AusafMo/NoteBook-Medicinal-Herb-Model-ResNet

A possible tool for assistance in Alternative Medicine using DeepLearning models of Resnet-50, looking to try more algorithms especially VGG16. If someone has any suggestions please feel obliged to guide me .

🙋‍♂️ The Training notebook(s) used to train/FineTune ResNet50 on Mendley Medicinal leaf dataset

2nd One is latest btw

The Code for the model serving can be found at the following repo links, There are two repos for Front-End and Back-End Components

• For Backend App see Backend Repo
• For Front-End see Training Notebook

What is this? :

• The Project aims to solve the problem of identification of medicinal herbs.
• The Machine Learning Model uses ResNet, with a validation accuracy of 98% and testing accuracy of 96%.
• The training and testing dataset contains over 1500 images across 50 medicinal herb species.
• The model is capable of classifying the user-given image into 30 species.

Tech Stack Used :

• Python 3
• Pandas
• Numpy
• Pillow
• PyTorch
• Tensorflow
• ResNet50 model trained on Mendeley Medicinal Leaf Dataset
• Dataset: S, Roopashree; J, Anitha (2020), “Medicinal Leaf Dataset”, Mendeley Data, V1, doi: 10.17632/nnytj2v3n5.1
  

Training, Validation and Test Accuracies for 54 Epochs (Early Stoping):

• Training Accuracy : 97.55%
• Validation Accuracy : 98.64%
• Top 1 Testing Accuracy : 96.43%
• Top 5 Testing Accuracy : 99.07%

Training Vs Validation Accuracy :

Sample Predictions :