Research and development of plateau portable landslide detection equipment based on Jetson Nano

Abstract

In response to the problems that landslides occur in remote locations, are difficult to monitor and landslide target detection is not easily deployed at the embedded end, a landslide detection device based on the Jetson nano edge device is developed. The technology detects information of landslide feature objects through YOLOV5m technology, and is able to output landslide area information as well as geographic location information while accurately detecting landslide targets. A dataset of 25,000 landslide images is used to build the data set, and the training and validation sets are divided 9:1. A deep learning network is used to extract landslide features and build a landslide target detection model. After that, the train.py file is placed on the cloud server for training, and the best.pt file is migrated to Jetson Nano and tested on the embedded platform. The experimental results show that the average running time of single frame of YOLOV5m model in the embedded device is 100ms, and the detection accuracy can be maintained above 80%, which can achieve accurate detection and information acquisition of landslide target on Jetson Nano device, and lay the foundation for the development of edge device module for landslide detection later.