Centroid Optimization of DNN Classification in DOA Estimation for UAV.

Area of Science:

• Signal Processing
• Machine Learning
• Artificial Intelligence

Background:

• Deep learning classification is used for direction of arrival (DOA) estimation.
• Limited DOA classes hinder accuracy for random azimuth signals in real-world applications.

Purpose of the Study:

• To present Centroid Optimization of deep neural network classification (CO-DNNC) for improved DOA estimation accuracy.
• To address limitations in current deep learning-based DOA classification methods.

Main Methods:

• The CO-DNNC method integrates signal preprocessing, a convolutional neural network (CNN) classification network, and Centroid Optimization.
• Centroid Optimization utilizes classified labels as coordinates and Softmax output probabilities to calculate signal azimuth.
• The classification network comprises convolutional and fully connected layers.

Main Results:

• CO-DNNC achieves precise and accurate DOA estimation, particularly effective in low signal-to-noise ratio (SNR) conditions.
• The method requires fewer classes for the same prediction accuracy and SNR compared to traditional methods.
• This leads to reduced deep neural network (DNN) complexity, saving training and processing time.

Conclusions:

• CO-DNNC offers a significant improvement in DOA estimation accuracy and efficiency.
• The approach is robust in challenging low SNR environments.
• CO-DNNC presents a computationally efficient alternative for DOA estimation in various applications.