Fig 1.
High-level study procedures.
Fig 2.
Data distribution in each target class.
Fig 3.
Sample leaf images.
Table 1.
Samples of target label mapping.
Fig 4.
Architecture of proposed DWS-CNN (finetuned Mobile Net).
Fig 5.
Standard convolution to produce one 8x8 output feature.
Fig 6.
Standard convolution to produce 256 8x8 image features.
Fig 7.
Depth-wise convolution to produce 3 channel 8x8 image feature.
Fig 8.
Point-wise convolution to produce one 8x8 image feature.
Fig 9.
Point-wise convolution to produce 256 8x8 image features.
Table 2.
The parameters used in the study experiment.
Fig 10.
Accuracy and loss of a model for training and testing datasets.
Table 3.
Testing accuracies of adopted models using learning rate 1*10-4.
Table 4.
Testing accuracies of adopted models using learning rate 1*10-3.
Table 5.
Correctly classified and misclassified classes performed by the proposed model (DWS-CNN).
Fig 11.
Confusion matrices of Proposed Model (Tuned DWS-CNN) for eight categories.
Table 6.
Classification report of the proposed model (Fine-tuned DWS-CNN) for eight categories.
Fig 12.
F1_score value of the models using learning rate = 1*10-4 and batch size of 32.
Table 7.
Testing accuracy of the models for each training.
Table 8.
Mean and standard deviation of accuracy and F1-score of each model.
Table 9.
Training time of proposed model and variants of CNN model.
Table 10.
Comparative analysis of previous work with the proposed method.
Fig 13.
Predicted medicinal plant using the proposed model.