Debris flow detection and velocity estimation using deep convolutional neural network and image processing

This study presents a novel method for the automatic detection of debris flow motion and velocity measurement using deep learning and image processing techniques. An advanced convolutional neural network (CNN) model based on the You Only Look Once algorithm was employed to identify debris flow motion from videos recorded by a camera system. An image processing technique was also proposed to calculate the front velocity of the detected debris flow along a channel. The CNN model was trained and tested on an image dataset (named Debrisflow21) derived from 12 debris flow videos (5950 frames) that were obtained from small flume tests, large flume tests, and several debris flow events. The results showed that the debris flow detection model using CNN achieved an average precision (AP) of 96.37% and an average intersection over union of 84.80% on the test datasets. The application results of the proposed CNN model to five additional videos reached approximately 39 frames per second with an AP over 99.72%. In addition, the accuracy of the velocity calculation results tested on small flume and large flume experiment videos ranged between 87.1 and 97.3%. The proposed method exhibited high accuracy and fast processing speed; thus, it can be applied for early detection and warning systems.

Automated summary

This study presents a novel method for automatically detecting debris flow motion and measuring velocity using deep learning and image processing techniques. A convolutional neural network model based on the You Only Look Once algorithm was used to accurately identify debris flow motion from recorded videos. The proposed method achieved high accuracy and fast processing speed, making it suitable for early detection and warning systems.