BLNN: Multiscale Feature Fusion-Based Bilinear Fine-Grained Convolutional Neural Network for Image Classification of Wood Knot Defects

Table 2

Parameters of BLNN layers.

Layer	Type	Patch size	Kernel sum	Stride	Output size	Neuron sum
Input	Input
Conv11	Convolution		16	1
BN11	BatchNorm
ReLU11	ReLU
Pool11	Avg-pooling			2
Conv12	Convolution		32	1
ReLU12	ReLU
Pool12	Avg-pooling			2
FC11	Fully connected		120
Conv21	Convolution		16	1
ReLU21	ReLU
Pool21	Avg-pooling			2
Conv22	Convolution		32	1
ReLU22	ReLU
Pool22	Avg-pooling			2
FC21	Fully connected		120
Cas	Cascade
FC31	Fully connected		50
FC32	Fully connected		4
Output	Output		4