Automatic Classification of Microscopic Hair Images Based on Improved Convolutional Neural Network

Table 1. Experimental environment configuration and parameters
View table
Table 1. Experimental environment configuration and parameters
Software/hardware Experimental configuration
CPU Intel(R)4 Core(TM)i5-7200U,2.50 GHz
GPU NVIDIA GeForce 940MX
RAM 8.00 GB
Frame Caffe
Toolsoftware Matlab2016a, Python2.7, VisualStudio 2013, Adobe Photoshop CS4

Table 2. Configuration parameters of Hair-Net

View table

Table 2. Configuration parameters of Hair-Net

Layer	Kernel size	Strides	Pad	Output size	Output number
Input				223×223	3
Convolution1+ ReLU1	9*9	4	1	55×55	96
Max pooling1	3*3	2	0	27×27	96
Convolution2+ReLU2	3*3	1	1	27×27	256
Convolution3+ReLU3	3*3	1	1	27×27	256
Max pooling3	3*3	2	0	13×13	256
Convolution4+ReLU4	3*3	1	1	13×13	384
Convolution5+ ReLU5	3*3	1	1	13×13	384
Convolution6+ReLU6	3*3	1	1	13×13	256
Max pooling6	3*3	2	0	6×6	256
Fully connected7+ ReLU7+ Dropout7				1×1	4096
Fully connected8+ ReLU8+ Dropout8				1×1	1024
Fully connected9				1×1	6
Output					6

Table 3. Effect of batch size on experimental results
View table
Table 3. Effect of batch size on experimental results
Batch_size Train accuracy /% Loss Test accuracy /%
16 76.15 1.149 72.43
32 83.42 0.822 80.97
64 88.50 0.709 86.39
128 94.81 0.388 93.66
256 Error Error Error

Table 4. Effects of LRN and BN on experimental results
View table
Table 4. Effects of LRN and BN on experimental results
Modelstructure Trainaccuracy /% Loss Testaccuracy /%
Hair-Net 96.15 0.247 95.32
Hair-Net+LRN 95.49 0.316 94.57
Hair-Net+BN 97.82 0.199 97.16

Tools

Get Citation

Copy Citation Text

Xiaojia Jiang, Shuhui Gao. Automatic Classification of Microscopic Hair Images Based on Improved Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2019, 56(23): 231002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites