Business Objectives

  • A lot of business data and information is available in the form of images. Some examples of such data could be image of damaged car for claims processing needs, image of finished product for quality checks etc.
  • While this image data may reveal valuable insights for a given business process, its very hard to process & consume such data in company’s business process.
  • Image Recognition models, especially Deep Learning models such as Convolutional Neural Network (CNN) or custom feature models built using OpenCV libraries can be used to train classifier models, which can help make sense of the image data.

Methodology

Data Analysis

Study images in detail.

Use OpenCV or equivalent to pre-process and do feature extraction.

Assess type of model to be applied for best results.

Images are studied in detail. Image data is pre-processed by applying different image filters (e.g. Sobel filter) using OpenCV or similar image libraries. Initial analysis of these features is done to look for patterns. An approach is identified for best results.

Model Training

Continue with Feature extraction.

Apply different models including CNN or primitive ML models.

Observe results on Train Test

Based on the chosen approach, feature extraction activity is done one more time and models are built and trained. Based on patterns observed and volume of data, either deep learning models or OpenCV feature extraction followed by normal Machine Learning model (such as Logistic regression, SVM etc.) or a combination is built. Trained models are tested on train:test split and results are published.

Present Results

Run models on blinded dataset.

Build API wrapper and build a simple GUI to access model in programs.

Host the API and models on customer’s IT infrastructure.

Once the models are ready with desired results on train:test data, models are tested on blinded datasets to confirm the results. Models may be wrapped into a RESTful API and hosted on the hosting platform. A simple or complex GUI (based on agreed scope) may be created to invoke the API and see the results.

Deliverables

  • Results on Train:Test:Validation data sets.
  • RESTful API hosted on hosting platform.
  • A simple or complex GUI program integrated with the model API

Case Studies

  • Learn how neurIOT helps a government department count children in village schools using Image Recognition techniques.
  • Learn how neurIOT helps farmers identify 9 different diseases in wheat crop using Image Recognition techniques.