0 引言
1 实验与方法
1.1 实验系统
Fig. 1 Schematic diagram of the system workflow图1 系统流程示意图 |
1.2 图像采集
1.3 模型构建
1.4 模型训练及评价
2 结果与讨论
2.1 模型训练优化
Fig. 2 Test accuracy and loss curves for the classification task optimization process in three models: (a) ResNet; (b) MobileNet; (c) EfficientNet图2 三种模型分类任务优化过程的测试精度和损失曲线:(a)ResNet;(b)MobileNet;(c)EfficientNet |
Fig. 3 Quantitative prediction of training process evaluation metrics SRME and R2 curves for three models: (a) ResNet; (b) MobileNet; (c) EfficientNet图3 三种模型定量预测训练过程的评价指标SRME和R2曲线:(a)ResNet;(b)MobileNet;(c)EfficientNet |
2.2 模型分类性能评价
Fig. 4 Confusion matrices for algae species recognition using three models: (a) ResNet; (b) MobileNet; (c) EfficientNet图4 三种模型对藻种识别的混淆矩阵:(a)ResNet;(b)MobileNet;(c)EfficientNet |
2.3 模型回归预测性能评价
Fig. 5 Biomass prediction for three types of algae using ResNet (a, d, g), MobileNet (b, e, h), and EfficientNet (c, f, i)图5 ResNet(a、d、g)、MobileNet(b、e、h)和EfficientNet(c、f、i)对三种藻类的生物量预测 |
Table 1 Images of three types of microalgae at various concentrations in an independent validation set表1 独立验证集中三种微藻在不同浓度下的图像 |
藻类 | 独立验证数据集各浓度图像/(g/L) | ||||
---|---|---|---|---|---|
小球藻 | 0.08 ![]() | 0.18 ![]() | 0.28 ![]() | 0.38 ![]() | 0.50 ![]() |
0.70 ![]() | 0.90 ![]() | 1.50 ![]() | 2.50 ![]() | 3.50 ![]() | |
红藻 | 0.08 ![]() | 0.18 ![]() | 0.28 ![]() | 0.38 ![]() | 0.50 ![]() |
0.70 ![]() | 0.90 ![]() | 1.50 ![]() | 2.50 ![]() | 3.50 ![]() | |
螺旋藻 | 0.08 ![]() | 0.18 ![]() | 0.28 ![]() | 0.38 ![]() | 0.50 ![]() |
0.70 ![]() | 0.90 ![]() | 1.50 ![]() | 2.50 ![]() | 3.50 ![]() |
Table 2 Evaluation metrics for algal biomass prediction performance of three models表2 三种模型的藻生物量预测性能评价指标 |
模型 | SRME | R2 | ||||
---|---|---|---|---|---|---|
小球藻 | 红藻 | 螺旋藻 | 小球藻 | 红藻 | 螺旋藻 | |
ResNet | 0.518 1 | 0.206 9 | 0.279 2 | 0.766 4 | 0.962 8 | 0.921 5 |
MobileNet | 0.699 5 | 0.280 0 | 0.595 1 | 0.574 3 | 0.931 8 | 0.686 4 |
EfficientNet | 0.453 4 | 0.267 0 | 0.445 4 | 0.821 2 | 0.938 0 | 0.827 4 |