AI understands service dogs better than veteran trainers

An international team of scientists at the University of Haifa has successfully developed an artificial intelligence (AI) model that can determine when a sniffer dog detects a target scent.

An international team of scientists at the University of Haifa (Israel) has successfully developed an artificial intelligence (AI) model that can determine when a police dog detects a target scent, with even higher accuracy than experienced trainers.

This step is expected to revolutionize the use of service dogs for security and rescue.

Police dogs have long been the effective "hearing arm" of the police, customs, military or rescue forces, thanks to their ability to sniff out far beyond humans, when searching for traces of explosives, drugs, cash or victims.

Veteran trainers often claim that they can “read” a dog’s success through small behaviors, especially tail movements. But until now, there has never been any scientific research to prove this.

In the study published in Royal Society Open Science, scientists trained eight dogs to recognize a synthetic odor, then filmed the entire sniffing process at varying levels of difficulty.

Specialized cameras record every movement of the tail: angle, speed, and wagging. From this data, the team applied computer vision technology to build an AI model capable of accurately predicting when a dog is approaching a target scent.

The results showed that the AI ​​model achieved 77% accuracy, while 190 professional trainers were only correct 46%.

According to the study authors, this demonstrates that dogs often send “silent signals” with tail movements before making an obvious action to signal that they have detected a scent – ​​and that AI can recognize those signals more sensitively than the human eye./.

(Vietnam News Agency/Vietnam+)

Source: https://www.vietnamplus.vn/ai-thau-hieu-cho-nghiep-vu-hon-ca-huan-luyen-vien-ky-cuu-post1058393.vnp