Methodological Selection of Optimal Features for Object Classification Based on Stereovision System.

Area of Science:

• Environmental Science
• Avian Ecology
• Artificial Intelligence

Background:

• Wind energy expansion increases collision risks for birds, particularly endangered species and birds of prey, due to their flight patterns.
• Accurate identification of flying objects is crucial for mitigating bird-turbine collisions.

Purpose of the Study:

• To develop and test an efficient object classification method using stereovision data.
• To distinguish endangered birds from common birds and other flying objects like aircraft.

Main Methods:

• Utilized a stereovision system (Bioseco BPS) for data acquisition.
• Extracted both motion and visual features from detected objects.
• Employed correlation-based and wrapper-type approaches with genetic algorithms (GAs) for feature selection and classifier optimization.

Main Results:

• Achieved 98.6% recall and 97% accuracy in distinguishing birds from aeroplanes.
• Successfully differentiated endangered birds from common birds with 93.5% recall and 77.2% accuracy.

Conclusions:

• The developed stereovision system and feature extraction methodology show high potential for identifying and protecting endangered avian species from wind turbine impacts.
• Optimized feature selection and classification are key to improving the accuracy of distinguishing between different types of flying objects and bird species.