Lightweight Multi-Task Apple Ripeness Classification Model (Invited)

Li Zhang; Xiaoge Wang; Chun Bao; Jie Cao; Qun Hao

doi:10.3788/LOP240953

Journals >Laser & Optoelectronics Progress >Volume 61 >Issue 20 >Page 2011012 > Article

Laser & Optoelectronics Progress
Vol. 61, Issue 20, 2011012 (2024)

Lightweight Multi-Task Apple Ripeness Classification Model (Invited)

Li Zhang¹, Xiaoge Wang², Chun Bao¹, Jie Cao^1,3,*, and Qun Hao⁴

Author Affiliations

¹School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, China

²School of Mechanical Engineering, Shandong University of Technology, Zibo 255022, Shandong , China

³Yangtze Delta Region Academy, Beijing Institute of Technology, Jiaxing 314003, Zhejiang , China

⁴School of Optoelectronic Engineering, Changchun University of Science and Technology, Changchun 130022, Jilin , China

show less

DOI: 10.3788/LOP240953 Cite this Article Set citation alerts

Li Zhang, Xiaoge Wang, Chun Bao, Jie Cao, Qun Hao. Lightweight Multi-Task Apple Ripeness Classification Model (Invited)[J]. Laser & Optoelectronics Progress, 2024, 61(20): 2011012 Copy Citation Text

show less

Fig. 1. Partial samples of apple fruit appearance defect dataset

Download full size

Fig. 2. Partial sample images

Download full size

Fig. 3. Luminance preprocessing based on gamma changes

Download full size

Fig. 4. Image preprocessing process based on Gamma changes under intense sunlight

Download full size

Fig. 5. Structure of multi-task classification architecture

Download full size

Grade No.	Grade 1	Grade 2	Grade 3	Grade 4
Typical image

Table 1. Comparison chart between apple ripen grades and fruit skin colors

Index	Training set	Testing set	Total
Defective	1280	250	1530
No defect	1305	250	1555
Total	2585	500	3085

Table 2. Dataset of fruit defects

Grade No.	Training set	Testing set	Total
Grade 1	1293	250	1543
Grade 2	1312	250	1562
Grade 3	1289	250	1539
Grade 4	1234	250	1484
Total	5128	1000	6128

Table 3. Maturity dataset

Model	AlexNet	ResNet-18	ResNet-34	VGG-16	D-Net
Recall	0.92	0.96	0.92	0.85	0.96
Precision	0.88	0.93	0.96	0.88	0.96
Accuracy	0.90	0.94	0.94	0.86	0.96

Table 4. Classification results of fruit defects by D-Net

Grade No.	Indicator	AlexNet	ResNet-18	ResNet-34	VGG-16	M-Net
Grade 1	Recall	0.88	0.92	0.92	0.87	0.89
	Precision	0.84	0.92	0.96	0.80	0.96
	F1 score	0.86	0.92	0.94	0.83	0.92
Grade 2	Recall	0.78	0.81	0.85	0.66	0.92
	Precision	0.85	0.84	0.88	0.76	0.88
	F1 score	0.81	0.82	0.86	0.70	0.90
Grade 3	Recall	0.84	0.84	0.92	0.73	0.96
	Precision	0.84	0.84	0.88	0.76	0.92
	F1 score	0.84	0.84	0.90	0.75	0.94
Grade 4	Recall	0.92	0.96	1.0	0.91	0.96
	Precision	0.88	0.92	0.96	0.80	0.96
	F1 score	0.90	0.94	0.98	0.85	0.96
Average accuracy		0.85	0.88	0.92	0.78	0.93

Table 5. Experimental comparison results of the maturity classification model

Li Zhang, Xiaoge Wang, Chun Bao, Jie Cao, Qun Hao. Lightweight Multi-Task Apple Ripeness Classification Model (Invited)[J]. Laser & Optoelectronics Progress, 2024, 61(20): 2011012

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享