Design a Monitoring System for Lignite Transferring by Dump Trucks in the Coal Mine Using Deep Learning at the Edge Combined with Cloud Service

Abstract

This research presents the design of a monitoring system for lignite transferring by dump trucks. The hardware developed has a camera to detect and process at the edge level. Two parallel processing models are applied for deep learning classification on the edge systems. The optimization model on the edge devices can be worked with performance and memory limitations in real-time. In the fields experiments, the three-class output model had high accuracy with low FPS, so it was reduced to one class. In an object-detection design, YoloV4 and EfficientDet-D4 provide 95% and 98% accuracy, respectively, with low FPS. To apply the model optimization using TensorFlow Lite, accuracy is achieved 97% and 96% with higher FPS and better CPU utilization performance that can reduce 30.7% in thermal performance of CPU. The dump truck classification used the mobileNet, EfficientNet, and Xception models. The results show the Xception was the most accurate at 99.0%, demonstrating the model optimization for processing. The results are on the edge that can be effectively deployed into the fields.