Improved: Pole lean detection

An upgraded pole lean algorithm improves how poles are isolated from surrounding objects in each image, reducing cases of extreme angle misreadings and delivering more dependable lean measurements. This gives reliability teams greater confidence when using lean severity to prioritize inspections and repairs.
​
​Improved: Transformer damage classification

Building on the oil leak and corrosion classes introduced earlier this year, the transformer damage model has been upgraded with a new classification approach that further improves how damage conditions are distinguished—supporting more confident triage of at-risk equipment.

​Improved: Pole-top damage detection

The pole-top damage model has been retrained on a substantially larger dataset, further improving the identification of frayed and damaged pole tops flagged during inspections.
​
​Improved: Insulator material classification

An updated insulator material model returns far fewer "unknown" classifications, meaning more components receive a definitive polymer or porcelain determination. This improvement gives utilities working through insulator upgrade programs a more complete picture of their inventory.

In progress: Cable classification

The cable segmentation model introduced last year is performing well at outlining and identifying individual cables on poles. With that foundation in place, development has shifted to the next step: classifying the type of each cable, which will help enable automated measurements for make-ready and joint-use workflows.

Improved: Day burner detection The day burner model has been retrained on a significantly larger dataset, delivering stronger performance in identifying streetlights that remain on during the day and further improving the reliability of lighting audits and energy-savings programs.

Improved: Light source classification

The light source model has been retrained to more reliably distinguish LED from non-LED fixtures, improving the accuracy of lighting inventories and conversion planning.

New: Vehicle route visualization

The route driven during each collection can now be toggled as a layer directly on the map. Teams can see exactly where a vehicle traveled, making it easier to verify coverage, spot gaps, and plan follow-up drives.

​New: Filters and map layers for CSP transformers and open wire secondaries

CSP transformers and open wire secondaries—two findings surfaced in the Noteworthy Insights panel earlier this year—can now be filtered across your data and toggled on as dedicated map layers, making it easy to surface and visualize where these replacement candidates are concentrated across your territory.

​Improved: Flexible detection date filtering

Previously, asset detections could only be filtered by their first or last captured date. A new filtering option surfaces any asset with at least one detection within a selected date range, making it much easier to isolate everything captured during a specific window.

​Improved: General quality-of-life and reliability updates

A series of smaller improvements make everyday work in Inspect Cloud smoother: image review no longer resets zoom or position after saving an annotation edit, front-end performance is faster for territories with very large asset counts, and fixes across filters, dashboards, asset editing, and region uploads ensure counts and edits behave consistently throughout the platform.

Improved: Raw data capture flexibility

Raw image capture can now be enabled or disabled without a full system restart, and any images still pending when capture is turned off now finish uploading automatically. This makes it much easier to manage large fleets during storm assessments, where raw capture may need to be switched on or off quickly across many units.

​Bug fix: Raw data location gaps
Resolved an issue where raw data collections were showing location points without associated images. Raw capture coverage is now more complete and dependable.