在对东北虎豹国家公园进行的生态学研究中，陷阱相机被广泛用作野生动物调查和监测的探测器。通过高灵敏度的触发器和相机拍摄图像，陷阱相机可以在野外的自然环境中获得大量图像数据，这些数据经过统计和汇总后可供研究人员进一步使用。传统情况下，这些图像数据是通过人工辨认和分类的，耗费大量人力且时间效率低下。随着计算机技术的发展，许多基于深度学习的方法被用于野生动物物种识别，并取得了令人满意的效果。然而，这些运用深度学习的方法大多适用于检测简单场景中的大型哺乳动物。当研究对象为野生环境下的鸟类时，由于其生活在复杂的丛林环境中，具有保护色，体型小等特点，导致深度学习应用在这种情况下的效果不够理想，人工智能方法在这些鸟类的识别的方法存在一定的困难与挑战。

本论文对东北虎豹国家公园的监测中获得的相机陷阱拍摄图像进行处理，构建了包括9000多张野生鸟类图像的数据集，其中包含15个常见物种，且鸟类均处于野外的复杂丛林环境中。为了得出最适用于复杂丛林环境下的鸟类检测方法，本研究挑选了八种最新的目标检测方法对该数据集进行检测，通过比较不同检测方法的平均精确率(Average  Precision，AP)来评估出有效方法，实现更为精确的鸟类物种的识别，便于研究人员进行下一步的生态学研究。

Trap cameras are widely used as detectors for wildlife surveys and monitoring in ecological studies conducted on the Northeast China Tiger and Leopard National Park. By using highly sensitive triggers, and cameras to capture images, trap cameras can obtain a large amount of image data in the natural environment in the field, which can be counted and aggregated for further use by researchers. Traditionally, these image data are identified and classified manually, which is labor-intensive and time-inefficient. With the development of computer technology, many deep learning-based methods have been used for wildlife species identification with satisfactory results. However, most of these methods using deep learning are suitable for detecting large mammals in simple scenarios. When the research object is birds in the wild environment, because they live in a complex forest environment with protective coloration and small size, resulting in less than ideal results of deep learning applications in this case, there are certain difficulties and challenges in the methods of artificial intelligence methods for the identification of these birds.

In this thesis, camera trap images obtained from monitoring in the Northeast China Tiger and Leopard National Park were processed to construct a dataset including more than 9000 images of wild birds, including 15 common species, and the birds were all in a complex forest environment in the wild. In order to derive the most suitable bird detection methods for complex forest environments, eight state-of-the-art target detection methods were selected to detect this dataset in this study. By comparing the Average Precision (AP) of the different detection methods to evaluate the effective methods, a more accurate identification of bird species can be achieved for researchers to carry out the next ecological study.