Acoustic Scene Classification and Visualization of Beehive Sounds Using Machine Learning Algorithms and Grad-CAM

Table 3

Performance comparison of different features using various CNNs.

CNN	Features	Learning rate	Batch size	Accuracy (%)
Shallow CNN	Mel spectrogram	1e-4	128	79.82
		1e-5	64	79.94
		1e-5	128	79.30
	MFCCs	1e-4	128	88.42
		1e-5	64	88.04
		1e-5	128	87.89
	CQT	1e-4	128	80.70
		1e-5	64	79.12
		1e-5	128	79.47
VGG-13	Mel spectrogram	1e-4	128	84.91
		1e-5	64	80.58
		1e-5	128	91.10
	MFCCs	1e-4	128	90.88
		1e-5	64	91.93
		1e-5	128	91.75
	CQT	1e-4	128	85.09
		1e-5	64	80.88
		1e-5	128	80.35