Fig. 8: Convolutional feature maps from the first DLoG layer at different learned orientations. | npj Acoustics

Fig. 8: Convolutional feature maps from the first DLoG layer at different learned orientations.

From: Towards accurate bird sound recognition through multi-scale texture-aware modeling

Fig. 8

Each subplot shows the response of a directional DLoG kernel to a bird spectrogram input: a ~0°, highlighting vertical structures; b ~45°, capturing ascending modulations; c ~90°, enhancing horizontal harmonics; and d ~135°, detecting descending contours. The filters demonstrate distinct sensitivity to orientation-specific acoustic patterns.

Back to article page