In a recent paper published in Scientific Reports, researchers introduced an innovative system for detecting and repelling wild birds using deep learning and laser technology. This system aims to identify various wild birds in real time and project a laser beam to scare them away.
Images of the (a) laser rotation mechanism and (b) proposed automatic wild bird repellent system. Image Credit: https://link.springer.com/article/10.1038/s41598-024-66920-2
Background
Wild birds can harm agriculture and poultry farming through crop damage, disease transmission, and feed contamination.
Methods like sound, visual, chemical, and physical techniques have been suggested to keep birds away, but many are ineffective, expensive, and harmful. Laser technology is widely used for bird repulsion because most wild birds have eyes sensitive to green light, which irritates them and forces them to leave the area.
However, existing systems are either manually operated or have limited scanning ranges, which reduces their effectiveness. They often struggle to detect birds accurately or adapt their scanning paths based on bird movements. Therefore, a more intelligent and automatic system is needed.
About the Research
In this study, the authors proposed a wild bird-repellent system with three components: a detection unit, a computing unit, and a laser control unit. The detection unit uses a camera to capture images of the field, which are sent to the computing unit. This unit runs a bird detection model trained with a mask region-based convolutional neural network (mask R-CNN) to find and locate wild birds.
The computing unit sends these coordinates to the laser control unit, which uses a relay and motor control chip to operate a 400-mW green laser. This laser is directed to project a beam around detected birds to repel them. The system operates continuously and automatically as long as birds are detected.
The wild bird detection model was trained using a dataset of images collected from various farms in Taiwan. This dataset included images of mynas, sparrows, Chinese bulbuls, doves, turtles, and pigeons, Taiwan's predominant wild bird species. The images were annotated with polygonal masks to label the birds.
The mask R-CNN model was optimized for detecting small pixel targets and trained using this dataset. The model’s performance was evaluated using precision and recall metrics under different settings of minimum detection confidence and intersection over union (IoU) thresholds.
The system was tested at a duck farm in Yunlin, Taiwan, where most wild birds were spotted. Two cameras recorded the experimental process. The system operated alternately for one hour on and one hour off. Four experiments were conducted over four consecutive days, using different laser scanning strategies to compare their effectiveness.
The number of wild birds was estimated using the same wild bird detection model on images captured by the recording cameras. The system's effectiveness was measured using the daily and hourly bird repulsion rates.
Research Findings
The outcomes showed that the model achieved the highest precision of 0.865 with a minimum detection confidence of 0.95 and an IoU threshold of 0.1. The model better detected wild birds with larger pixels than smaller ones.
Visual examples demonstrated the model's ability to detect wild birds accurately and precisely against complex backgrounds, with a low rate of falsely detecting ducks as wild birds.
Field experiments indicated that the automatic wild bird repellent system effectively reduced the number of wild birds on the farm. The highest daily bird repulsion rate was 40.3%, and the highest hourly was 97.1%. Various factors influenced system performance, including ambient light, weather conditions, laser scanning strategy, and wild bird habituation.
The system performed better in the morning than in the afternoon when the ambient light was too bright for the birds to see the laser. It also performed better in dry weather than in rainy weather when the birds were less active.
The laser scanning strategy that targeted the region with the most detected wild birds was the most effective. However, performance gradually decreased as the birds became accustomed to the laser scanning path. Statistical analysis showed that the presence of the system significantly reduced the number of wild birds, confirming its effectiveness.
This novel system can protect crops and poultry from wild bird damage and disease transmission, improving food security and animal health. Additionally, it can prevent bird strikes at airports, reducing the risk of aviation accidents and human casualties.
The system can also control the population and distribution of invasive or nuisance bird species, mitigating their negative impacts on the ecosystem and biodiversity.
Conclusion
In summary, the novel system proved effective for detecting and repelling wild birds in real-time, offering several advantages over conventional methods. It was effective, low-cost, environmentally friendly, and adaptable.
This system could be applied in various scenarios where wild bird repulsion is needed. Future improvements could address limitations like weather, bird habituation, and small target detection by changing scanning paths, increasing laser numbers, and employing advanced detection techniques.
Journal Reference
Chen, YC., Chu, JF., Hsieh, KW. et al. Automatic wild bird repellent system that is based on deep-learning-based wild bird detection and integrated with a laser rotation mechanism. Sci Rep 14, 15924 (2024). doi: 10.1038/s41598-024-66920-2, https://link.springer.com/article/10.1038/s41598-024-66920-2
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.