Achieve 99% accuracy in air quality monitoring with supervised learning

Recent strides in the efficacy of AI have made it possible to mitigate risks associated with poor air quality. Real-time...

Achieve 99% accuracy in air quality monitoring with supervised learning

Recent strides in the efficacy of AI have made it possible to mitigate risks associated with poor air quality. Real-time air particle monitoring means we can manage air quality to the highest standards. We achieved just that for our clients in the air quality monitoring sector. Harsh environmental conditions can affect the performance of their sensor systems. To overcome these issues, Sparkd developed an effective API capable of autonomously and accurately monitoring air quality.  

Humans take approximately 20,000 breaths every day. With each breath, we could be inhaling toxic particulate matter, without even realising it. Eye irritation, inflamed airways, and respiratory issues are just a few of the side effects of particle pollution. For air quality monitoring, particulate matter is defined by its diameter. Particles with a diameter of 10 micrometres or less (PM10) are inhalable and can have major health implications. Particles with a diameter of 2.5 microns or less (PM2.5) are particularly dangerous. PM2.5 particles are so minute, they can only be detected by an electron microscope. That’s why it’s no surprise that detecting particulate matter is a major challenge, especially in harsh weather conditions.

Sensors like optical particle counters have the power to revolutionise health and safety standards. This is particularly true in environments like industrial plants, construction sites, and mines. Metal, gravel, concrete, wood, and sand are difficult to detect. While monitoring the levels of these particles is integral to onsite health and safety compliance, it’s not always achievable. Accurate sensors that can detect all potential non-compliant events are rarely found in the current market. 

The challenge of detecting particles in high-humidity environments

The effectiveness of an air particle monitoring system is subject to a number of variables. High humidity levels are particularly problematic. In circumstances where the relative humidity level is high, sensors may give false readings. This is because such sensors often detect mist or fog droplets as particles. Low-cost particle sensors are commonly used in numerous applications. Unfortunately, most of them have no system in place to remove water from the sample before measurement. While the presence of water in particles is not harmful to humans, inaccurate air quality readings can be. Such errors undermine the integrity of air particle monitoring systems. As a result, low-cost sensors are not suitable for industry compliance applications.

Can Artificial Intelligence increase the performance of air quality sensors?

Machine learning technology can strengthen sensor systems and overcome environmental variables. ML increases the performance of air quality sensors by reducing false alarms, and improving early warning alerts. AI-enabled smart sensors account for environmental factors like wind direction, humidity, and temperature. Through deep learning algorithms they can predict weather conditions and prepare for them. These sensors allow to meet compliance standards and environmental targets in any weather conditions. On top of that, by pairing AI components with sensor, platform managers can be notified of changes in real-time. Data will be accessible to them in an instant via a dashboard, helping to streamline operations. 

Gemmo Solution: 99% accuracy in particle detection

Challenges to air quality monitoring are what inspired Gemmos particle detection algorithms. The solution we built is underpinned by the understanding that the success of any AI application depends on data. When developing a solution for our client, we collected a large quantity of annotated data from their monitoring devices. This data allowed us to train the algorithm using many different attributes. Doing so, powered a solution that understands weather conditions and detects certain particles in the air.

The algorithm’s effectiveness is owed to a supervised learning approach. This approach is notoriously more effective in classifying events of interest in comparison to an unsupervised approach. The supervised learning approach was enabled by our access to data collected in the wild. We perfected the solution by testing multiple approaches, but we defaulted towards a simple XGBoost classifier. This choice was due to the speed of the algorithm. After a process of data mining, machine learning, and algorithm configuration, the Gemmo solution was born: A highly accurate embedded algorithm capable of detecting particles with 99% precision.

Final Thoughts

At Gemmo we are passionate about building computer vision solutions that have real-world impact. Often this means, creating solutions for the most fundamental constituents of our environment. On a large scale, environmental tracking via AI-enabled sensors allows us to control harmful conditions. With this system, we can monitor everything from noise levels down to the air we breathe. Our bespoke air particle monitoring solution overcomes environmental challenges and provides reliable, automated results for our customers. 

 

Work with Gemmo today for your AI needs.

 

Check out our other articles: 

 

Start your AI Journey Today

 

Subscribe to Newsletter

Author: Johanna Walsh 

Start your Ai journey today!

Related Articles

Gemmo's noise classification case study with Sonitus