Infrared Monitoring Dataset of Wildlife in Dense Forests

In today's ecological scientific research, observing and monitoring wildlife activities in their natural environment is a significant challenge, mainly due to the complexity of dense forest environments and the elusive nature of animals. Traditional monitoring methods such as direct observation and setting traps are often constrained by terrain, weather changes, and limited human resources, resulting in inefficient and inaccurate data collection. The Infrared Monitoring Dataset of Wildlife in Dense Forests addresses these challenges by capturing animals' natural behaviors without disturbance using infrared-triggered cameras, solving the monitoring difficulties of hidden and nocturnal animals. Data is captured by infrared cameras placed in tropical and temperate dense forests, which automatically take photos upon detecting movement. To ensure data accuracy and consistency, multiple rounds of annotations are conducted by an expert team with biological backgrounds, followed by rigorous consistency checks and reviews. Additionally, data preprocessing includes image quality enhancement and image cropping to ensure the needed image quality. Data is organized by date and location, stored in JPG format for easy retrieval and analysis. The annotation accuracy of this dataset is over 98%, employing innovative multi-label classification methods and data augmentation techniques to significantly enhance recognition accuracy. Compared to similar datasets, it offers wider species coverage and more comprehensive ecological information, applicable to practical issues such as species conservation and ecological impact assessment, significantly enhancing model predictive ability and analytical performance. This dataset is not only unique for containing images of rare species but also has strong scalability and generalizability due to its applicability across multiple ecological environments.