生物多样性（鸟类）特征数据集
收藏资源简介：
生物多样性是地球生命共同体的血脉和根基。习近平总书记在《生物多样性公约》第十五次缔约方大会上强调，要加强生物多样性保护，推动生物多样性主流化。持续研究高空遥感、环境DNA、AI识别等技术手段，加快观测手段多样化、观测过程智能化与观测数据集约化进程，加强生物多样性智慧管理系统建设，用翔实的数据和生动的案例，充分展现江苏贯彻落实习近平生态文明思想的实践成效。本产品按照《生物多样性观测技术导则鸟类-HJ710.4-2014》标准为指导，采集了以无锡市为代表的江南地区较为常见的约 700种鸟类上万张不同视角的图片数据，经过对数据的标定特别是矢量化处理后建立了鸟类学习数据库(样本数据集)，依托面向GPU和DPU相结合的算力支撑，运用公司自主研发的图神经网络系统进行连续无监督学习，构建了覆盖700多种鸟类的高质量特征数据集（特征数据集）。特征数据集能够支撑的识别准确率超90%。同时系统采用不定期增量学习机制确保鸟类数据库（样本数据集）定期扩容，推动数据产品定期更新，确保产品与时俱进，满足数据服务识别需求，最后形成用户上传鸟类图片的识别数据集（成果数据集）。
Biodiversity is the lifeblood and foundation of the Earth's community of life. General Secretary Xi Jinping emphasized at the 15th Conference of the Parties to the Convention on Biological Diversity (COP15) that efforts should be strengthened to protect biodiversity and promote its mainstreaming. It is necessary to continuously research technical means such as high-altitude remote sensing, environmental DNA (eDNA), and AI recognition, accelerate the diversification of monitoring methods, the intelligentization of monitoring processes, and the intensification of monitoring data, strengthen the construction of smart biodiversity management systems, and fully demonstrate the practical achievements of Jiangsu Province in implementing Xi Jinping Thought on Ecological Civilization with detailed data and vivid cases. This product is developed under the guidance of *Guidelines for Biodiversity Monitoring Techniques - Birds - HJ710.4-2014*. It has collected tens of thousands of images of approximately 700 common bird species in the Jiangnan region represented by Wuxi, taken from different perspectives. After data calibration, especially vectorization processing, a bird learning database (sample dataset) was established. Relying on computing power supported by a combination of Graphics Processing Units (GPUs) and Data Processing Units (DPUs), the company's independently developed Graph Neural Network (GNN) system was used for continuous unsupervised learning, constructing a high-quality feature dataset (feature dataset) covering more than 700 bird species. The recognition accuracy supported by this feature dataset exceeds 90%. Additionally, the system adopts an irregular incremental learning mechanism to ensure regular expansion of the bird database (sample dataset) and periodic updates of the data product, keeping the product up-to-date to meet the recognition needs of data services, and finally forming the recognition dataset (result dataset) for users to upload bird images.
