How IOT and AI can help identify flailing trees and prevent accidents
2 min read . Updated: 16 Jul 2019, 12:39 PM IST
Premium- The researchers examined over 100 large trees scattered in the campus of City University of Hong Kong
- Several other methods have been used in the past to evaluate the health of trees
New Delhi: In June 2019, a strong dust storm killed 17 people and injured 11 in Uttar Pradesh. According to the state government's findings, the accidents were caused by overturning of trees or collapsing buildings on the victims.
During storms and heavy rain, bigger and older trees often lose branches or even fall down completely due to weak roots. People often grow trees in their garden or backyard of the house for the greenery, fruits or aesthetic value. However, proper management including the risk assessment of big trees are not taken seriously by many.
A new study by Department of Electrical Engineering, City University of Hong Kong, claims improperly grown trees are hazardous for the environment and people living in their vicinity, as factors like climate change and soil erosion have a significant impact on their health. However, an early warning system based on ML (machine learning) and IOT (Internet of things) to assess the potential poor tree health can help people identify trees that are likely to fall during storm and heavy rain so they can get rid of them in advance. The study was published on July 15 in journal Sensors.
The researchers examined over 100 large trees scattered in the campus of City University of Hong Kong, Lok Fu Park, and Fa Hui Park to estimate a defoliation and discoloration of the leaves on a scale of 0–3. This was followed by a measurement of proximity environmental feature (PEF) of the trees using sensors connected to a controller integrated module. The function of the module was to issue commands related to timing and duration of data gathering. The module was then connected to the IoT module and the collected data was transferred to the server by the IoT network.
The environmental factors that affect tree growth such as air temperature, humidity, oxygen concentration, carbon dioxide concentration, illumination intensity and soil humidity were taken into account while measuring PEF. The report claims, if one of these features is not in the normal range trees cannot survive.
The experiment went on for several months, and the weather at the three experimental sites changed from time to time. To rule out the negative effects produced by interference factors like weather on the PEF dataset, an ADI (active data identifying) algorithm was developed to remove abnormal data. The researchers point out that the ADI algorithm improved the performance of the evaluation model.
Then the researchers used a radial basis function neural network (NN), a machine learning algorithm, to correlate tree health and PEFs and build a PTA (proximity tree health assessment) algorithm to evaluate, and monitor tree health remotely.
The researchers claim the RBF NN is a typical feed-forward neural network and is notable for its simple structure and fast training capability, which allows it to approximate arbitrary non-linear functions at a high learning speed.
Several other methods have been used in the past to evaluate the health of trees. For instance, the assessment based on satellite data and hyper-spectral image work well in open areas with green cover, in urban areas background materials like buildings have been found to affect the reliability of the sensing data. Also, a tree with too many leaves also obstructs the data.
IOT sensors based monitoring can clearly be more effective. However, a lot more research is needed before such processes can be used in cities vying to become smart and avoid accidents caused by falling trees.
