Hyperspectral Imaging-Based Quality Classification for Kiwifruit by Incorporating Three-Dimensional Convolution Neural Network and Haar Wavelet Filter

Ke Jin; Zhiqiang Guo; Yunliu Zeng; Gang Ding

doi:10.3788/LOP223142

Laser & Optoelectronics Progress, Volume. 60, Issue 20, 2010003(2023)

Hyperspectral Imaging-Based Quality Classification for Kiwifruit by Incorporating Three-Dimensional Convolution Neural Network and Haar Wavelet Filter

Ke Jin¹, Zhiqiang Guo^1、*, Yunliu Zeng^2、**, and Gang Ding²

Author Affiliations

¹College of Information Engineering, Wuhan University of Technology, Wuhan 430070, Hubei , China

²Key Laboratory of Horticultural Plant Biology, Ministry of Education, National R&D Center for Citrus Preservation, College of Horticulture and Forestry Sciences, Huazhong Agricultural University, Wuhan 430070, Hubei , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Convolution process. (a) 2D Convolution; (b) 3D Convolution

Download full size

Fig. 2. Overall network structure

Download full size

Fig. 3. Steps of acquiring region of interest mask

Download full size

Fig. 4. Average reflectivity of center, edge and background area. (a) Sampling area; (b) average reflectance spectra curve

Download full size

Fig. 5. Accuracy and loss curves of different models. (a) Accuracy curves on training set; (b) accuracy curves on validation set; (c) loss curves on training set; (d) loss curves on validation set

Download full size

Fig. 6. Confusion matrix of control group. (a) Confusion matrix of CARS_SpectralNet; (b) confusion matrix of Without_2D-DWT; (c) confusion matrix of 3DCNN_SpectralNet

Download full size

Fig. 7. t-SNE diagram of different models on the test set. (a) t-SNE diagram of features extracted from Without_2D-DWT model; (b) t-SNE diagram of features extracted from 3DCNN_SpectralNet model

Download full size

Table 1. Parameters of each module

View table

Table 1. Parameters of each module

Layer name	Input size	Output size	Keral_size	Padding
3DCov_1	30 $\times$ 30 $\times$ 150 $\times$ 1	28 $\times$ 28 $\times$ 30 $\times$ 8	$3 \times 3 \times 5,8, s = 1 \times 1 \times 5$
3DCov_2	28 $\times$ 28 $\times$ 30 $\times$ 8	26 $\times$ 26 $\times$ 28 $\times$ 1	$3 \times 3 \times 3,1, s = 1$
2D-DWT	26 $\times$ 26 $\times$ 28	13 $\times$ 13 $\times$ 112
2DCov_1_x	13 $\times$ 13 $\times$ 112	13 $\times$ 13 $\times$ 128	$[\begin{matrix} 3 \times 3,128, s = 1 \\ 3 \times 3,128, s = 1 \end{matrix}]$	‘same’
2DCov_2_x	26 $\times$ 26 $\times$ 28	13 $\times$ 13 $\times$ 128	$[\begin{matrix} 3 \times 3,64, s = 1 \\ 3 \times 3,128, s = 2 \end{matrix}]$	‘same’
2DCov_3_x	13 $\times$ 13 $\times$ 112	13 $\times$ 13 $\times$ 128	$[\begin{matrix} 3 \times 3,128, s = 1 \\ 3 \times 3,128, s = 1 \end{matrix}]$	‘same’
2DCov_4_x	13 $\times$ 13 $\times$ 256	13 $\times$ 13 $\times$ 128	$[\begin{matrix} 3 \times 3,256, s = 1 \\ 3 \times 3,128, s = 1 \end{matrix}]$	‘same’
2DCov_5_x	13 $\times$ 13 $\times$ 256	9 $\times$ 9 $\times$ 64	$[\begin{matrix} 3 \times 3,256, s = 1 \\ 3 \times 3,128, s = 1 \\ 3 \times 3,64, s = 1 \end{matrix}]$	$[\begin{matrix} ‘ s a m e ’ \\ - \\ - \end{matrix}]$

Table 2. Dataset Statistics
View table
Table 2. Dataset Statistics
Category Mature group Semimature group Immature group
Number of fruit 228 164 101
Number of data block 31540 23746 15130

Table 3. Classification results of different models on test sets

View table

Table 3. Classification results of different models on test sets

Model	Acc	Mature kiwi			Half_Mature kiwi			Unripe kiwi
Model	Acc	PPV	R	F1	PPV	R	F1	PPV	R	F1
EfficientNet_B0	0.909	0.932	0.942	0.937	0.879	0.904	0.892	0.909	0.847	0.877
GoogLeNet	0.932	0.948	0.953	0.951	0.919	0.921	0.920	0.916	0.900	0.907
ResNet18	0.824	0.857	0.895	0.876	0.793	0.787	0.790	0.797	0.731	0.763
ResNet34	0.879	0.906	0.918	0.912	0.858	0.852	0.855	0.854	0.840	0.847
HybridSN	0.750	0.820	0.844	0.831	0.702	0.664	0.682	0.676	0.691	0.684
3DCNN_SpectralNet	0.973	0.977	0.980	0.979	0.968	0.969	0.969	0.970	0.961	0.965

Table 4. Experimental results of control group

View table

Table 4. Experimental results of control group

Model	Acc	Mature kiwi			Half_Mature kiwi			Unripe kiwi
Model	Acc	PPV	R	F1	PPV	R	F1	PPV	R	F1
CARS_SpectralNet	0.917	0.937	0.951	0.944	0.902	0.898	0.900	0.900	0.878	0.889
Without_2D-DWT	0.899	0.936	0.925	0.930	0.880	0.885	0.883	0.854	0.866	0.860
3DCNN_SpectralNet	0.973	0.977	0.980	0.979	0.968	0.969	0.969	0.970	0.961	0.965

Tools

Get Citation

Copy Citation Text

Ke Jin, Zhiqiang Guo, Yunliu Zeng, Gang Ding. Hyperspectral Imaging-Based Quality Classification for Kiwifruit by Incorporating Three-Dimensional Convolution Neural Network and Haar Wavelet Filter[J]. Laser & Optoelectronics Progress, 2023, 60(20): 2010003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites