Accompanying lecture notes & lab workshop: [ Ссылка ]

[ Using Deep Learning and Transfer Learning to Conduct Customized Image Analysis ]
by: Sam Gu [ Data Science Trainer ]
May 2017

1. Lecture: Agenda
Refer to lecture notes: Notes/Notes_Image_Analysis.pdf
* Deep Learning Basics for Image Analysis
* Real World Image Analysis Needs
* Idea of Transfer Learning
* Architecture of Transfer Learning
* Hands-on Datalab Workshop on GCP


2. Lab: Hands-on Datalab Workshop on GCP
Refer to lab workshop: Lab/Lab_Image_Analysis.ipynb
In this lab, you will carry out a transfer learning example based on Google Inception-v3 image recognition neural network.
* Explore images in customer’s industry.
* Reposition a pre-trained deep neural net for new image recognition task.
* Perform feature extraction.
* Obtain deep feature representation of customer’s original image.
* Train a simple machine learning model for new classification task.
* Evaluate results of this transfer learning model.

Credit: This python notebook was adapted based on: [ Ссылка ]

Image Analysis Lab
Detect Normal or Abnormal Industrial Valves, Using Transfer Learning Technology upon Google's Pre-Trained Deep Neural Network

The use case here is to use drone to provide regular surveillance on remote or dangerous areas, capturing image of industrial equipment like valves, them steam the image back for automatic malfunction diagnosis, using machine intelligence. This improves safety and efficiency compared to current human-involved processes, without large investment on wired sensor infrastructure. The core part of this solution involves advanced image analysis in real world.

свернуть