Multi-Input Convolutional Neural Network for Flower Grading

Table 2

Architecture of models with one convolutional layer before merging.

Name of models	Convolutional layers
	C1–C3	C4	C5	C6	C7	C8	FC1	FC2	ACC
M12	3 × 3,32	3 × 3,32	3 × 3,32	3 × 3,64	3 × 3,64	3 × 3,64	32	3
M13	3 × 3,32	3 × 3,32	3 × 3,32	3 × 3,64	3 × 3,64	—	32	3
M14	3 × 3,32	3 × 3,32	3 × 3,32	3 × 3,64	—	—	32	3
M15	5 × 5,16	3 × 3,32	3 × 3,32	3 × 3,32	—	—	32	3
M16	7 × 7,16	3 × 3,32	3 × 3,32	3 × 3,32	—	—	32	3
M17	3 × 3,32	3 × 3,48	3 × 3,48	3 × 3,48	—	—	32	3
M18	3 × 3,16	3 × 3,32	3 × 3,32	—	—	—	32	3
M19	3 × 3,16	3 × 3,32	3 × 3,16	—	—	—	16	3
M20	5 × 5,16	3 × 3,32	3 × 3,16	—	—	—	16	3
M21	5 × 5,16	3 × 3,64	3 × 3,32	—	—	—	32	3
M22	7 × 7,16	3 × 3,32	3 × 3,16	—	—	—	16	3
M23	3 × 3,16	5 × 5,32	3 × 3,32	—	—	—	32	3

C1–C3 represent the layers in three branches before merging. Pooling layers are ignored in this table. Generally, every convolutional layer is followed by a pooling layer. All the sizes and strides of pooling layers are set to 2 × 2.