The information from the sensors are processed in a miniature computer. The small size is a prerequisite for future applications. Credit: Olov Planthaber
On June 14, 2017, a high-rise fire at Grenfell Tower in London resulted in the death of 72 people. The fire burned for two days with many efforts to reach the top floors unsuccessful. One the fire was declared extinguished, firefighters used a drone to search for casualties.
When news of that method reached Donatella Puglisi, a professor at Linköping University (Sweden), she turned it into a research project.
“Society has a genuine need for our research,” said Puglisi. “We received high interest to develop an innovative scientifically robust technical solution.”
That solution comes in the form of a portable gas sensor system that can be trained through machine learning to identify human remains—kind of like cadaver dogs.
“Dogs have an extraordinary olfactory capability, which is reportedly 10 000 to 100 000 times better than that of people. If trained properly, they can be exceptionally helpful. They are considered the most rapid and efficient tool for odor detection among the police community. However, there are some legal and ethical dilemmas regarding the use of the dog response as evidence in court,” said Puglisi.
Beyond court, Puglisi says there are a few other reasons why a technical solution that can complement cadaver dogs is necessary. For example, in the case of the Grenfell Tower fire, it would have been ethically wrong to send a dog to the top floors of the tower to search for people since the structure was compromised.
“In a situation like that it would have been possible to use drones equipped with gas sensors, which could have searched through the building and found the bodies in a safe manner,” said Puglisi.
Additionally, it is difficult and time-consuming to train cadaver dogs. In Sweden, this has led to a shortage of cadaver dogs, which has made the Swedish police even more interested in Puglisi’s sensor.
A decaying corpse consists of many different tissues that smell differently, and the smell varies depending on the degree of decay. For Puglisi’s gas sensor to determine whether the remains are human rather than animal, every odor must have a “fingerprint” that can be identified and categorized by gas chromatography/mass spectrometry (GC/MS).
But a biological system based on millions of years of evolution is hard to emulate. Dogs have more than 200 million odor receptors in their nose, and their sense of smell takes up about one-third of their brain. Meanwhile, humans have about 5 million odor receptors, and our sense of smell takes up just 5 percent of our brain.
That’s why Puglisi is applying machine learning techniques to “train” the portable gas sensor to detect all the associated smells of decaying flesh. The information from the sensors will then be processed by an on-board miniature computer.
The “electronic nose” also boasts consistency, as opposed to cadaver dogs that can be negatively affected by their surroundings or just be having a bad day.
“We aim to develop an air analysis device based on smart sensors and data engineering that may effectively help the searching,” said Puglisi. “[It] could serve as a powerful complementary tool to cadaver-detection dogs.”