๐Ÿช„Methodology

Our workflow has been designed methodically to ensure that we emphasize on the quality of the shapes that we provide while also focusing on the scalability, allowing us to meet the global demand for data.

The accuracy of the polygons is directly linked to the quality of the remote sensory image used. To ensure Shapesโ€™ offering is top-tier, we embrace a multi-sourcing strategy that allows us to source the highest quality imagery to ensure we provide a holistic view of the world we live in. Through this approach, we are able to provide the following:

  • Data Validation: Using a single source of data can introduce bias and limitations. Through multi-sourcing, we are able to ensure that the world is depicted in an accurate way, as it enables cross-verification and validation of the data. Hence, enhancing the dataโ€™s reliability.

  • Data Coverage: The ever-changing landscapes and urban developments worldwide pose a challenge in ensuring comprehensive coverage from a single source. Multisourcing allows us to expand our reach, encompassing a wider range of geographies, including remote or less accessible areas. This approach ensures that our dataset is not only more comprehensive but also more accurate.

The Shapes methodology is divided into five main steps:

Step 1: POI Identification

Our Places product is in a constant state of development as we expand our offering to new countries. To maximize value for our customers, we make sure that every new POI we introduce is paired with a visual shape representing its boundaries.

Step 2: Location Identification

To obtain the most accurate polygons for each POI, we use the geocode associated with the POI to identify the POI location on the remote sensing imagery. This enables us to pinpoint the POI's location within remote sensing imagery, ensuring accurate verification and providing spatial context. This transformation elevates a POI from a simple point to a tangible, well-defined location.

Step 3: Deploy Building Segmentation Models

Once we have identified the location of the POI of interest, we deploy our Building Segmentation Model that harnesses the capabilities of computer vision and machine learning models to detect and segment the boundary of each POI, allowing us to generate a polygon that is a true representation of the POI.

Step 4: Polygon Refinement

One of the hurdles we face in employing ML models for image recognition is the presence of noise along the edges of our polygons. This noise compromises the accuracy of the generated polygons.

To overcome this challenge, Echo has integrated specialized tools within our machine learning models to post-process the data, guaranteeing the precision of our polygons by reducing noise.

This meticulous process ultimately delivers superior quality polygons.

Step 5: Quality Assurance

Once we have derived and associated the best quality polygons for a POI, we incorporate human verification to a sample of our data, ensuring the accuracy of the generated polygons and upholding the highest quality standards. This process not only assesses our quality but also fosters ongoing model improvement through iterative training.

At Echo, we firmly believe that continuous improvement is the cornerstone of any data-driven enterprise. Consequently, we maintain an unwavering commitment to the ongoing training of our machine learning models with fresh datasets to constantly tune our models. This practice is integral to our quest for the most accurate and up-to-date data possible.

Last updated