Multiple Musical Instrument Signal Recognition Based on Convolutional Neural Network

Table 4

Changes of feature size in the classification model based on the attention network.

Input size	Operation	Output size
176 × 165 × 6	2 × 1 Convolution kernel, 352 channels	352 × 164 × 6
352 × 164 × 6	3 × 1 Maximum pooling	352 × 54 × 6
352 × 52 × 6	3 × 1 Convolution kernel, 704 channels	704 × 52 × 6
704 × 52 × 6	3 × 1 Channels	704 × 17 × 6
704 × 17 × 6	2 × 1 Channels, 11 channels	11 × 8 × 6
704 × 17 × 6	Attention subnet	Six attention weights
11 × 8 × 6	The sum using the weighting of attention weight	11 × 8