Image recognition
Picture recognition through machine learning
Picture recognition through machine learning provides the points of interest; i.e. the damage types and severity classes.
Initially, the model is optimized to identify damages in sewage systems related to surfaces, fractures and joints. This focus in optimization is based on close dialogue with companies and contractors that specifically look to surfaces, fractures and joints when assessing damages and necessary remedies.
The picture recognition model has been trained in certain, specific steps
We have collected data from sewage inspections from a number of wastewater companies. Our base training data consists of short om 1 mill. observations.
We have gone through data and tagged damage types and severity on a scale from 0 to 4 with 0 indicating no damage or class.
We have separated our observation data in a training set and a test set to assess and track the statistical robustness of our picture recognition models.
We have performed quality assurance in several steps, e.g. we have verified damage type and class in our data with professional annotators and with utility company experience, we have tracked statistical performance, and we have comprised and augmented data to accurately reflect actual data.
We have prioritized development work to focus on most important damage types to secure a strong statistical picture recognition model.
We have adjusted output to include points of interests and possible points of interest including false positives from a principle about ‘better safe than sorry’.
We continually develop and improve the model. We do not expect the model to work completely independent of utility company experience at any point.
We expect utility company experience to be powered by the picture recognition model as it lists points of interests and help the utility company employees save valuable time. The utility company and its employees have last say in interpretation and prioritization.
Select performance indicators…
Professional drone inspections
Fill out the form below to get started. Our expert team will contact you to discuss your specific needs and provide a tailored solution that meets your requirements.