Prediction Pipeline

At the heart of Algasat is our Prediction Pipeline. When you create a job in Algasat, the Prediction Pipeline is run on your satellite image. The pipeline extracts features from your satellite image, transforms these features into a data structure that can be ingested by our machine learning model, and once the model has made a prediction, converts these predictions into a report, including a heatmap image representing the seaweed biomass. Below we go into detail about each stage of the pipeline.

The stages of the prediction pipeline

Pending

The job has been created and is waiting for the pipeline to begin.

Processing Image

The input image is transformed, and various features of the image are extracted. These features are used to perform calculations, that are then combined and passed as input into our machine learning model.

Calculating

The input is passed to our machine learning model, and a prediction is made. Our model is a Convolutional Neural Network (CNN) which is commonly used to identify patterns in images. A CNN is made up of individual nodes that form layers. The inputs to nodes in a single layer have a weight assigned to them that changes the effect that parameter has on the overall prediction. So, at each layer of our CNN, patterns are being identified, passing the result to the next layer, to eventually give a prediction. Our CNN has been tuned to work effectively at predicting the biomass of Ascophyllum nodosum.

Generate Report

The predictions from our model are then transformed into a new GeoTiff file, a heatmap representing the predicted seaweed biomass in the image. We also calculate some statistics at this stage, such as the biomass range and standard deviation in the image.

Completed

The pipeline has now been completed; you will now be able to view the results.