖程度。然而，当前用于蝴蝶自动识别研究的现有数据集均基于标本图像，缺乏自然生态场景下的蝴蝶影像。为此，本团队构建了一套采集自中国野外生态环境的蝴蝶图像数据集，并配套提供PASCAL VOC 2007格式的标注文件。该数据集共包含721张蝴蝶图像，涵盖94个蝴蝶物种。每张图像均对应一份标注文件，其中包含图像尺寸、蝴蝶的分类学信息以及其在图像中的位置坐标。本数据集可支撑蝴蝶自动识别、目标检测以及昆虫分类学等相关研究工作。