Optical Instruments, Volume. 45, Issue 2, 26(2023)
Fruit damage detection and classification based on attention mechanism
Figures & Tables(19)
Fig. 8. Visualization thermodynamic diagram of each layer feature extracted by mode
Fig. 9. Comparison of accuracy results of each network model for the validation set
Fig. 10. Comparison of the loss value of each network model for the train set
Table 1. Comparison of classification performance of each network model
View tableView in ArticleTable 1. Comparison of classification performance of each network model
模型 准确率/% 精确率/% 召回率/% ResNet34 97.9 97.9 97.9 ResNet34+SE 98.7 98.7 98.8 ResNet34+CBAM 98.3 98.4 98.4 ResNet34+SE+CBAM 98.8 98.8 99.0
Table 2. Results of each variety in test set
View tableView in ArticleTable 2. Results of each variety in test set
类别 精确率/% 召回率/% 特异度/% 新鲜苹果 98.2 99.3 99.5 新鲜香蕉 100.0 99.8 100.0 新鲜橙子 98.8 99.3 99.8 腐烂苹果 98.3 97.7 99.5 腐烂香蕉 99.6 99.3 99.9 腐烂橙子 99.5 98.2 99.9
Table 3. Comparison of experimental test set results
View tableView in ArticleTable 3. Comparison of experimental test set results
类别 精确率/% 召回率/% F1-Score/% ResNet34+SE+CBAM 文献[ 5 ]ResNet34+SE+CBAM 文献[ 5 ]ResNet34+SE+CBAM 文献[ 5 ]新鲜苹果 98.2 98 99.3 99 98.7 97 新鲜香蕉 100.0 99 99.8 98 99.8 97 新鲜橙子 98.8 99 99.3 99 99.0 99 腐烂苹果 98.3 98 97.7 99 97.9 98 腐烂香蕉 99.6 99 99.3 98 99.4 98 腐烂橙子 99.5 99 98.2 99 98.8 99
Table 4. Comparison of performance effects before and after network improvement before the data is added into the validation set
View tableView in ArticleTable 4. Comparison of performance effects before and after network improvement before the data is added into the validation set
类别 模型 精确率/% 召回率/% 特异度/% 新鲜苹果 ResNet34 97.8 98.8 99.2 ResNet34+SE 98.0 99.2 99.4 ResNet34+CBAM 97.9 98.9 99.3 ResNet34+SE+CBAM 98.2 99.3 99.5 腐烂苹果 ResNet34 97.1 97.2 99.1 ResNet34+SE 98.0 97.5 99.5 ResNet34+CBAM 97.7 97.2 99.4 ResNet34+SE+CBAM 98.3 97.7 99.5
Table 5. Comparison of performance effects before and after network improvement after the data is added into the validation set
View tableView in ArticleTable 5. Comparison of performance effects before and after network improvement after the data is added into the validation set
类别 模型 精确率/% 召回率/% 特异度/% 新鲜苹果 ResNet34 93.8 91.3 99.0 ResNet34+SE 97.1 91.4 99.3 ResNet34+CBAM 94.1 91.4 99.1 ResNet34+SE+CBAM 97.3 91.6 99.5 腐烂苹果 ResNet34 96.0 89.2 97.6 ResNet34+SE 96.5 92.6 97.4 ResNet34+CBAM 96.1 95.1 97.8 ResNet34+SE+CBAM 96.7 96.2 97.9
Table 6. Comparison of performance effects after data enhancement before and after network improvement
View tableView in ArticleTable 6. Comparison of performance effects after data enhancement before and after network improvement
类别 损失值 准确率/% ResNet34 0.245 97.4 ResNet34+SE 0.195 98.0 ResNet34+CBAM 0.235 97.8 ResNet34+SE+CBAM 0.193 98.9
Tools
Get Citation
Copy Citation Text
Jie ZHANG, Chunlei XIA, Rongfu ZHANG, Julaiti HALIZHATI, Yi LIU. Fruit damage detection and classification based on attention mechanism[J]. Optical Instruments, 2023, 45(2): 26
Paper Information
Category: APPLICATION TECHNOLOGY
Received: Sep. 17, 2022
Accepted: --
Published Online: Jun. 12, 2023
The Author Email: XIA Chunlei (xiachunlei@usst.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20