Lightweight AAC Audio Steganalysis Model Based on ResNeXt

Table 3

Comparison of detection results of neural network steganalysis methods.

Steganalysis model	Steganography algorithm	Accuracy (%)					Parameters (K)	Model size (MB)
		0.1	0.2	0.3	0.5	1.0
Our model—LARXNet	MIN	98.65	99.05	99.12	99.34	99.90	102	1.49
	SIGN	94.27	96.65	98.54	99.12	99.87
	HCM	85.50	91.65	95.13	96.85	99.55
Spec-ResNet	MIN	94.31	95.72	96.43	98.81	99.85	206	2.49
	SIGN	92.12	93.29	94.79	98.48	99.82
	HCM	72.50	81.53	88.03	93.27	98.35
WASDN	MIN	93.57	94.52	95.96	98.70	99.61	73000	692.27
	SIGN	90.24	92.19	93.25	98.35	99.57
	HCM	69.32	73.47	81.10	85.32	92.95