Tools and resource

1. Python - Tested on Python 3.9+
2. Tensorflow - Tested on Tensorflow 2.13.0+
3. FastAPI
4. Docker

Steps for model training

1. First download the datasets from the following link: https://drive.google.com/drive/folders/17_vJzfcs4KFM9Nkqv1Bjt4HIpGLQYtMo?usp=drive_link

   
   
   

2. Pre-processing the data and clean it up by separating the data into training, validation, and test.

   
   
   

3. Then, rescale the image and also determine the target image size that we will use. We use a target size of 224px.

   
   
   

4. After splitting into 3 parts, then each data is ready for model training.

5. For our training, we use the existing pre-trained model Xception as our base model.

   
   
   

6. We also created our own model by adding a 2D Convolutional layer, MaxPooling, and also added 2 Neural Layers.

   
   
   

7. With this model, we can predict a class from the equipment data that users will use later.

   
   
   

8. We got an accuracy 99% and validation accuracy 87%

   
   
   

9. We do deployments using Docker, FastAPI, and also Cloud Run.