Autonomous Canal Following by a Micro-Aerial Vehicle Using Deep CNN

Abstract

Globally, large-scale irrigation canal networks serve as the backbone of agriculture in many important river basins. However, these water channels are in a constant threat of erosion, silt accumulation and structural damages over time which significantly reduces the water carrying capacity. Therefore, periodic inspections of the canals are required for critical operations and maintenance tasks. Due to the vast lengths of the channels and time-critical operations, automation has become a necessity. In this paper, we have proposed an aerial autonomous canal traversal system using ResNet50 inspired deep convolutional neural network. Given the uniqueness of our problem, we have generated our dataset for supervised learning and validation and later evaluated the proposed approach on a real canal. We have implemented our approach on a COTS micro-aerial vehicle. We have designed our system in such a way that it takes 200ms from perception to action thereby making the system real-time. We compare the superior performance of our Res Net 50 inspired network with other state-of-the-art CNNs trained on canal datasets.