Lightweight AAC Audio Steganalysis Model Based on ResNeXt

Table 1

Detection accuracies of different network structures.

Number	Network structure parameters	Accuracy (%)
#1	Proposed model—LARXNet	95.13
#2	Remove the high-pass filter layer	93.56
#3	A set of ResNeXt modules are added	94.25
#4	The cardinality of ResNeXt module is adjusted to 16	93.85
#5	The cardinality of ResNeXt module is adjusted to 4	93.82
#6	ResNeXtBlock1 shortcut replaced with convolution kernel	94.10