Technology Solutions Firm Leverages AI Custom Vision to Streamline its Image Accuracy and Reduce Manual Identifiers

OVERVIEW

The role of Artificial Intelligence (AI) continues to grow in importance as organizations continue to find ways to leverage the technology in day-to-day operations to streamline processes, improve decision-making and reduce manual errors.

A technology solutions firm was intrigued by the innovative opportunity to create something that would help stabilize its current process for collecting data and images as its current operational system was not fully optimized and vulnerable to human error. With its current process, data and images were subject to an individual checking these images and sending out a report that could potentially hold inaccurate representations if human error occurred.

The firm approached Concurrency to create an AI Azure Custom Vision solution after learning about how this technology has supported success at other organizations. The goal was to create vision-based, trainable machine learning algorithms to accelerate and improve the accuracy of the firm’s services provided to its customers and contractors.

SOLUTION

Microsoft Azure Custom Vision is an image recognition service that lets users build, deploy, and improve their own image identifiers. The image identifier applies labels to images according to visual characteristics, then allows users to specify the labels and train custom models to detect them. The following visual showcases how Azure Custom Vision works as the model is trained to understand what components should be included in the image and then recognize and alert the user if/when an item is missing.

To start, Concurrency worked with the firm to train the first five objects out of over fifty possible items to attain an accuracy no less than 60% for all objects inching closer to and past the accuracy of humans at 75%. After 160 hours of training the models, the accuracy increased to 80%. Training the models included uploading images and calling out labels for the machine learning algorithms to pick up on certain elements.

Eventually, five models were created with an accuracy of over 95%, demonstrating the ease of using Power Apps as a tool to prove the concept. Users were taught how to scale data collections and storage and trained on how to re-train and update the Azure Custom Vision models when needed.

RESULTS 

The Azure AI Custom Vision solution reduced the amount of time an individual would spend manually going through images to check if they were set up accurately, meaning projects were completed quicker than before and with higher accuracy.

After a few months of the solution’s implementation, the firm had handled over 3400 events/images that were monitored by a human and by Azure Custom Vision. Out of these thousands of instances, only eleven events were rejected by the Azure Custom Vision solution that the human approved. The plan is to train the model to catch those and bring down the rejections closer to between two to three instances.

The firm was able to change how they achieved business objectives and transform its organization using a simple yet compelling AI solution. It also opens doors for future projects to explore what else can be streamlined and automated with AI.