Research pick: A new twist on spotting fires starting - "Bidirectional attention network for real-time segmentation of forest fires based on UAV images"

A new system aimed at improving the monitoring and detection of forest fires through advanced real-time image processing is reported in the International Journal of Information and Communication Technology. The work could lead to faster and more accurate detection and so help improve the emergency response to reduce the environmental, human, and economic impacts.

Zhuangwei Ji and Xincheng Zhong of Changzhi College, in Shanxi, China, describe an image segmentation model based on STDCNet, an enhanced version of the BiseNet model. Image segmentation involves classifying areas within an image to allow flames and forest background to be differentiated. The STDCNet approach can extract relevant features efficiently without demanding excessive computational resources.

The team explains how their approach uses a bidirectional attention module (BAM). This allows it to focus on distinct characteristics of different image features and determine the relationships between adjacent areas in the image within the same feature. This dual approach improves the precision of fire boundary detection, particularly for small-scale fires that are often missed until they have grown much larger.

Tests with the model on a public dataset showed better performance than existing approaches in terms of both accuracy and computational efficiency. This bolsters the potential for real-time fire detection, where early identification can prevent fires from spreading uncontrollably.

The new system has several advantages over standard fire detection methods, such as ground-based sensors and satellite imagery. These have limitations such as high maintenance costs, signal transmission issues, and interference from environmental factors such as clouds and rugged terrain. The researchers suggest that drones equipped with the new image processing technology could offer a more adaptable and cost-effective alternative to sensors or satellites, allowing fire detection to be carried out in different weather conditions and in challenging environments.

Ji, Z. and Zhong, X. (2024) ‘Bidirectional attention network for real-time segmentation of forest fires based on UAV images’, Int. J. Information and Communication Technology, Vol. 25, No. 6, pp.38–51.