Artificial Intelligence (AI) bird counting is a rapidly growing field that is revolutionizing the way we monitor bird populations. Bird counting is an essential tool for wildlife conservation and management, as it provides vital information about population trends, habitat preferences, and migration patterns. AI bird counting has the potential to make bird counting more accurate, efficient, and cost-effective. Let’s talk about the application, limitation and recently development of AI bird counting in Hong Kong.
Application:
AI bird counting has several applications in the field of wildlife conservation and management. One of the significant benefits of AI bird counting is its ability to detect and identify bird species accurately and quickly. This technology can provide real-time data about bird populations, which can be used to monitor and manage bird habitats. Furthermore, AI bird counting can automate the bird counting process, which can save time and resources.
Point count is a common method used by ornithologists to estimate bird populations. This technique involves observing birds at a fixed point for a set amount of time. However, this method is time-consuming and often needs more than one experienced observer. Especially, in the migration bottlenecks where bird migrations occur intensively in specific periods, the observer has to detect and count birds and keep records. Furthermore, the success of point counts is much obliged to the human census and affected by the observer’s ability and experience, environmental and topographic variables and birds’ detectability. Detection of birds in bad weather conditions such as intense cloud cover, rainy and foggy weather, and low light intensity might not be possible. As such, researchers are continually developing new techniques and technologies, such as AI bird counting, to improve the accuracy and efficiency of bird population monitoring.
Limitations:
Despite the potential benefits of AI bird counting, there are also some limitations to this technology. One of the significant limitations is its accuracy in detecting and identifying bird species in complex environments. AI bird counting systems may have difficulty distinguishing between similar-looking bird species, which can lead to errors in population estimates. Additionally, AI bird counting systems may not be effective in detecting birds that are hidden or camouflaged, such as birds in dense vegetation.
Major misdetections occur due to very small bird sizes. As the distance to birds increases, such birds often correspond to very small details (i.e., several pixels) for even the human eye to differentiate. Small object detection is also an open-problem in many other datasets. Other main reasons for the misdetections are intra-species variation, unusual bird poses, occlusions by other birds and plants, cast shadows, and background clutter.
Development:
AI bird counting technology is rapidly evolving, and researchers are continually working to improve its accuracy and efficiency. One of the significant developments in AI bird counting is the use of machine learning algorithms, which can learn from large datasets and improve their accuracy over time. We partner with Hong Kong Bird Watching Society (HKBWS), World Wide Fund for Nature Hong Kong (WWF-Hong Kong) and APlus Digital Academy to explore AI bird detection and counting in Hong Kong. We deployed fixed and mobile cameras in multiple locations to evaluate AI bird detection performance.
We find the AI detection confidence score will significantly affect the bird counting performance. If we require AI detection with a higher confidence score, there will be more false negatives and the system will underestimate the bird population. Vice versa, if we accept AI detection with a lower confidence score, there will be more false positives and the software will overestimate the number of birds. We are looking for the most effective confidence score values in our future research.
Additionally, we are developing new sensors that can capture high-quality data in challenging environments. Another development is the use of citizen science initiatives to gather data for AI bird counting. Citizen scientists can take photos of birds and upload them to databases, which can then be used to train AI algorithms. This not only helps to improve the accuracy of the algorithms but also engages the public in conservation efforts.
In conclusion, AI bird counting is a promising technology that has the potential to transform the way we monitor and manage bird populations. While there are limitations to this technology, ongoing developments in AI algorithms and sensing technology are likely to improve its accuracy and effectiveness in the future. As such, AI bird counting is a field to watch for conservationists and researchers who are committed to protecting the world’s bird species.
Reference:
- Automated Bird Counting with Deep Learning for Regional Bird Distribution Mapping
- AI Wildfire Detection Accuracy: Precision, Recall, mAP
The original article was published on andrecheung.wordpress.com
