Recent Advancements in Fruit Detection and Classification Using Deep Learning Techniques

Table 4

Implementation details of the CNN classifier.

Name	Kernel	Stride	Padding	Input	Output	Param#	Activation
Conv2d-L1	3 × 3	1 × 1	1 × 1	3 × 100 × 100	32 × 100 × 100	896	ReLU
Conv2d-L2	3 × 3	1 × 1	1 × 1	32 × 100 × 100	64 × 100 × 100	18,496	ReLU
MaxP2d	2	2	0	64 × 100 × 100	64 × 50 × 50	—	—
Conv2d-L3	3 × 3	1 × 1	1 × 1	64 × 64	128 × 50 × 50	73,856	ReLU
Conv2d-L4	3 × 3	1 × 1	1 × 1	128 × 50 × 50	128 × 50 × 50	147, 584	ReLU
MaxP2d	2	2	0	128 × 50 × 50	128 × 25 × 25	—	—
Conv2d-L5	3 × 3	1 × 1	1 × 1	128 × 25 × 25	256 × 25 × 25	295,168	ReLU
Conv2d-L6	3 × 3	1 × 1	1 × 1	256 × 25 × 25	256 × 25 × 25	590,080	ReLU
MaxP2d	5	5	—	256 × 25 × 25	256 × 5 × 5	—	—
FC-L	—	—	15,000	6,400	1024	6,554,624	ReLU
FC-L	—	—	—	1,024	512	524,800	ReLU
FC-L	—	—	—	512	120	67,203	—

Param#: parameters, Act: activation, Conv2d-L₁, L₂, L₃, L₄, L₅, L₆: convolution 2 dimension layer 1, 2, 3, 4, 5, and 6, MaxP2d: maximum pooling 2 dimension, and FC-L: fully connected layer.